Multicast Capacity of Optical WDM Packet Ring for Hotspot Traffic

Packet-switching WDM ring networks with a hotspot transporting unicast, multicast, and broadcast traffic are important components of high-speed metropolitan area networks. For an arbitrary multicast fanout traffic model with uniform, hotspot destinat…

Authors: Matthias an der Heiden, Michel Sortais, Michael Scheutzow

Multicast Capacity of Optical WDM Packet Ring for Hotspot Traffic
MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC MA TTHIAS AN DER HEIDEN, MICHEL SOR T AIS, MICHAEL SCHEUTZO W, MAR TIN REISSLEIN, AND MAR TIN MAIER Abstra t. P a k et-swit hing WDM ring net w orks with a hotsp ot transp orting uniast, m ultiast, and broadast tra are imp ortan t omp onen ts of high-sp eed metrop olitan area net w orks. F or an arbitrary m ultiast fanout tra mo del with uniform, hotsp ot destination, and hotsp ot soure pa k et tra, w e analyze the maxim um a hiev able long-run a v erage pa k et throughput, whi h w e refer to as multi ast  ap aity , of bi-diretional shortest-path routed WDM rings. W e iden tify three segmen ts that an exp eriene the maxim um utilization, and th us, limit the m ultiast apait y . W e  haraterize the segmen t utilization probabilities through b ounds and appro ximations, whi h w e v er- ify through sim ulations. W e diso v er that shortest-path routing an lead to utilization probabilities ab o v e one half for mo derate to large p ortions of hotsp ot soure m ulti- and broadast tra, and onsequen tly m ultiast apaities of less than t w o sim ultaneous pa k et transmissions. W e outline a one-op y routing strategy that guaran tees a m ultiast apait y of at least t w o sim ultaneous pa k et transmissions for arbitrary hotsp ot soure tra. Keyw ords: Hotsp ot tra, m ultiast, pa k et throughput, shortest path routing, spatial reuse, w a v elength division m ultiplexing (WDM). 1. Intr odution Optial pa k et-swit hed ring w a v elength division m ultiplexing (WDM) net w orks ha v e emerged as a promising solution to alleviate the apait y shortage in the metrop olitan area, whi h is ommonly referred to as metro gap. P a k et-swit hed ring net w orks, su h as the Resilien t P a k et Ring (RPR) [1, 2, 3 ℄, o v erome man y of the shortomings of iruit-swit hed ring net w orks, su h as lo w pro visioning exibilit y for pa k et data tra [4℄. In addition, the use of m ultiple w a v elength  hannels in WDM ring net w orks, see e.g., [5 , 6, 7, 8, 9, 10, 11 , 12, 13℄, o v eromes a k ey limitation of RPR, whi h w as originally designed for a single-w a v elength  hannel in ea h ring diretion. In optial pa k et-swit hed ring net w orks, the destination no des t ypially remo v e (strip) the pa k ets destined to them from the ring. This destination stripping allo ws the destination no de as w ell as other no des do wnstream to utilize the w a v elength  hannel for their o wn transmissions. With this so-alled sp atial wavelength r euse , m ultiple sim ultaneous transmissions an tak e plae on an y giv en w a v elength  hannel. Spatial w a v elength reuse is maximized through shortest path routing, whereb y the soure no de sends a pa k et Supp orted b y the DF G Resear h Cen ter Ma theon Mathematis for k ey te hnologies in Berlin. M. an der Heiden, M. Sortais, and M. S heutzo w are with the Departmen t of Mathematis, T e h- nial Univ ersit y Berlin, 10623 Berlin, German y (e-mail: Matthias.an.der.Heidenalumni.TU- Berl in.DE , sortaismath-info.univ-paris5. fr , msmath.tu-berlin.de ). M. Reisslein is with the Dept. of Eletrial Eng., Arizona State Univ., T emp e, AZ 852875706, USA (e-mail: reissleinasu.edu ). M. Maier is with the Institut National de la Re her he Sien tique (INRS), Mon tréal, QC, H5A 1K6, CANAD A (e-mail: maierieee.org ). 1 MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 2 in the ring diretion that rea hes the destination with the smallest hop distane, i.e., tra v ersing the smallest n um b er of in termediate net w ork no des. Multiast tra is widely exp eted to aoun t for a large p ortion of the metro area tra due to m ulti-part y omm uniation appliations, su h as tele-onferenes [ 14 ℄, virtual priv ate net w ork in teronnetions, in terativ e distane learning, distributed games, and on ten t distribution. These m ulti-part y appliations are exp eted to demand substan tial bandwidths due to the trend to deliv er the video omp onen t of m ultimedia on ten t in the High-Denition T elevision (HDTV) format or in video formats with ev en higher resolutions, e.g., for digital inema and tele-immersion appliations. While there is at presen t san t quan titativ e information ab out the m ultiast tra v olume, there is ample anedotal evidene of the emerging signiane of this tra t yp e [15, 16℄. As a result, m ultiasting has b een iden tied as an imp ortan t servie in optial net w orks [17 , 18℄ and has b egun to attrat signian t atten tion in optial net w orking resear h as outlined in Setion 1.1 . Metrop olitan area net w orks onsist t ypially of edge rings that in teronnet sev eral aess net w orks (e.g., Ethernet P assiv e Optial Net w orks) and onnet to a metro ore ring [ 4℄. The metro ore ring in teronnets sev eral metro edge rings and onnets to the wide area net w ork. The no de onneting a metro edge ring to the metro ore ring is t ypially a tra hotsp ot as it ollets/distributes tra destined to/originating from other metro edge rings or the wide area net w ork. Similarly , the no de onneting the metro ore ring to the wide area net w ork is t ypially a tra hotsp ot. Examining the apait y of optial pa k et-swit hed ring net w orks for hotsp ot tra is therefore v ery imp ortan t. In this pap er w e examine the m ultiast apait y (maxim um a hiev able long run a v erage m ulti- ast pa k et throughput) of bidiretional WDM optial ring net w orks with a single hotsp ot for a general fanout tra mo del omprising uniast, m ultiast, and broadast tra. W e onsider an arbitrary tra mix omp osed of uniform tra, hotsp ot destination tra (from regular no des to the hotsp ot), and hotsp ot soure tra (from the hotsp ot to regular no des). W e study the widely onsidered no de ar hiteture that allo ws no des to transmit on all w a v elength  hannels, but to reeiv e only on one  hannel. W e initially examine shortest path routing b y deriving b ounds and appro xima- tions for the ring segmen t utilization probabilities due to uniform, hotsp ot destination, and hotsp ot soure pa k et tra. W e pro v e that there are three ring segmen ts (in a giv en ring diretion) that go v ern the maxim um segmen t utilization probabilit y . F or the lo  kwise diretion in a net w ork with no des 1 , 2 , . . . , N and w a v elengths 1 , 2 , . . . , Λ (with N/ Λ ≥ 1 ), whereb y no de 1 reeiv es on w a v e- length 1, no de 2 on w a v elength 2, . . . , no de Λ on w a v elength Λ , no de Λ + 1 on w a v elength 1, and so on, and with no de N denoting the index of the hotsp ot no de, the three ritial segmen ts are iden tied as: MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 3 ( i ) the segmen t onneting the hotsp ot, no de N , to no de 1 on w a v elength 1, ( ii ) the segmen t onneting no de Λ − 1 to no de Λ on w a v elength Λ , and ( iii ) the segmen t onneting no de N − 1 to no de N on w a v elength Λ . The utilization on these three segmen ts limits the maxim um a hiev able m ultiast pa k et throughput. W e observ e from the deriv ed utilization probabilit y expressions that the utilizations of the rst t w o iden tied segmen ts exeed 1/2 (and approa h 1) for large frations of hotsp ot soure m ulti- and broadast tra, whereas the utilization of the third iden tied segmen t is alw a ys less than or equal to 1/2. Th us, shortest path routing a hiev es a long run a v erage m ultiast throughput of less than t w o sim ultaneous pa k et transmissions (and approa hing one sim ultaneous pa k et transmission) for large p ortions of hotsp ot soure m ulti- and broadast tra. W e sp eify one-op y routing whi h sends only one pa k et op y for hotsp ot soure tra, while uniform and hotsp ot destination pa k et tra is still serv ed using shortest path routing. One-op y routing ensures a apait y of at least t w o sim ultaneous pa k et transmissions for arbitrary hotsp ot soure tra, and at least appro ximately t w o sim ultaneous pa k et transmissions for arbitrary o v er- all tra. W e v erify the auray of our b ounds and appro ximations for the segmen t utilization probabilities, whi h are exat in the limit N/ Λ → ∞ , through omparisons with utilization proba- bilities obtained from disrete ev en t sim ulations. W e also quan tify the gains in maxim um a hiev able m ultiast throughput a hiev ed b y the one-op y routing strategy o v er shortest path routing through sim ulations. This pap er is strutured as follo ws. In the follo wing subsetion, w e review related w ork. In Setion 2, w e in tro due the detailed net w ork and tra mo dels and formally dene the m ultiast apait y . In Setion 3, w e establish fundamen tal prop erties of the ring segmen t utilization in WDM pa k et rings with shortest path routing. In Setion 4, w e deriv e b ounds and appro ximations for the ring segmen t utilization due to uniform, hotsp ot destination, and hotsp ot soure pa k et tra on the w a v elengths that the hotsp ot is not reeiving on, i.e., w a v elengths 1 , 2 , . . . , Λ − 1 in the mo del outlined ab o v e. In Setion 5, w e deriv e similar utilization probabilit y b ounds and appro ximations for w a v elength Λ that the hotsp ot reeiv es on. In Setion 6, w e pro v e that the three sp ei segmen ts iden tied ab o v e go v ern the maxim um segmen t utilization and m ultiast apait y in the net w ork, and disuss impliations for pa k et routing. In Setion 7, w e presen t n umerial results obtained with the deriv ed utilization b ounds and appro ximations and ompare with v erifying sim ulations. W e onlude in Setion 8. 1.1. Related W ork. There has b een inreasing resear h in terest in reen t y ears for the wide range of asp ets of m ultiast in general mesh iruit-swit hed WDM net w orks, inluding ligh tpath design, MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 4 see for instane [19 ℄, tra gro oming, see e.g., [20 ℄, routing and w a v elength assignmen t, see e.g., [21 , 22, 23 ℄, and onnetion arrying apait y [24℄. Similarly , m ultiasting in pa k et-swit hed single-hop star WDM net w orks has b een in tensely in v estigated, see for instane [25 , 26, 27, 28℄. In on trast to these studies, w e fo us on pa k et-swit hed WDM ring net w orks in this pap er. Multiasting in iruit-swit hed WDM rings, whi h are fundamen tally dieren t from the pa k et- swit hed net w orks onsidered in this pap er, has b een extensiv ely examined in the literature. The s heduling of onnetions and ost-eetiv e design of bidiretional WDM rings w as addressed, for instane in [29 ℄. Cost-eetiv e tra gro oming approa hes in WDM rings ha v e b een studied for instane in [30 , 31℄. The routing and w a v elength assignmen t in reongurable bidiretional WDM rings with w a v elength on v erters w as examined in [32 ℄. The w a v elength assignmen t for m ultiasting in iruit-swit hed WDM ring net w orks has b een studied in [33 , 34, 35, 36, 37 , 38℄. F or uniast tra, the throughputs a hiev ed b y dieren t iruit-swit hed and pa k et-swit hed optial ring net w ork ar hitetures are ompared in [39 ℄. Optial p aket-swithe d WDM ring net w orks ha v e b een exp erimen tally demonstrated, see for in- stane [13, 40 ℄, and studied for uniast tra, see for instane [5, 41 , 6 , 7, 8 , 9 , 10 , 11, 12 , 13 ℄. Multiasting in pa k et-swit hed WDM ring net w orks has reeiv ed inreasing in terest in reen t y ears [ 42, 10 ℄. The photonis lev el issues in v olv ed in m ultiasting o v er ring WDM net w orks are explored in [43℄, while a no de ar hiteture suitable for m ultiasting is studied in [44 ℄. The general net w ork ar hiteture and MA C proto ol issues arising from m ultiasting in pa k et-swit hed WDM ring net w orks are addressed in [40 , 45 ℄. The fairness issues arising when transmitting a mix of uniast and m ultiast tra in a ring WDM net w ork are examined in [46℄. The m ultiast apait y of pa k et-swit hed WDM ring net w orks has b een examined for uniform pa k et tra in [47 , 48 , 49, 50℄. In on trast, w e onsider non-uniform tra with a hotsp ot no de, as it ommonly arises in metro edge rings [51 ℄. Studies of non-uniform tra in optial net w orks ha v e generally fo used on issues arising in iruit- swit hed optial net w orks, see for instane [52, 53 , 54, 55 , 56 , 57, 58 ℄. A omparison of iruit-swit hing to optial burst swit hing net w ork te hnologies, inluding a brief omparison for non-uniform tra, w as onduted in [59 ℄. The throughput  harateristis of a mesh net w ork in teronneting routers on an optial ring through b er shortuts for non-uniform uniast tra w ere examined in [60℄. The study [61℄ onsidered the throughput  harateristis of a ring net w ork with uniform uniast tra, where the no des ma y adjust their send probabilities in a non-uniform manner. The m ultiast apait y of a single-w a v elength pa k et-swit hed ring with non-uniform tra w as examined in [62℄. In on trast to these w orks, w e onsider non-uniform tra with an arbitrary fanout, whi h aommo dates a wide range of uniast, m ultiast, and broadast tra mixes, in a WDM ring net w ork. MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 5 x x x x x x x x x x x x x x x x 1 8 9 13 14 15 N=16 2 3 Λ=4 5 6 7 10 11 12 Figure 2.1. Illustration of the lo  kwise w a v elength  hannels of a WDM ring net- w ork with N = 16 no des and Λ = 4 w a v elength  hannels. 2. System Model and Not a tions W e let N denote the n um b er of net w ork no des, whi h w e index sequen tially b y i, i = 1 , . . . , N , in the lo  kwise diretion and let M := { 1 , . . . , N } denote the set of net w ork no des. F or on v eniene, w e lab el the no des mo dulo N , e.g., no de N is also denoted b y 0 or − N . W e onsider the family of no de strutures where ea h no de an transmit on an y w a v elength using either one or m ultiple tunable transmitters ( T T s ) or an arra y of Λ xed-tuned transmitters  F T Λ  , and reeiv e on one w a v elength using a single xed-tuned reeiv er ( F R ) . F or N = Λ , ea h no de has its o wn home  hannel for reeption. F or N > Λ , ea h w a v elength is shared b y η := N / Λ no des, whi h w e assume to b e an in teger. F or 1 ≤ i ≤ N , w e let y u i denote the lo  kwise orien ted ring segmen t onneting no de i − 1 to no de i . Analogously , w e let x u i denote the oun ter lo  kwise orien ted ring segmen t onneting no de i to no de i − 1 . Ea h ring deplo ys the same set of w a v elength  hannels { 1 , . . . , Λ } , one set on the lo  kwise ring and another set on the oun terlo  kwise ring. The no des n = λ + k Λ with k ∈ { 0 , 1 , . . . , η − 1 } share the drop w a v elength λ . W e refer to the inoming edges of these no des, i.e., the edges y u λ + k Λ and x u λ +1+ k Λ , as riti al e dges on λ . F or m ultiast tra, the sending no de generates a op y of the m ultiast pa k et for ea h w a v elength that is drop w a v elength for at least one destination no de. Denote b y S the no de that is the sender. W e in tro due the random set of destinations (fanout set) F ⊂ ( { 1 , 2 , . . . , N } \ { S } ) . Moreo v er, w e dene the set of ativ e no des A as the union of the sender and all destinations, i.e., A := F ∪ { S } . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 6 W e onsider a tra mo del om bining a p ortion α of uniform tra, a p ortion β of hotsp ot destination tra, and a p ortion γ of hotsp ot soure tra with α, β , γ ≥ 0 and α + β + γ = 1 : Uniform tra: A giv en generated pa k et is a uniform tr a pa k et with probabilit y α . F or su h a pa k et, the sending no de is  hosen uniformly at random amongst all net w ork no des { 1 , 2 , . . . , N } . One the sender S is  hosen, the n um b er of reeiv ers (fanout) l ∈ { 1 , 2 , . . . , N − 1 } is  hosen at random aording to a disrete probabilit y distribution ( µ l ) N − 1 l =1 . One the fanout l is  hosen, the random set of destinations (fanout set) F ⊂ ( { 1 , 2 , . . . , N } \ { S } ) is  hosen uniformly at random amongst all subsets of { 1 , 2 , . . . , N } \ { S } ha ving ardinalit y l . W e denote b y P α the probabilit y measure asso iated with uniform tra. Hotsp ot destination tra: A giv en pa k et is a hotsp ot destination tr a pa k et with prob- abilit y β . F or a hotsp ot destination tra pa k et, no de N is alw a ys a destination. The sending no de is  hosen uniformly at random amongst the other no des { 1 , 2 , . . . , N − 1 } . One the sender S is  hosen, the fanout l ∈ { 1 , 2 , . . . , N − 1 } is  hosen at random aording to a disrete probabilit y distribution ( ν l ) N − 1 l =1 . One the fanout l is  hosen, a random fanout subset F ′ ⊂ ( { 1 , 2 , . . . , N − 1 } \ { S } ) is  hosen uniformly at random amongst all subsets of { 1 , 2 , . . . , N − 1 } \ { S } ha ving ardinalit y ( l − 1) , and the fanout set is F = F ′ ∪ { N } . W e denote b y Q β the probabilit y measure asso iated with hotsp ot destination tra. Hotsp ot soure tra: A giv en pa k et is a hotsp ot sour  e tr a pa k et with probabilit y γ . F or su h a pa k et, the sending no de is  hosen to b e no de N . The fanout 1 ≤ l ≤ ( N − 1) is  hosen at random aording to a disrete prob. distribution ( κ l ) N − 1 l =1 . One the fanout l is  hosen, a random fanout set F ⊂ { 1 , 2 , . . . , N − 1 } is  hosen uniformly at random amongst all subsets of { 1 , 2 , . . . , N − 1 } ha ving ardinalit y l . W e denote b y Q γ the probabilit y measures asso iated with hotsp ot soure tra. While our analysis assumes that the tra t yp e, the soure no de, the fanout, and the fanout set are dra wn indep enden tly at random, this indep endene assumption is not ritial for the analysis. Our results hold also for tra patterns with orrelations, as long as the long run a v erage segmen t utilizations are equiv alen t to the utilizations with the indep endene assumption. F or instane, our results hold for a orrelated tra mo del where a giv en soure no de transmits with a probabilit y p < 1 to exatly the same set of destinations as the previous pa k et sen t b y the no de, and with probabilit y 1 − p to an indep enden tly randomly dra wn n um b er and set of destination no des. W e denote b y P l α the probabilit y measure P α onditioned up on |F | = l , and dene Q l β and Q l γ analogously . W e denote the set of no des with drop w a v elength λ b y (2.1) M λ := { λ + k Λ | k ∈ { 0 , . . . , η − 1 }} . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 7 The set of all destinations with drop w a v elength λ is then (2.2) F λ := F ∩ M λ . Moreo v er, w e use the follo wing notation: F or ℓ ∈ { 1 , . . . , N − 1 } w e denote the probabilit y of ℓ destinations on w a v elength λ b y µ λ,ℓ , ν λ,ℓ , and κ λ,ℓ for uniform, hotsp ot destination, and hotsp ot soure tra, resp etiv ely . Sine the fanout set is  hosen uniformly at random among all subsets of { 1 , 2 , . . . , N } \ { S } ha ving ardinalit y l , these usage-probabilities an b e expressed b y µ l , ν l , and κ l . Dep ending on whether the sender is on the drop-w a v elength or not, w e obtain sligh tly dieren t expressions. As will b eome eviden t shortly , it sues to fo us on the ase where the sender is on the onsidered drop w a v elength λ , i.e., S ∈ M λ , sine the relev an t probabilities are estimated through omparisons with transformations (enlarged, redued or righ t/left-shifted ring in tro dued in App endix A) that put the sender in M λ . Through elemen tary om binatorial onsiderations w e obtain the follo wing probabilit y distribu- tions: F or uniform tra, the probabilit y for ha ving ℓ ∈ { 0 , . . . , l ∧ η } destinations on w a v elength λ is (2.3) µ λ,ℓ := N − 1 X l =max(1 ,ℓ )  η ℓ  N − η l − ℓ   N l  µ l . F or hotsp ot destination tra, w e obtain for w a v elengths λ 6 = Λ and ℓ ∈ { 0 , . . . , ( l − 1) ∧ η } (2.4) ν λ,ℓ := N − 1 X l =max(1 ,ℓ )  η ℓ  N − η − 1 l − ℓ − 1   N − 1 l − 1  ν l , as w ell as for w a v elength Λ homing the hotsp ot and ℓ ∈ { 1 , . . . , l ∧ η } (2.5) ν Λ ,ℓ := N − 1 X l =1  η − 1 ℓ − 1  N − η l − ℓ   N − 1 l − 1  ν l . Finally , for hotsp ot soure tra, w e obtain for λ 6 = Λ and ℓ ∈ { 0 , . . . , l ∧ η } (2.6) κ λ,ℓ := N − 1 X l =max(1 ,ℓ )  η ℓ  N − 1 − η l − ℓ   N − 1 l  κ l and for λ = Λ and ℓ ∈ { 0 , . . . , l ∧ ( η − 1) } (2.7) κ Λ ,ℓ := N − 1 X l =max(1 ,ℓ )  η − 1 ℓ  N − η l − ℓ   N − 1 l  κ l . F or a giv en w a v elength λ , w e denote b y p ℓ α,λ the probabilit y measure P α onditioned up on |F λ | = ℓ , and dene q ℓ β ,λ and q ℓ γ ,λ analogously . R emark 2.1 . Whenev er it is lear whi h w a v elength λ is onsidered w e omit the subsript λ and write p ℓ α , q ℓ β , or q ℓ γ . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 8 W e in tro due the set of ativ e no des A λ on a giv en drop w a v elength λ as (2.8) A λ := F λ ∪ { S } . W e order the no des in this set in inreasing order of their no de indies, i.e., (2.9) A λ = { X λ, 1 , X λ, 2 , . . . , X λ,ℓ +1 } , 1 ≤ X λ, 1 < X λ, 2 < . . . < X λ,ℓ +1 ≤ N , and onsider the gaps (2.10) X λ, 1 + ( N − X λ,ℓ +1 ) , ( X λ, 2 − X λ, 1 ) , . . . , ( X λ,ℓ +1 − X λ,ℓ ) , b et w een suessiv e no des in the set A λ . W e ha v e denoted here again b y ℓ ≡ ℓ λ the random n um b er of destinations with drop w a v elength λ . F or shortest path routing, i.e., to maximize spatial w a v elength reuse, w e determine the largest of these gaps. Sine there ma y b e a tie among the largest gaps (in whi h ase one of the largest gaps is  hosen uniformly at random), w e denote the seleted largest gap as  C LG λ   (for Chosen Largest Gap). Supp ose the C LG λ is b et w een no des X λ,i − 1 and X λ,i . With shortest path routing, the pa k et is then sen t from the sender S to no de X λ,i − 1 , and from the sender S to no de X λ,i in the opp osite diretion. Th us, the largest gap is not tra v ersed b y the pa k et transmission. Note that b y symmetry , P { y u 1 is used } = P { x u N is used } , and P { y u N is used } = P { x u 1 is used } . More generally , for reasons of symmetry , it sues to ompute the utilization probabilities for the lo  kwise orien ted edges. F or n ∈ { 1 , . . . , N } , w e abbreviate (2.11) y n λ := y u n is used on w a v elength λ. It will b e on v enien t to all no de N also no de 0 . W e let G λ , G λ = 0 , . . . , N − 1 , b e a random v ariable denoting the rst no de b ordering the  hosen largest gap on w a v elength λ , when this gap is onsidered lo  kwise. The utilization probabilit y for the lo  kwise segmen t n on w a v elength λ is giv en b y (2.12) P  y n λ  = η X ℓ =0  α · p ℓ α  y n λ  · µ λ,ℓ + β · q ℓ β  y n λ  · ν λ,ℓ + γ · q ℓ γ  y n λ  · κ λ,ℓ  . Our primary p erformane metri is the maxim um pa k et throughout (stabilit y limit). More sp eif- ially , w e dene the (eetiv e) m ultiast apait y C M as the maxim um n um b er of pa k ets (with a giv en tra pattern) that an b e sen t sim ultaneously in the long run, and note that C M is giv en as the reipro al of the largest ring segmen t utilization probabilit y , i.e., C M := 1 max n ∈{ 1 ,...,N } max λ ∈{ 1 ,..., Λ } P  y n λ  . (2.13) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 9 3. General Pr oper ties of Segment Utiliza tion First, w e pro v e a general reursion form ula for shortest path routing. Prop osition 3.1. L et λ ∈ { 1 , . . . , Λ } b e a xe d wavelength. F or al l no des n ∈ { 0 , . . . , N − 1 } , P  y ( n + 1) λ  = P  y n λ  + P ( S = n ) − P ( G λ = n ) . (3.1) Pr o of. There are t w o omplemen tary ev en ts leading to y ( n + 1) λ : (A) the pa k et tra v erses (on w a v e- length λ ) b oth the lo  kwise segmen t y u n +1 and the preeding lo  kwise segmen t y u n , i.e., the sender is a no de S 6 = n , and (B) no de n is the sender ( S = n ) and transmits the pa k et in the lo  kwise diretion, so that it tra v erses segmen t y u n +1 follo wing no de n (in the lo  kwise diretion). F ormally , P  y ( n + 1) λ  = P  y n λ and y ( n + 1) λ  + P  S = n and y ( n + 1) λ  . (3.2) Next, note that the ev en t that the lo  kwise segmen t y u n is tra v ersed an b e deomp osed in to t w o omplemen tary ev en ts, namely (a) segmen ts y u n and y u n +1 are tra v ersed, and (b) segmen t y u n is tra v ersed, but not segmen t y u n +1 , i.e., P  y n λ  = P  y n λ and y ( n + 1) λ  + P  y n λ and not y ( n + 1) λ  . (3.3) Similarly , w e an deomp ose the ev en t of no de n b eing the sender as (3.4) P ( S = n ) = P  S = n and y ( n + 1) λ  + P  S = n and not y ( n + 1) λ  . Hene, w e an express P  y ( n + 1) λ  as P  y ( n + 1) λ  = P  y n λ  − P  y n λ and not y ( n + 1) λ  + P ( S = n ) − P  S = n and not y ( n + 1) λ  . (3.5) No w, note that there are t w o omplemen tary ev en ts that result in the CLG to start at no de n , su h that lo  kwise segmen t n + 1 is inside the CLG: ( i ) no de n is the last destination no de rea hed b y the lo  kwise transmission, i.e., segmen t n is used, but segmen t n + 1 is not used, and ( ii ) no de n is the sender and transmits only a pa k et op y in the oun ter lo  kwise diretion. Hene, P ( G λ = n ) = P  y n λ and not y ( n + 1) λ  + P  S = n and not y ( n + 1) λ  . (3.6) Therefore, w e obtain the general reursion (3.7) P  y ( n + 1) λ  = P  y n λ  + P ( S = n ) − P ( G λ = n ) .  MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 10 W e in tro due the left (oun ter lo  kwise) shift and the righ t (lo  kwise) shift of no de n to b e ⌊ n ⌋ λ and ⌈ n ⌉ λ giv en b y (3.8) ⌊ n ⌋ λ :=  n − λ Λ  Λ + λ and ⌈ n ⌉ λ :=  n − λ Λ  Λ + λ. The oun ter lo  kwise shift maps a no de n not homed on λ on to the nearest no de in the oun ter lo  kwise diretion that is homed on λ . Similarly , the lo  kwise shift maps a no de n not homed on λ on to the losest no de in the lo  kwise diretion that is homed on λ . F or the tra on w a v elength λ , w e obtain b y rep eated appliation of Prop osition 3.1 P  y ( ⌈ n ⌉ λ ) λ  = P  y n λ  + ⌈ n ⌉ λ − 1 X i = n P ( S = i ) − ⌈ n ⌉ λ − 1 X i = n P ( G λ = i ) (3.9) = P  y n λ  + P ( S ∈ { n , . . . , ⌈ n ⌉ λ − 1 } ) − P ( G λ ∈ { n, . . . , ⌈ n ⌉ λ − 1 } ) . (3.10) Note that the CLG on λ an only start ( i ) at the soure no de, irresp etiv e of whether it is on λ , or ( ii ) at a destination no de on λ . Consider a giv en no de n that is not on λ , then the no des in { n, n + 1 , . . . , ⌈ n ⌉ λ − 1 } are not on λ . (If no de n is on λ , i.e., n = ⌈ n ⌉ λ , then trivially the set { n, n + 1 , . . . , ⌈ n ⌉ λ − 1 } is empt y and P  y ( ⌈ n ⌉ λ ) λ  = P  y n λ  .) Hene, the CLG on λ an only start at a no de in { n, n + 1 , . . . , ⌈ n ⌉ λ − 1 } if that no de is the soure no de, i.e., (3.11) P ( G λ ∈ { n, . . . , ⌈ n ⌉ λ − 1 } ) = P ( G λ = S ∈ { n, . . . , ⌈ n ⌉ λ − 1 } ) . Next, note that the ev en t that a no de in { n, n + 1 , . . . , ⌈ n ⌉ λ − 1 } is the soure no de an b e deomp osed in to the t w o omplemen tary ev en ts ( i ) a no de in { n, n + 1 , . . . , ⌈ n ⌉ λ − 1 } is the soure no de and the CLG on λ starts at that no de, and ( ii ) a no de in { n, n + 1 , . . . , ⌈ n ⌉ λ − 1 } is the soure no de and the CLG do es not start at that no de. Hene, (3.12) P ( S ∈ { n, . . . , ⌈ n ⌉ λ − 1 } ) = P ( G λ = S ∈ { n, . . . , ⌈ n ⌉ λ − 1 } ) + P ( S ∈ { n, . . . , m − 1 } , G λ 6 = S ) . Inserting (3.11 ) and (3.12) in (3.10 ) w e obtain (3.13) P  y ( ⌈ n ⌉ λ ) λ  = P  y n λ  + P ( S ∈ { n , . . . , m − 1 } , G λ 6 = S ) whi h diretly leads to Corollary 3.2. The usage of non-riti al se gments is smal ler than the usage of riti al se gments, mor e pr e isely for n ∈ { 0 , . . . , N − 1 } : P  y n λ  = P  y ( ⌈ n ⌉ λ ) λ  − P ( S ∈ { n , . . . , ⌈ n ⌉ λ − 1 } , G λ 6 = S ) . (3.14) T o ompare the exp eted length of the largest gap on a w a v elength in the WDM ring with the exp eted length of the largest gap in the single w a v elength ring, w e in tro due the enlarged and MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 11 redued ring in App endix A. In brief, in the enlarged ring, an extra no de is added on the onsidered w a v elength b et w een the λ -neigh b ors of the soure no de. This enlargemen t results in ( a ) a set of η + 1 no des homed on the onsidered w a v elength, and ( b ) an enlarged set of ativ e no des A + λ on taining the original destination no des plus the added extra no de (whi h in a sense represen ts the soure no de on the onsidered w a v elength) for a total of ℓ + 1 ativ e no des. The exp eted length of the largest gap on this enlarged w a v elength ring with ℓ + 1 ativ e no des among η + 1 no des homed on the w a v elength (A) is equiv alen t to Λ times the exp eted length of the largest gap on a single w a v elength ring with l = ℓ destination no des and one soure no de among N no des homed on the ring, and (B) pro vides an upp er b ound on the exp eted length of the largest gap on the original w a v elength ring (b efore the enlargemen t). In the redued ring, the left- and righ t-shifted soure no de are merged in to one no de on the onsidered w a v elength, resulting ( a ) in a set of η − 1 no des homed on the onsidered w a v elength, and ( b ) a set A − λ of ℓ − 1 , ℓ , or ℓ + 1 ativ e no des. The exp eted length of the largest gap dereases with inreasing n um b er of ativ e no des, hene w e onsider the ase with ℓ + 1 ativ e no des for a lo w er b ound. The exp eted length of the largest gap on the redued w a v elength ring with ℓ + 1 ativ e no des among η − 1 no des homed on the w a v elength (A) is equiv alen t to Λ times the exp eted length of the largest gap on a single w a v elength ring with l = ℓ destination no des and one soure no de among N no des homed on the ring, and (B) pro vides a lo w er b ound on the exp eted length of the largest gap on the original w a v elength ring (b efore the redution). F rom these t w o onstrutions, whi h are formally pro vided in App endix A, w e diretly obtain: Prop osition 3.3. Given that the  ar dinality of F λ is ℓ , the exp e te d length of the CLG on wavelength λ is b ounde d by: (3.15) Λ · g ( ℓ, η − 1) ≤ E ℓ ( | C LG λ | ) ≤ Λ · g ( ℓ, η + 1) , wher e g ( l, N ) denotes the exp e te d length of the CLG for a single wavelength ring with N no des, when the ative set is hosen uniformly at r andom fr om al l subsets of { 1 , . . . , N } with  ar dinality ( l + 1) . The exp eted length of the largest gap g ( l, N ) [ 63 ℄ is giv en for l = 0 , . . . , N − 1 , b y g ( l, N ) = P N k =1 k · q l,N ( k ) , where q l,N ( · ) denotes the distribution of the length of the largest gap. Let p l,N ( k ) =  N − k − 1 l − 1  /  N − 1 l  denote the probabilit y that an arbitrary gap has k hops. Then the distribution q l,N ma y b e omputed using the reursion q l,N ( k ) = p l,N ( k ) · k X m =1 q l − 1 ,N − k ( m ) + k − 1 X m =1 p l,N ( m ) · q l − 1 ,N − m ( k ) (3.16) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 12 together with the initialization q 0 ,N ( k ) = δ N ,k and q N − 1 ,N ( k ) = δ 1 ,k , where δ N ,k denotes the Kro- ne k er Delta. Whereb y , q 0 ,N ( k ) = δ N ,k means a ring with only one ativ e no de has only one gap of length N , hene the largest gap has length N with probabilit y one. Similarly , q N − 1 ,N ( k ) = δ 1 ,k means a ring with all no des ativ e (broadast ase) has N gaps with length one, hene the largest gap has length 1 with probabilit y one. This initialization diretly implies g (0 , N ) = N as w ell as g ( N − 1 , N ) = 1 . Ob viously , w e ha v e to set g ( l, N ) = 0 for l ≥ N . 4. Bounds on Segment Utiliza tion f or λ 6 = Λ 4.1. Uniform T ra. In the setting of uniform tra, one has for all n ∈ {− Λ + λ + 1 , . . . , λ } and k ∈ { 0 , . . . , η − 1 } , for reasons of symmetry: (4.1) P α  y n λ  = P α  y ( n + k Λ) λ  . F or n ∈ {− Λ + λ + 1 , . . . , λ } , the dierene b et w een ritial and non-ritial edges, orresp onding to Corollary 3.2, an b e estimated b y 0 ≤ P α ( S ∈ { n, . . . , λ − 1 } , G λ 6 = S ) ≤ P α ( S ∈ { n, . . . , λ − 1 } ) = λ − n N . (4.2) With shortest path routing, on a v erage N − E α ( | C LG | λ ) segmen ts are tra v ersed on λ to serv e a uniform tra pa k et. Equiv alen tly , w e obtain the exp eted n um b er of tra v ersed segmen ts b y sum- ming the utilization probabilities of the individual segmen ts, i.e., as P N n =1 P α  y n λ  + P N n =1 P α  x n λ  , whi h, due to symmetry , equals 2 P N n =1 P α  y n λ  . Hene, N − E α ( | C LG | λ ) = 2 N X n =1 P α  y n λ  (4.3) and E α ( | C LG | λ ) = N − 2 N X n =1 P α  y n λ  (4.4) = N − 2 η λ X k = − Λ+ λ +1 P α  y k λ  . (4.5) Expressing P α  y k λ  using Corollary 3.2 , w e obtain (4.6) E α ( | C LG | λ ) = N − 2 N P α  y λ λ  + 2 η λ X k = − Λ+ λ +1 P α ( S ∈ { k , . . . , λ − 1 } , G λ 6 = S ) . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 13 Solving for P α  y λ λ  yields (4.7) P α  y λ λ  = 1 2 − 1 2 N E α ( | C LG | λ ) + 1 Λ λ X k = − Λ+ λ +1 P α ( S ∈ { k , . . . , λ − 1 } , G λ 6 = S ) . Hene, the inequalities (4.2) lead to (4.8) 1 2 − 1 2 N E α ( | C LG | λ ) ≤ P α  y λ λ  ≤ 1 2 − 1 2 N E α ( | C LG | λ ) + Λ − 1 2 N . Emplo ying the b ounds for E α ( | C LG | λ ) from Prop osition 3.3 giv es (4.9) 1 2 − 1 2 η η X ℓ =0 g ( ℓ, η + 1) µ λ,ℓ ≤ P α  y λ λ  ≤ 1 2 − 1 2 η η − 2 X ℓ =0 g ( ℓ, η − 1) µ λ,ℓ + Λ − 1 2 N . 4.2. Hotsp ot Destination T ra. The only dierene to uniform tra is that N annot b e a sender, sine it is already a destination, i.e., (4.10) q ℓ β  y n λ  = p ℓ α  y n λ | S 6 = N  . Using p ℓ α ( S = N ) = 1 N , w e obtain q ℓ β  y n λ  = N N − 1 p ℓ α  y n λ  − 1 N − 1 p ℓ α  y n λ | S = N  (4.11) = N N − 1 p ℓ α  y n λ  − 1 N − 1 q ℓ γ  y n λ  . (4.12) Due to the fator 1 N − 1 , the seond term is negligible in the on text of large net w orks. 4.3. Hotsp ot Soure T ra. Sine no de N is the sender (and giv en that there is at least one destination no de on λ ), it sends a pa k et op y o v er segmen t y u n on w a v elength λ if the CLG on λ starts at a no de with index n or higher. Hene, the usage probabilit y of a segmen t an b e omputed as (4.13) q ℓ γ  y n λ  = q ℓ γ ( G λ ≥ n ) for n ∈ { 1 , . . . , N } . W e notie immediately that q ℓ γ  y n λ  is monotone dereasing in n . Moreo v er, for all n ∈ { 1 , . . . , ( η − 1) Λ + λ } , Equation (3.14) simplies to q ℓ γ  y n λ  = q ℓ γ  y ( ⌈ n ⌉ λ ) λ  (4.14) sine the sender is no de N ≡ 0 and onsequen tly P ( S ∈ { n, . . . , ⌈ n ⌉ λ − 1 } , G λ 6 = S ) = 0 for the onsidered n ∈ { 1 , . . . , ( η − 1) Λ + λ } . Sine q ℓ γ  y n λ  is monotone dereasing in n , the maximally used ritial segmen t on w a v elength λ is y u λ . With no de N b eing the sender, the CLG on λ an only start at the soure no de N ≡ 0 , or at a destination no de homed on λ . If the CLG do es not start at N ≡ 0 , the segmen t y u λ leading to the MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 14 rst no de homed on λ , namely no de λ , is utilized. Hene, (4.15) q ℓ γ  y λ λ  = q ℓ γ ( G λ 6 = 0) . Observ e that (4.16) q ℓ γ ( G λ = 0) < q ℓ γ ( G Λ − λ = 0) for λ < Λ 2 , whi h is exploited in Setion 4.4. Enlarging the ring leads to (4.17) q ℓ γ ( G λ = 0) ≤ q ℓ γ  G + λ = 0  = 1 ℓ + 1 , sine the gaps b ordering no de 0 are enlarged whereas the lengths of all other gaps are un hanged. A righ t shifting of S yields the follo wing lo w er b ound: q ℓ γ ( G λ = 0) ≥ q ℓ γ ( G → λ = 0 | λ / ∈ F λ ) q ℓ γ ( λ / ∈ F λ ) (4.18) = 1 ℓ + 1  1 − ℓ η  . (4.19) Th us, 1 − 1 ℓ + 1 ≤ q ℓ γ  y λ λ  ≤ 1 − 1 ℓ + 1  1 − ℓ η  . (4.20) 4.4. Summary of Segmen t Utilization Bounds and Appro ximation for λ 6 = Λ . F or λ 6 = Λ w e obtain from (2.12 ) and (4.12 ) P  y n λ  = η X ℓ =0  p ℓ α  y n λ   αµ λ,ℓ + N N − 1 β ν λ,ℓ  + q ℓ γ  y n λ   γ κ λ,ℓ − 1 N − 1 β ν λ,ℓ  . (4.21) Using Corollary 3.2 for p ℓ α and (4.13 ) for q ℓ γ yields (4.22) max n ∈ M P  y n λ  = P  y λ λ  , i.e., the segmen t n um b er λ exp erienes the maxim um utilization on w a v elength λ . Moreo v er, in- equalit y (4.16 ) yields (4.23) max λ 6 =Λ max n ∈ M P  y n λ  = P  y 1 1  , i.e., the rst segmen t on w a v elength 1, exp erienes the maxim um utilization among all segmen ts on all w a v elengths λ 6 = Λ . F rom (4.21 ) in onjuntion with (4.9 ) and (4.20 ) w e obtain P  y 1 1  ≥ 1 2  α + N N − 1 β  − 1 2 η η X ℓ =0 g ( ℓ, η + 1)  αµ 1 ,ℓ + N N − 1 β ν 1 ,ℓ  + + η X ℓ =0 ℓ ℓ + 1  γ κ 1 ,ℓ − 1 N − 1 β ν 1 ,ℓ  =: p 1 l (4.24) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 15 and P  y 1 1  ≤ 1 2  1 + Λ − 1 N   α + N N − 1 β  − 1 2 η η X ℓ =0 g ( ℓ, η − 1)  αµ 1 ,ℓ + N N − 1 β ν 1 ,ℓ  + η X ℓ =0 ℓ ( η + 1) ( ℓ + 1) η  γ κ 1 ,ℓ − 1 N − 1 β ν 1 ,ℓ  =: p 1 u. (4.25) W e obtain an appro ximation of the segmen t utilization b y onsidering the b eha vior of these b ounds for large η = N Λ . Large η imply η +1 η ∼ 1 as w ell as N N − 1 ∼ 1 , and g ( ℓ, η − 1) ∼ g ( ℓ, η + 1) . In tuitiv ely , this last relation means that the exp eted length of the largest gap on a ring net w ork with ℓ destination no des among η − 1 no des is appro ximately equal to the largest gap when there are ℓ destination no des among η + 1 no des. With these onsiderations w e an simplify the b ounds giv en ab o v e and obtain the appro ximation (v alid for large η ): P  y 1 1  ∼ 1 2 ( α + β ) − 1 2 η η X ℓ =0 g ( ℓ, η ) ( αµ 1 ,ℓ + β ν 1 ,ℓ ) + γ η X ℓ =0 ℓ ℓ + 1 κ 1 ,ℓ =: p 1 a. (4.26) 5. Bounds on Segment Utiliza tion f or λ = Λ F or uniform tra this ase, of ourse, do es not dier from the ase λ 6 = Λ . 5.1. Hotsp ot Destination T ra. Sine N is a destination no de, b y symmetry it is rea hed b y a lo  kwise transmission with probabilit y one half, i.e., (5.1) Q β  y N Λ  = 1 2 . F or hotsp ot destination tra, no de N an not b e the sender, i.e., Q β ( S = N ) = 0 . Hene, b y Prop osition 3.1 : (5.2) Q β  y 1 Λ  = 1 2 − Q β ( G Λ = 0) . Moreo v er, w e ha v e from Corollary 3.2 with n = 1 and λ = Λ : (5.3) Q β  y Λ Λ  = Q β  y 1 Λ  + Q β ( S ∈ { 1 , . . . , Λ − 1 } , G Λ 6 = S ) . T o estimate Q β ( G Λ = 0) , w e in tro due, as b efore, the left- resp. righ t-shift of S , giv en b y (5.4) ⌊ S ⌋ Λ :=  S Λ  Λ and ⌈ S ⌉ Λ :=  S Λ  Λ . Left and righ t shifting of S leads to the follo wing b ounds for the probabilit y q ℓ β ( G Λ = 0) , whi h are pro v en in App endix B. Prop osition 5.1. F or hotsp ot destination tr a,  onditioning on the  ar dinality of F Λ to b e ℓ , the pr ob ability that the CLG starts at no de 0 is b ounde d by: (5.5) 1 ℓ + 1  1 − 1 ℓη  ≤ q ℓ β ( G Λ = 0) ≤ 1 ℓ + 1  1 + 1 η  . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 16 Inserting the b ounds from Prop osition 5.1 and noting that 0 ≤ Q β ( S ∈ { 1 , . . . , Λ − 1 } , G Λ 6 = S ) ≤ (Λ − 1) / (2 N ) leads to (5.6) Q β  y Λ Λ  ≤ 1 2 − η X ℓ =1 ν Λ ,ℓ 1 ℓ + 1  1 − 1 ℓη  + Λ − 1 2 N and (5.7) Q β  y Λ Λ  ≥ 1 2 − N − 1 X ℓ =1 ν Λ ,ℓ 1 ℓ + 1  1 + 1 η  . 5.2. Hotsp ot Soure T ra. Sine w e kno w that N is the sender and has drop w a v elength Λ , w e ha v e a symmetri setting on F Λ and an diretly apply the results of the single w a v elength setting [62℄. In partiular, w e obtain from Setion 3.1.3 in [62 ℄ for ℓ ∈ { 0 , . . . , η − 1 } (5.8) q ℓ γ  y N Λ  = 0 and (5.9) q ℓ γ  y Λ Λ  = q ℓ γ ( G Λ 6 = 0) = ℓ ℓ + 1 . 5.3. Summary of Segmen t Utilization Bounds and Appro ximation for λ = Λ . Inserting the b ounds deriv ed in the preeding setions in (2.12 ), w e obtain P  y Λ Λ  ≥ 1 2 α 1 − 1 η η X ℓ =0 g ( ℓ, η + 1) µ Λ ,ℓ ! + 1 2 β 1 − η X ℓ =1 2( η + 1) ( ℓ + 1) η ν Λ ,ℓ ! + (5.10) + γ η − 1 X ℓ =0 ℓ ℓ + 1 κ Λ ,ℓ =: pLl and P  y Λ Λ  ≤ 1 2 α 1 + Λ − 1 N − 1 η η X ℓ =0 g ( ℓ, η − 1) µ Λ ,ℓ ! + 1 2 β 1 + Λ − 1 N − η X ℓ =1 2( ℓη − 1) ( ℓ + 1) ℓη ν Λ ,ℓ ! + + γ η − 1 X ℓ =0 ℓ ℓ + 1 κ Λ ,ℓ =: pLu, (5.11) whereb y µ Λ ,ℓ is giv en b y setting λ = Λ in (2.3 ). Moreo v er, P  y N Λ  ≥ 1 2 α 1 − 1 η η X ℓ =0 g ( ℓ, η + 1) µ Λ ,ℓ ! + 1 2 β =: pN l (5.12) and P  y N Λ  ≤ 1 2 α 1 + Λ − 1 N − 1 η η X ℓ =0 g ( ℓ, η − 1) µ Λ ,ℓ ! + 1 2 β =: pN u. (5.13) Considering again these b ounds for large η , w e obtain the appro ximations: P  y Λ Λ  ∼ 1 2 ( α + β ) − α 2 η η X ℓ =0 g ( ℓ, η ) µ Λ ,ℓ − β η X ℓ =1 1 ℓ + 1 ν Λ ,ℓ + γ η − 1 X ℓ =0 ℓ ℓ + 1 κ Λ ,ℓ =: pLa (5.14) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 17 as w ell as P  y N Λ  ∼ 1 2 ( α + β ) − α 2 η η X ℓ =0 g ( ℓ, η ) µ Λ ,ℓ =: pN a. (5.15) 6. Ev alua tion of Lar gest Segment Utiliza tion and Seletion of R outing Stra tegy With (4.23 ) and a detailed onsideration of w a v elength λ = Λ , w e pro v e in App endix C the main theoretial result: Theorem 6.1. The maximum se gment utilization pr ob ability is (6.1) max n ∈{ 1 ,...,N } max λ ∈{ 1 ,..., Λ } P  y n λ  = max  P  y 1 1  , P  y Λ Λ  , P  y N Λ  . It th us remains to ompute the three probabilities on the righ t hand side. W e ha v e no exat result in the most general setting (it w ould b e p ossible to giv e reursiv e form ulae, but these w ould b e prohibitiv ely omplex). Ho w ev er, w e ha v e giv en upp er and lo w er b ounds and appro ximations in Setions 4.4 and 5.3, whi h mat h rather w ell in most situations, as demonstrated in the next setion, and ha v e the same asymptotis when η → ∞ while Λ remains xed. T o w ard assessing the onsidered shortest-path routing strategy , w e diretly observ e, that P  y N Λ  is alw a ys less or equal to 1 2 . On the other hand, the rst t w o usage probabilities will, for γ large enough, b eome larger than 1 2 , esp eially for hotsp ot soure tra with mo derate to large fanouts. Hene, shortest-path routing will result in a m ultiast apait y of less than t w o for large p ortions of hotsp ot soure m ulti- and broadast tra, whi h ma y arise in on ten t distribution, su h as for IP TV. The in tuitiv e explanation for the high utilization of the segmen ts y 1 1 and y Λ Λ with shortest-path routing for m ulti- and broadast hotsp ot soure tra is a follo ws. Consider the transmission of a giv en hotsp ot soure tra pa k et with destinations on w a v elength Λ homing the hotsp ot. If the pa k et has a single destination uniformly distributed among the other η − 1 no des homed on w a v elength Λ , then the CLG is adjaen t and to the left (i.e., in the oun ter lo  kwise sense) of the hotsp ot with probabilit y one half. Hene, with probabilit y one half a pa k et op y is sen t in the lo  kwise diretion, utilizing the segmen t y Λ Λ . With an inreasing n um b er of uniformly distributed destination no des on w a v elength Λ , it b eomes less lik ely that the CLG is adjaen t and to the left of the hotsp ot, resulting in inreased utilization of segmen t y Λ Λ . In the extreme ase of a broadast destined from the hotsp ot to all other η − 1 no des homed on Λ , the CLG is adjaen t and to the left of the hotsp ot with probabilit y 1 /η , i.e., segmen t y Λ Λ is utilized with probabilit y 1 − 1 /η . With probabilit y 1 − 2 /η the CLG is not adjaen t to the hotsp ot, resulting in t w o pa k et op y transmissions, i.e., a pa k et op y is sen t in ea h ring diretion. MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 18 F or w a v elength 1, the situation is subtly dieren t due to the rotational oset of the no des homed on w a v elength 1 from the hotsp ot. That is, no de 1 has a hop distane of 1 from the hotsp ot (in the lo  kwise diretion), whereas the highest indexed no de on w a v elength 1, namely no de ( η − 1)Λ + 1 has a hop distane of Λ − 1 from the hotsp ot (in the oun ter lo  kwise diretion). As for w a v elength Λ , for a giv en pa k et with a single uniformly distributed destination on w a v elength 1, the CLG is adjaen t and to the left of the hotsp ot with probabilit y one half, and the pa k et onsequen tly utilizes segmen t y 1 1 with probabilit y one half. With inreasing n um b er of destinations, the probabilit y of the CLG b eing adjaen t and to the left of the hotsp ot dereases, and the utilization of segmen t y 1 1 inreases, similar to the ase for w a v elength Λ . F or a broadast destined to all η no des on w a v elength 1, the situation is dieren t from w a v elength Λ , in that the CLG is nev er adjaen t to the hotsp ot, i.e., the hotsp ot alw a ys sends t w o pa k et opies, one in ea h ring diretion. 6.1. One-Cop y (OC) Routing. T o o v erome the high utilization of the segmen ts y 1 1 and y Λ Λ due to hotp ot soure m ulti- and broadast tra, w e prop ose one- opy (OC) r outing : With one-op y routing, uniform tra and hotsp ot destination tra are still serv ed using shortest path routing. Hotsp ot soure tra is serv ed using the follo wing oun ter-based p oliy . W e dene the oun ter Y λ to denote the n um b er of no des homed on λ that w ould need to b e tra v ersed to rea h all destinations on λ with one pa k et transmission in the lo  kwise diretion (whereb y the nal rea hed destination no de oun ts as a tra v ersed no de). If Y λ < η / 2 , then one pa k et op y is sen t in the lo  kwise diretion to rea h all destinations. If Y λ > η / 2 , then one pa k et op y is sen t in the oun ter lo  kwise diretion to rea h all destinations. Ties, i.e., Y λ = η / 2 , are serv ed in either lo  kwise or oun ter lo  kwise diretion with probabilit y one half. F or hotsp ot soure tra with arbitrary tra fanout, this oun ter-based one-op y routing ensures a maxim um utilization of one half on an y ring segmen t. Note that the oun ter-based p oliy onsiders only the no des homed on the onsidered w a v elength λ to ensure that the rotational oset b et w een the w a v elength Λ homing the hotsp ot and the onsidered w a v elength λ do es not aet the routing deisions. W e prop ose the follo wing strategy for swit hing b et w een shortest path (SP) and one-op y (OC) routing. Shortest path routing is emplo y ed if b oth (4.26 ) and (5.14 ) are less than one half. If (4.26 ) or (5.14 ) exeeds one half, then one-op y routing is used. F or the pratial implemen tation of this swit hing strategy , the hotsp ot an p erio dially estimate the urren t tra parameters, i.e., the tra p ortions α , β , and γ as w ell as the orresp onding fanout distributions µ l , ν l , and κ l , l = 1 , . . . , N − 1 , for instane, through a om bination of tra measuremen ts and histori tra patterns, similar to [64 , 65, 66, 67, 68℄. F rom these tra parameter estimates, the hotsp ot an then ev aluate (4.26 ) and (5.14 ). MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 19 T o obtain a more rened riterion for swit hing b et w een shortest path routing and one-op y routing w e pro eed as follo ws. W e  haraterize the maxim um segmen t utilization with shortest path routing more expliitly b y inserting (4.26 ), (5.14 ), and (5.15 ) in (6.2 ) to obtain: max n ∈{ 1 ,...,N } max λ ∈{ 1 ,..., Λ } P  y n λ  = 1 2 ( α + β ) − α 2 η η X ℓ =0 g ( ℓ, η ) µ 1 ,ℓ + max ( 0 , − β 2 η η X ℓ =0 g ( ℓ, η ) ν 1 ,ℓ + γ η X ℓ =0 ℓ ℓ + 1 κ 1 ,ℓ , − β η X ℓ =1 1 ℓ + 1 ν Λ ,ℓ + γ η − 1 X ℓ =0 ℓ ℓ + 1 κ Λ ,ℓ ) , (6.2) whereb y w e noted that the denition of µ λ,ℓ in (2.3 ) diretly implies that µ λ,ℓ is indep enden t of λ . Clearly , the hotsp ot soure tra do es not inuene the maxim um segmen t utilization as long as γ ≤ γ th 1 , 1 := β 2 η P η ℓ =0 g ( ℓ, η ) ν 1 ,ℓ P η ℓ =1 ℓ ℓ +1 κ 1 ,ℓ (6.3) and γ ≤ γ th 1 , Λ := β P η ℓ =1 1 ℓ +1 ν Λ ,ℓ P η − 1 ℓ =1 ℓ ℓ +1 κ Λ ,ℓ . (6.4) Th us, if γ ≤ γ th 1 = min( γ th 1 , 1 , γ th 1 , Λ ) , then all tra is serv ed using shortest path routing. W e next note that Theorem 6.1 do es not hold for the one-op y routing strategy . W e therefore b ound the maxim um segmen t utilization probabilit y with one-op y routing b y observing that (4.9 ) together with Prop osition 3.2 and (4.2 ) implies that asymptotially for all λ ∈ { 1 , . . . , Λ } (6.5) P α  y n λ  ∼ 1 2 − 1 2 η η − 1 X ℓ =0 g ( ℓ, η ) µ λ,ℓ . Hene, P α  y n λ  is asymptotially onstan t. Moreo v er, similar as in the single w a v elength ase [ 62 ℄, w e ha v e (6.6) P β  y n λ  ≤ P β  y N Λ  = 1 2 . Therefore, the maxim um segmen t utilization with one-op y routing is (appro ximately) b ounded b y max n ∈{ 1 ,...,N } max λ ∈{ 1 ,..., Λ } P  y n λ  ≤ 1 2 ( α + β + γ ) − α 2 η η − 1 X ℓ =0 g ( ℓ, η ) µ 1 ,ℓ . (6.7) Comparing (6.7 ) with (6.2 ) w e observ e that the maxim um segmen t utilization with one-op y routing is smaller than with shortest path routing if the follo wing threshold onditions hold: • If P η ℓ =1 ℓ ℓ +1 κ 1 ,ℓ > 1 2 , then set γ th 2 , 1 = β 2 η P η ℓ =0 g ( ℓ, η ) ν 1 ,ℓ P η ℓ =1 ℓ ℓ +1 κ 1 ,ℓ − 1 2 , (6.8) otherwise set γ th 2 , 1 = ∞ . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 20 • If P η − 1 ℓ =1 ℓ ℓ +1 κ Λ ,ℓ > 1 2 , then set γ th 2 , Λ := β P η ℓ =1 1 ℓ +1 ν Λ ,ℓ P η − 1 ℓ =1 ℓ ℓ +1 κ Λ ,ℓ − 1 2 , (6.9) otherwise set γ th 2 , Λ = ∞ . If γ ≥ γ th 2 = max( γ th 2 , 1 , γ th 2 , Λ ) , then one-op y routing is emplo y ed. F or γ v alues b et w een γ th 1 and γ th 2 , the hotsp ot ould n umerially ev aluate the maxim um seg- men t utilization probabilit y of shortest path routing with the deriv ed appro ximations. The hotsp ot ould also obtain the segmen t utilization probabilities with one-op y routing through disrete ev en t sim ulations to determine whether shortest path routing or one-op y routing of the hotsp ot tra is preferable for a giv en set of tra parameter estimates. 7. Numerial and Simula tion Resul ts In this setion w e presen t n umerial results obtained from the deriv ed b ounds and appro ximations of the utilization probabilities as w ell as v erifying sim ulations. W e initially sim ulate individual, sto  hastially indep enden t pa k ets generated aording to the tra mo del of Setion 2 and routed aording to the shortest path routing p oliy . W e determine estimates of the utilization probabilities of the three segmen ts y 1 1 , y Λ Λ , and y N Λ and denote these probabilities b y p 1 s , pLs , and pN s . Ea h sim ulation is run un til the 99% ondene in terv als of the utilization probabilit y estimates are less than 1% of the orresp onding sample means. W e onsider a net w orks with Λ = 4 w a v elength  hannels in ea h ring diretion. 7.1. Ev aluation of Segmen t Utilization Probabilit y Bounds and Appro ximations for Shortest P ath Routing. W e examine the auray of the deriv ed b ounds and appro ximations b y plotting the segmen t utilization probabilities as a funtion of the n um b er of net w ork no des N = 8 , 12 , 16 , . . . , 256 and omparing with the orresp onding sim ulation results. F or the rst set of ev aluations, w e onsider m ultiast tra with xed fanout µ 1 = ν 1 = κ 1 = 1 / 4 and µ l = ν l = κ l = 3 / (4 ( N − 2)) for l = 2 , . . . , N − 1 . W e examine inreasing p ortions of hotsp ot tra b y setting α = 1 , β = γ = 0 for Fig. 7.1, α = 0 . 6 , β = 0 . 1 , and γ = 0 . 3 for Fig. 7.2, and α = 0 . 2 , β = 0 . 2 , and γ = 0 . 6 for Fig. 7.3. W e onsider these senarios with hotsp ot tra dom- inated b y hotsp ot soure tra, i.e., with γ > β , sine man y m ultiast appliations in v olv e tra distribution b y a hotsp ot, e.g., for IP TV. W e also onsider a xed tra mix α = 0 . 2 , β = 0 . 2 , and γ = 0 . 6 for inreasing fanout. W e onsider uniast (UC) tra with µ 1 = ν 1 = κ 1 = 1 in Fig. 7.4, mixed tra (MI) with µ 1 = ν 1 = κ 1 = 1 / 2 and µ l = ν l = κ l = 1 / (2( N − 2)) for l = 2 , . . . , N − 1 in Fig. 7.5, m ultiast (MC) tra MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 21 0 0.1 0.2 0.3 0.4 0.5 0.6 0 50 100 150 200 250 Utilization Probability Number of Nodes p1u p1s p1a p1l 0 0.1 0.2 0.3 0.4 0.5 0.6 0 50 100 150 200 250 Utilization Probability Number of Nodes pLu pLs pLa pLl 0 0.1 0.2 0.3 0.4 0.5 0.6 0 50 100 150 200 250 Utilization Probability Number of Nodes pNu pNs pNa pNl (a) P  y 1 1  (b) P  y 4 4  () P  y 64 4  Figure 7.1. Segmen t utilization probabilit y as a funtion of n um b er of No des N for α = 1 , β = 0 , γ = 0 , and µ 1 = ν 1 = κ 1 = 1 / 4 and µ l = ν l = κ l = 3 / (4( N − 2)) for l = 2 , . . . , N − 1 . 0.1 0.2 0.3 0.4 0.5 0.6 0 50 100 150 200 250 Utilization Probability Number of Nodes p1u p1s p1a p1l 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 50 100 150 200 250 Utilization Probability Number of Nodes pLu pLs pLa pLl 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 50 100 150 200 250 Utilization Probability Number of Nodes pNu pNs pNa pNl (a) P  y 1 1  (b) P  y 4 4  () P  y 64 4  Figure 7.2. Segmen t utilization probabilit y as a funtion of n um b er of No des N for α = 0 . 6 , β = 0 . 1 , γ = 0 . 3 , and µ 1 = ν 1 = κ 1 = 1 / 4 and µ l = ν l = κ l = 3 / (4( N − 2)) for l = 2 , . . . , N − 1 . 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0 50 100 150 200 250 Utilization Probability Number of Nodes p1u p1s p1a p1l 0.1 0.2 0.3 0.4 0.5 0.6 0 50 100 150 200 250 Utilization Probability Number of Nodes pLu pLs pLa pLl 0.1 0.12 0.14 0.16 0.18 0.2 0.22 0 50 100 150 200 250 Utilization Probability Number of Nodes pNu pNs pNa pNl (a) P  y 1 1  (b) P  y 4 4  () P  y 64 4  Figure 7.3. Segmen t utilization probabilit y as a funtion of n um b er of No des N for α = 0 . 2 , β = 0 . 2 , γ = 0 . 6 , and µ 1 = ν 1 = κ 1 = 1 / 4 and µ l = ν l = κ l = 3 / (4( N − 2)) for l = 2 , . . . , N − 1 . with µ l = ν l = κ l = 1 / ( N − 1) for l = 1 , . . . , N − 1 in Fig. 7.6 , and broadast (BC) tra with µ N − 1 = ν N − 1 = κ N − 1 = 1 in Fig. 7.7. MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 22 0 0.05 0.1 0.15 0.2 0.25 0.3 0 50 100 150 200 250 Utilization Probability Number of Nodes p1u p1s p1a p1l 0 0.05 0.1 0.15 0.2 0.25 0 50 100 150 200 250 Utilization Probability Number of Nodes pLu pLs pLa pLl 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 50 100 150 200 250 Utilization Probability Number of Nodes pNu pNs pNa pNl (a) P  y 1 1  (b) P  y 4 4  () P  y 64 4  Figure 7.4. Segmen t utilization probabilit y as a funtion of n um b er of No des N for α = 0 . 2 , β = 0 . 2 , γ = 0 . 6 , and uniast (UC) tra with µ 1 = ν 1 = κ 1 = 1 . 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 50 100 150 200 250 Utilization Probability Number of Nodes p1u p1s p1a p1l 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 50 100 150 200 250 Utilization Probability Number of Nodes pLu pLs pLa pLl 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22 0 50 100 150 200 250 Utilization Probability Number of Nodes pNu pNs pNa pNl (a) P  y 1 1  (b) P  y 4 4  () P  y 64 4  Figure 7.5. Segmen t utilization probabilit y as a funtion of n um b er of No des N for α = 0 . 2 , β = 0 . 2 , γ = 0 . 6 , for mixed (MI) tra with µ 1 = ν 1 = κ 1 = 1 / 2 and µ l = ν l = κ l = 1 / (2( N − 2)) for l = 2 , . . . , N − 1 . 0.3 0.4 0.5 0.6 0.7 0 50 100 150 200 250 Utilization Probability Number of Nodes p1u p1s p1a p1l 0.2 0.3 0.4 0.5 0.6 0.7 0 50 100 150 200 250 Utilization Probability Number of Nodes pLu pLs pLa pLl 0.1 0.12 0.14 0.16 0.18 0.2 0.22 0 50 100 150 200 250 Utilization Probability Number of Nodes pNu pNs pNa pNl (a) P  y 1 1  (b) P  y 4 4  () P  y 64 4  Figure 7.6. Segmen t utilization probabilit y as a funtion of n um b er of No des N for α = 0 . 2 , β = 0 . 2 , γ = 0 . 6 , for m ultiast (MC) tra with µ l = ν l = κ l = 1 / ( N − 1) for l = 1 , . . . , N − 1 . W e observ e from these gures that the b ounds get tigh t for mo derate to large n um b ers of no des N and that the appro ximations  haraterize the atual utilization probabilities fairly aurately for the full range of N . F or instane, for N = 64 no des, the dierene b et w een the upp er and lo w er MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 23 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0 50 100 150 200 250 Utilization Probability Number of Nodes p1u p1s p1a p1l 0.3 0.4 0.5 0.6 0.7 0.8 0 50 100 150 200 250 Utilization Probability Number of Nodes pLu pLs pLa pLl 0.14 0.16 0.18 0.2 0.22 0.24 0 50 100 150 200 250 Utilization Probability Number of Nodes pNu pNs pNa pNl (a) P  y 1 1  (b) P  y 4 4  () P  y 64 4  Figure 7.7. Segmen t utilization probabilit y as a funtion of n um b er of No des N for α = 0 . 2 , β = 0 . 2 , γ = 0 . 6 , for broadast (BC) tra with µ N − 1 = ν N − 1 = κ N − 1 = 1 . b ound is less than 0.06, for N = 128 this dierene shrinks to less than 0.026. The magnitudes of the dierenes b et w een the utilization probabilities obtained with the analytial appro ximations and the atual sim ulated utilization probabilities are less than 0.035 for N = 64 no des and less than 0.019 for N = 128 for the wide range of senarios onsidered in Figs. 7.1 7.7 . (When exluding the broadast ase onsidered in Fig. 7.7 , these magnitude dierenes shrink to 0.02 for N = 64 no des and 0.01 for N = 128 no des.) F or some senarios w e observ e for small n um b er of no des N sligh t osillations of the atual utiliza- tion probabilities obtained through sim ulations, e.g., in Fig. 7.4(a) and 7.5 (a). More sp eially , w e observ e p eaks of the utilization probabilities for o dd η and v alleys for ev en η . These osillations are due to the disrete v ariations in the n um b er of destination no des leading to segmen t tra v ersals. F or instane, for the hotsp ot soure uniast tra that aoun ts for a γ = 0 . 6 p ortion of the tra in Fig. 7.4 (a), the utilization of segmen t y 1 1 is as follo ws. F or ev en η , there are η / 2 p ossible destination no des that result in tra v ersal of segmen t y 1 1 , ea h of these destination no des o urs with probabilit y 1 / ( N − 1) ; hene, segmen t y 1 1 is tra v ersed with probabilit y N/ [2Λ( N − 1)] . On the other hand, for o dd η , there are ( η + 1) / 2 p ossible destination no des that result in tra v ersal of segmen t y 1 1 ; hene, segmen t y 1 1 is tra v ersed with probabilit y ( N + Λ) / [2 Λ( N − 1)] . Ov erall, w e observ e from Fig 7.1 that for uniform tra, the three segmen ts go v erning the max- im um utilization probabilit y are ev enly loaded. With inreasing frations of non-uniform tra (with hotsp ot soure tra dominating o v er hotsp ot destination tra), the segmen ts y 1 1 and y 4 4 exp eriene inreasing utilization probabilities ompared to segmen t y 64 4 , as observ ed in Figs. 7.2 and 7.3 . Similarly , for the non-uniform tra senarios with dominating hotsp ot soure tra, w e observ e from Figs. 7.47.7 inreasing utilization probabilities for the segmen ts y 1 1 and y 4 4 ompared to segmen t y 64 4 with inreasing fanout. (In senarios with dominating hotsp ot destination tra, not MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 24 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Max Util Prob gamma BC, SP BC, OC MC, SP MC, OC MI, SP MI, OC UC, SP UC, OC 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Max Util Prob gamma BC, SP BC, OC MC, SP MC, OC MI, SP MI, OC UC, SP UC, OC (a) β = 0 . 1 (b) β = 0 . 2 Figure 7.8. Maxim um segmen t utilization probabilit y as a funtion of fration of hotsp ot soure tra γ (with α = 1 − β − γ ) for shortest path (SP) and one-op y routing (OC) for xed fration of hotsp ot tra β for uniast (UC) tra, mixed (MI) tra, m ultiast (MC) tra, and broadast (BC) tra. sho wn here due to spae onstrain ts, the utilization of segmen t y 64 4 inreases ompared to segmen ts y 1 1 and y 4 4 .) In Figs. 7.3, 7.6, and 7.7 , the utilization probabilities for segmen ts y 1 1 and y 4 4 exeed one half for senarios with mo derate to large n um b ers of no des (and orresp ondingly large fanouts), indiating the p oten tial inrease in m ultiast apait y b y emplo ying one-op y routing. 7.2. Comparison of Segmen t Utilization Probabilities for SP and OC Routing. In Fig. 7.8 w e ompare shortest path routing (SP) with one-op y routing (OC) for uniast (UC) tra, mixed (MI) tra, m ultiast (MC) tra, and broadast (BC) tra with the fanout distributions dened ab o v e for a net w ork with N = 128 no des. The orresp onding thresholds γ th 1 and γ th 2 are rep orted in T able 1. F or SP routing, w e plot the maxim um segmen t utilization probabilit y obtained from the analytial appro ximations. F or OC routing, w e estimate the utilization probabilities of all segmen ts in the net w ork through sim ulations and then sear h for the largest segmen t utilization probabilit y . F o using initially on uniast tra, w e observ e that b oth SP and OC routing attain the same maxim um utilization probabilities. This is to b e exp eted sine the routing b eha viors of SP and OC are iden tial when there is a single destination on a w a v elength. F or β = 0 . 1 , w e observ e with inreasing p ortion of hotsp ot soure tra γ an initial derease, a minim um v alue, and subsequen t inrease of the maxim um utilization probabilit y . The v alue of the maxim um utilization probabilit y for γ = 0 is due to the uniform and hotsp ot destination tra hea vily loading segmen t y 64 4 . With inreasing γ and onsequen tly dereasing α , the load on segmen t y 64 4 diminishes, while the load on segmen ts y 1 1 and y 4 4 inreases. F or appro ximately γ = 0 . 4 , the three segmen ts y 1 1 , y 4 4 , and y 64 4 are MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 25 F anout γ th 1 γ th 2 β = 0 . 1 UC 0.397 ∞ MI 0.059 7.32 MC 0.011 0.030 BC 0.0004 0.006 β = 0 . 2 UC 0.794 ∞ MI 0.118 14.64 MC 0.022 0.061 BC 0.0008 0.013 T able 1. Thresholds γ th 1 and γ th 2 for senarios onsidered in Fig. 7.8 ab out equally loaded. As γ inreases further, the segmen ts y 1 1 and y 4 4 exp eriene roughly the same, inreasing load. F or β = 0 . 2 w e observ e only the derease of the maxim um utilization probabilit y , whi h is due to the load on segmen t y 64 4 dominating the maxim um segmen t utilization. F or this larger fration of hotsp ot destination tra w e do not rea h the regime where segmen ts y 1 1 and y 4 4 go v ern the maxim um segmen t utilization. T urning to broadast tra, w e observ e that SP routing giv es higher maxim um utilization proba- bilities than OC routing for essen tially the en tire range of γ , rea hing utilization probabilities around 0.9 for high prop ortions of hotsp ot soure tra. This is due to the high loading of segmen ts y 1 1 and y 4 4 . In on trast, with OC routing, the maxim um segmen t utilization sta ys lose to 0.5, resulting in signian tly inreased apait y . The sligh t exursions of the maxim um OC segmen t utilization probabilit y ab o v e 1/2 are due to uniform tra. The segmen t utilization probabilit y with uniform tra is appro ximated (not b ounded) b y (6.5), making exursions ab o v e 1/2 p ossible ev en though hotsp ot destination and hotsp ot soure tra result in utilization probabilities less than (or equal) to 1/2. F or mixed and m ultiast tra, w e observ e for inreasing γ an initial derease, minim um v alue, and subsequen t inrease of the maxim um utilization probabilit y for b oth SP and OC routing. Similarly to the ase of uniast tra, these dynamis are aused b y initially dominating loading of segmen t y 64 4 , then a derease of the loading of segmen t y 64 4 while the loads on segmen ts y 1 1 and y 4 4 inrease. W e observ e for the mixed and m ultiast tra senarios with the same fanout for all three tra t yp es onsidered in Fig 7.8 that SP routing and OC routing giv e essen tially the same maxim um segmen t utilization for small γ up to a knee p oin t in the SP urv es. F or larger γ , OC routing giv es signian tly smaller maxim um segmen t utilizations. W e observ e from T able 1 that for relativ ely large fanouts (MC and BC), the ranges b et w een γ th 1 and γ th 2 are relativ ely small, limiting the need for resorting to n umerial ev aluation and sim ulation for determining whether to emplo y SP or OC MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 26 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Max Util Prob gamma SP,d=127 OC,d=127 SP,d=64 OC,d=64 SP,d=1 OC,d=1 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Max Util Prob gamma SP,d=127 OC,d=127 SP,d=64 OC,d=64 SP,d=1 OC,d=1 (a) ν 8 = 1 , κ d = 1 (b) ν d = 1 , κ 64 = 1 Figure 7.9. Maxim um segmen t utilization probabilit y as a funtion of fration of hotsp ot soure tra γ . Fixed parameters: N = 128 no des, β = 0 . 4 , µ l = 1 / 16 for l = 1 , . . . , 16 . Senario γ th 1 γ th 2 κ d = 1 d = 127 0.122 0.283 d = 64 0.126 0.302 d = 1 0.972 ∞ ν d = 1 d = 127 0.0017 0.028 d = 64 0.025 0.073 d = 1 0.212 0.456 T able 2. Thresholds γ th 1 and γ th 2 for senarios onsidered in Fig. 7.9 routing. F or small fanouts (UC and MI), the γ thresholds are far apart; further rened deision riteria for routing with SP or OC are therefore an imp ortan t diretion for future resear h. W e ompare shortest path (SP) and one-op y (OC) routing for senarios with dieren t fanout distribution for the dieren t tra t yp es in Fig. 7.9 for a ring with N = 128 no des. W e observ e from Fig. 7.9 (a) that for hotsp ot soure tra with large fanout, SP routing a hiev es signian tly smaller maxim um segmen t utilizations than OC routing for γ v alues up to a ross-o v er p oin t, whi h lies b et w een γ th 1 and γ th 2 . Similarly , w e observ e from Fig. 7.9 (b) that for small γ , SP routing a hiev es signian tly smaller maxim um segmen t utilizations than OC routing for hotsp ot destination tra with small fanout. F or example, for uniast hotsp ot destination tra (i.e., ν 1 = 1 ), for γ = 0 . 21 , SP routing giv es a m ultiast apait y of C M = 3 . 72 ompared to C M = 3 . 19 with OC routing. By swit hing from SP routing to OC routing when the fration of hotsp ot soure tra γ exeeds 0.31, the smaller maxim um utilization probabilit y , i.e., higher m ultiast apait y an b e a hiev ed aross the range of frations of hotsp ot soure tra γ . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 27 8. Conlusion W e ha v e analytially  haraterized the segmen t utilization probabilities in a bi-diretional WDM pa k et ring net w ork with a single hotsp ot. W e ha v e onsidered arbitrary mixes of uniast, m ultiast, and broadast tra in om bination with an arbitrary mix of uniform, hotsp ot destination, and hotsp ot soure tra. F or shortest-path routing, w e found that there are three segmen ts that an attain the maxim um utilization, whi h in turn limits the maxim um a hiev able long-run a v erage m ultiast pa k et throughput (m ultiast apait y). Through v erifying sim ulations, w e found that our b ounds and appro ximations of the segmen t utilization probabilities, whi h are exat in the limit for man y no des in a net w ork with a xed n um b er of w a v elength  hannels, are fairly aurate for net w orks with on the order of ten no des reeiving on a w a v elength. Imp ortan tly , w e observ ed from our segmen t utilization analysis that shortest-path routing do es not maximize the a hiev able m ultiast pa k et throughput when there is a signian t p ortion of m ulti- or broadast tra emanating from the hotsp ot, as arises with m ultimedia distribution, su h as IP TV net w orks. W e prop osed a one-op y routing strategy with an a hiev able long run a v erage m ultiast pa k et throughout of ab out t w o sim ultaneous pa k et transmissions for su h distribution senarios. This study fo used on the maxim um a hiev able m ultiast pa k et throughput, but did not on- sider pa k et dela y . A thorough study of the pa k et dela y in WDM ring net w orks with a hotsp ot transp orting m ultiast tra is an imp ortan t diretion for future resear h. Appendix A. Definition of Enlar ged and Redued Ring as well as of Left ( A ← λ ) and Right Shifting ( A → λ ) of Set of A tive Nodes In this app endix, w e rst dene the enlarging and reduing of the set of " λ -ativ e no des" A λ := F λ ∪ { S } . Supp ose that |F λ | = ℓ . Dep ending on the setting, and with M λ denoting the set of no des homed on a giv en w a v elength λ , the set F λ is  hosen uniformly at random among • all subsets of M λ (uniform tra and for λ 6 = Λ also hotsp ot destination and soure tra), or • all subsets of M λ that on tain N (hotsp ot destination tra for λ = Λ sine N is alw a ys a destination for hotsp ot destination tra), or • all subsets of M λ that do not on tain N (hotsp ot soure tra for λ = Λ sine N is alw a ys the soure for hotsp ot soure tra). Assuming S / ∈ M λ , w e dene: MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 28 enlarged ring: W e enlarge the set M λ b y injeting an extra no de homed on λ b et w een ⌊ S ⌋ λ and ⌈ S ⌉ λ (and orresp ondingly Λ − 1 no des homed on the other w a v elengths). After a re- n umeration starting with 0 at the new no de (whi h is aordingly homed on w a v elength Λ after the re-n umeration), w e obtain M Λ ,η + 1 :=  m Λ   m ∈ { 0 , . . . , η }  . W e dene the enlarged set F + λ to equal the ren um b ered set F λ united with the new no de. This pro edure leads to a random set of ativ e no des A + λ = F + λ that is uniformly distributed among all subsets of M Λ ,η + 1 with ardinalit y ( ℓ + 1) on taining no de 0 . Note that the largest gap of the enlarged set is larger or equal to the largest gap of A λ . 4 =0 X 3 X X 1 X 2 S 3 X enlarge 1 X 2 X = =2 =3 4 Λ Λ Λ Λ Figure A.1. Example of enlarging M 3 for N = 16 , Λ = 4 . The sender homed on w a v elength 1 is represen ted b y S in the left illustration. The no des of M 3 are indiated b y longer ti k marks and the no des of F 3 are irled. The enlarged ring has a total of N + Λ = 20 no des, with η + 1 = 5 no des homed on ea h w a v elength. The added no de on w a v elength 3 is n um b ered with 0 and lies b et w een the former ⌊ S ⌋ λ and ⌈ S ⌉ λ . redued ring: W e transform the set M λ b y merging the no des ⌊ S ⌋ λ and ⌈ S ⌉ λ to a single ativ e no de (eliminating the Λ − 1 no des in b et w een). After re-n umeration starting with 0 at this merged no de, w e obtain an ativ e set A − λ on M Λ ,η − 1 . Dep ending on the ardinalit y of F λ ∩ {⌊ S ⌋ λ , ⌈ S ⌉ λ } the new ativ e set A − λ has ℓ + 1 , ℓ , or ℓ − 1 elemen ts. More sp eially , if neither the left- nor the righ t-shifted soure no de w as a destination no de, then |A − λ | = ℓ + 1 . If either the left- or the righ t-shifted soure no de w as a destination no de, then |A − λ | = ℓ . If b oth the left- and righ t-shifted soure no de w ere destination no des, then |A − λ | = ℓ − 1 . In ea h of these ases A − λ is uniformly distributed among all subsets of M λ,η − 1 with ardinalit y   A − λ   that on tains no de 0 . Observ e that in all ases, the largest gap of A − λ is smaller or equal to the largest gap of A λ . W e also dene the follo wing transformations: MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 29 3 X =0 X X X 3 1 2 S reduce 1 X 2 X = =2 Λ Λ Figure A.2. Example of reduing for N = 16 , Λ = 4 . The sender is represen ted b y S and the no des of M 3 ha v e longer ti k marks. The no des of F 3 are irled. The no des ⌊ S ⌋ λ and ⌈ S ⌉ λ (as w ell as the 3 no des in b et w een) are merged in to the no de n um b ered 0 in the righ t illustration. Left (oun ter lo  kwise) shifting: Sine S is uniformly distributed on { 1 , . . . , N } , the set (A.1) A ← λ := F λ ∪ {⌊ S ⌋ λ } is a random subset of M λ . W e an think of A ← λ as b eing  hosen uniformly at random among all subsets of M λ ha ving ardinalit y |A ← λ | and sub jet to the same onditions as F λ . Notie that |A ← λ | = |F λ | if ⌊ S ⌋ λ ∈ F λ and |A ← λ | = |F λ | + 1 otherwise. 3 X X 1 S 4 =0 X X 2 3 X left shift 1 X 2 X = =2 =3 Λ Λ Λ Figure A.3. Example of left shifting for N = 16 , Λ = 4 . The destination no des are irled on the left, and the ativ e no des are irled on the righ t. The no des are ren um b ered after the shifting, starting with the former sender at 0. Also, the ativ e no des is ren um b ered, starting with X 1 > 0 , the rst ativ e no de after the former sender. The former sender is therefore the last ativ e no de, i.e., X 4 = 0 . Righ t (lo  kwise) shifting: Analogously w e dene (A.2) A → λ := F λ ∪ {⌈ S ⌉ λ } . This is a random set  hosen uniformly at random among all subsets of M λ ha ving ardinalit y |A → λ | and sub jet to the same onditions as F λ . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 30 3 X X 1 X 2 S 3 X =0 right shift 1 X 2 X 3 =2 = Λ Λ Λ Λ Figure A.4. Example of righ t shifting for N = 16 , Λ = 4 . After ren um b ering, the former sender is X 3 = 0 . Appendix B. Pr oof of Pr oposition 5.1 on Bounds f or Pr obability tha t CLG st ar ts a t Node 0 f or Hotspot Destina tion Traffi f or λ = Λ Pr o of. Conditioned on S ∈ M Λ , w e obtain (B.1) q ℓ β ( G Λ = 0 | S ∈ M Λ ) = 1 ℓ + 1 . Hene, w e only ha v e to onsider the ase S / ∈ M Λ . W e will not expliitly write do wn this ondition. Consider the righ t shifting and denote b y G → Λ the starting p oin t of the  hosen largest gap of A → Λ . Sine N ≡ 0 is the only xed ativ e no de, the rst gap, i.e., { 0 , . . . , X Λ , 1 } , is the only one that nev er shrinks, while the last gap, i.e., { X Λ ,ℓ +1 , . . . , N } , is the only one that nev er gro ws. Therefore, q ℓ β ( G Λ = 0) ≤ q ℓ β ( G → Λ = 0) . (B.2) F or reasons of symmetry , w e ha v e q ℓ β ( G → Λ = 0 | ⌈ S ⌉ Λ / ∈ F Λ ) = q ℓ γ ( G Λ = 0) = 1 ℓ + 1 , (B.3) and q ℓ β ( G → Λ = 0 | ⌈ S ⌉ Λ ∈ F Λ ) = q ℓ − 1 γ ( G Λ = 0) = 1 ℓ . (B.4) The remaining probabilities an b e omputed as q ℓ β ( ⌈ S ⌉ Λ ∈ F Λ | S / ∈ M Λ ) = ℓ η , leading to the desired upp er b ound. Analogously , the left shifting yields a lo w er b ound, namely q ℓ β ( G Λ = 0 | ⌊ S ⌋ Λ 6 = 0) ≥ q ℓ β ( G ← Λ = 0 | ⌊ S ⌋ Λ 6 = 0) . (B.5) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 31 Again for reasons of symmetry , w e obtain (B.6) q ℓ β ( G ← Λ = 0 | ⌊ S ⌋ Λ / ∈ F Λ ) = 1 ℓ + 1 and (B.7) q ℓ β ( G ← Λ = 0 | ⌊ S ⌋ Λ ∈ F Λ \ { 0 } ) = 1 ℓ . Finally , w e ha v e, of ourse, q ℓ β ( ⌊ S ⌋ Λ ∈ F Λ | S / ∈ M Λ ) = ℓ η and q ℓ β ( ⌊ S ⌋ Λ ∈ F Λ \ { 0 } | S / ∈ M Λ ) = ℓ − 1 .  Appendix C. Pr oof of Theorem 6.1 on the Maximal Segment Utiliza tion Pr o of. Due to Equation (4.23), w e only ha v e to pro v e the ase of drop w a v elength Λ . Corollary 3.2 tell us that it sues to onsider the ritial segmen ts. Let n ≡ δ Λ with 1 ≤ δ < η b e a ritial segmen t for Λ . Analogously to the pro of of the domination priniple in [62℄, w e redue the domination priniple for hotsp ot destination tra to the statemen t (C.1) q ℓ β ( n ≥ G Λ > n − Λ) ≥ 1 η − δ q ℓ β ( G Λ > n − Λ) , and for hotsp ot soure tra to: (C.2) q ℓ γ ( G Λ = n ) ≥ 1 η − δ q ℓ γ ( G Λ ≥ n ) . 1 N Figure C.1. Illustration of statemen t (C.1): the mean slop e of a ertain p erio d is bigger or equal than the mean slop e o v er all later p erio ds In the γ (hotsp ot soure tra) setting, w e kno w that N is the sender, and th us A Λ ⊂ M Λ . Hene, w e do not need to onsider the no des on the other drop w a v elengths and the pro of is exatly the same as in the single w a v elength ase [ 62 ℄, see also gure C.2. W e will no w use the same strategy for the more ompliated pro of in the β (hotsp ot destination tra) setting. Let K n denote the n um b er of ativ e no des nding themselv es b et w een the no des N and n (lo  kwise), i.e., (C.3) K n := |A Λ ∩ { 1 , . . . , n − Λ } | . MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 32 1 N Figure C.2. Gamma setting: the usage probabilit y sta ys onstan t on non ritial edges F or k ∈ { 0 , . . . , ( n − 1) ∧ ( ℓ − 1) } w e denote q ℓ,k γ for the probabilit y measure q ℓ γ onditioned on K n = k . W e denote again n ≡ δ Λ for δ ∈ { 1 , . . . , η − 1 } . W e will sho w that (C.4) q ℓ β ( n − Λ < G Λ ≤ n ) ≥ 1 η − δ q ℓ β ( G Λ > n − Λ) . 1 N Figure C.3. Beta setting: the usage probabilit y  hanges along ea h segmen t In ase that S ∈ M Λ w e an again use the pro of of the one w a v elength senario. This is also true if S ∈ { 1 , . . . , n − Λ } , sine w e do not laim an ything ab out these no des. Hene, w e only ha v e to in v estigate the ase S ∈ { n − Λ + 1 , . . . , N } \ M Λ . F rom no w on w e assume this to b e the ase. W e deomp ose the left hand side in to t w o parts, (C.5) q ℓ β ( n − Λ < G Λ ≤ n ) = q ℓ β ( G Λ = n ) + q ℓ β ( G Λ = S, n − Λ < S < n ) . F or the rst summand of (C.5 ), w e pro eed similarly to the ase of a single w a v elength, namely q ℓ,k β ( G Λ = n ) = q ℓ,k β ( G Λ = n, G Λ ≥ n, ⌊ S ⌋ Λ 6 = n, n ∈ F Λ ) = q ℓ,k β  G Λ = n   G Λ ≥ n, ⌊ S ⌋ Λ 6 = n, n ∈ F Λ  × × q ℓ,k β  G Λ ≥ n   ⌊ S ⌋ Λ 6 = n, n ∈ F Λ  × q ℓ,k β ( ⌊ S ⌋ Λ 6 = n, n ∈ F Λ ) . (C.6) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 33 W e obtain q ℓ,k β ( G Λ = n | G ≥ n, ⌊ S ⌋ Λ 6 = n, n ∈ F Λ ) = q ℓ − k β ,N − n +Λ ( G Λ = 1 | ⌊ S ⌋ Λ 6 = 1 , 1 ∈ F Λ ) ≥ q ℓ − k β ,N − n +Λ ( G ← Λ = 1 | ⌊ S ⌋ Λ 6 = 1 , 1 ∈ F Λ ) . (C.7) This probabilit y an b e omputed preisely q ℓ − k β ,N − n +Λ ( G ← Λ = 1 | ⌊ S ⌋ Λ 6 = 1 , 1 ∈ F Λ ) = q ℓ − k − 1 γ ,N − n ( G Λ = 0) q ℓ − k β ,N − n +Λ ( ⌊ S ⌋ Λ / ∈ F Λ | ⌊ S ⌋ Λ 6 = 1 , 1 ∈ F Λ ) + + q ℓ − k − 2 γ ,N − n ( G Λ = 0) q ℓ − k β ,N − n +Λ ( ⌊ S ⌋ Λ ∈ F Λ | ⌊ S ⌋ Λ 6 = 1 , 1 ∈ F Λ ) = 1 ℓ − k  1 − ℓ − k − 1 η − δ  + 1 ℓ − k − 1 ℓ − k − 1 η − δ = 1 ℓ − k  1 + 1 η − δ  . (C.8) W e no w use the fat that, onditionally on S ∈ { n − Λ + 1 , . . . , N } \ M Λ , (C.9) q ℓ,k β ( ⌊ S ⌋ Λ 6 = n ∈ F Λ ) = q ℓ,k β ( ⌊ S ⌋ Λ 6 = n ) q ℓ,k β ( n ∈ F Λ ) and (C.10) q ℓ,k β  G Λ ≥ n   ⌊ S ⌋ Λ 6 = n ∈ F Λ  = q ℓ,k β  G Λ ≥ n   n ∈ F Λ  q ℓ,k β ( ⌊ S ⌋ Λ 6 = n ) . Hene, w e obtain with q ℓ,k β ( n ∈ F Λ ) = ℓ − k − 1 η − δ that q ℓ,k β ( G Λ = n ) ≥ 1 η − δ q ℓ,k β  G Λ ≥ n   n ∈ F Λ  ×  1 − 1 ℓ − k   1 + 1 η − δ  . (C.11) F or the seond part of (C.5 ), w e obtain q ℓ,k β ( G Λ ∈ I δ \ n ) = q ℓ,k β ( G Λ = S, G Λ ≥ S, ⌈ S ⌉ Λ = n, n / ∈ F Λ ) = q ℓ,k β  G Λ = S   G Λ ≥ S, ⌈ S ⌉ Λ = n, n / ∈ F Λ  × × q ℓ,k β  G Λ ≥ S   ⌈ S ⌉ Λ = n / ∈ F Λ  q ℓ,k β ( ⌈ S ⌉ Λ = n, n / ∈ F Λ ) . (C.12) W e ha v e q ℓ,k β ( G Λ = S | G Λ ≥ S, ⌈ S ⌉ Λ = n, n / ∈ F Λ ) = q ℓ − k β ,N − n +Λ ( G Λ = S | ⌈ S ⌉ Λ = 1 , 1 / ∈ F Λ ) . (C.13) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 34 No w, w e use that |A → Λ | = |F Λ + 1 | for ⌈ S ⌉ Λ / ∈ F Λ . Hene, w e obtain q ℓ − k β ,N − n +Λ ( G Λ = S | ⌈ S ⌉ Λ = 1 , 1 / ∈ F Λ ) ≥ q ℓ − k β ,N − n +Λ ( G → Λ = 1 | ⌈ S ⌉ Λ = 1 , 1 / ∈ F Λ ) = q ℓ − k − 1 γ ,N − n ( G Λ = 0) = 1 ℓ − k . (C.14) Note that, onditioned on S ∈ { n − Λ + 1 , . . . , N } \ M Λ , w e ha v e (C.15) q ℓ,k β ( ⌈ S ⌉ Λ = n / ∈ F Λ ) = q ℓ,k β ( ⌈ S ⌉ Λ = n ) q ℓ,k β ( n / ∈ F Λ ) and (C.16) q ℓ,k β  G Λ ≥ S   ⌈ S ⌉ Λ = n / ∈ F Λ  = q ℓ,k β  G Λ ≥ S   ⌈ S ⌉ Λ = n  q ℓ,k β ( n / ∈ F Λ ) . Summarizing, w e obtain, using q ℓ,k β ( ⌈ S ⌉ Λ = n ) = 1 η − δ , that q ℓ,k β ( G Λ ∈ I δ \ n ) ≥ 1 η − δ 1 ℓ − k q ℓ,k β  G Λ ≥ S   ⌈ S ⌉ Λ = n  . (C.17) It remains to sho w that q ℓ,k β ( G Λ > n − Λ) ≤  1 − 1 ℓ − k   1 + 1 η − δ  × × q ℓ,k β ( G Λ ≥ n | n ∈ F Λ ) + + 1 ℓ − k  1 − ℓ − k − 1 η − δ  × × q ℓ,k β ( G Λ ≥ S | ⌈ S ⌉ Λ = n, n / ∈ F Λ ) . (C.18) This an b e sho wn b y q ℓ,k β ( G Λ > n − Λ) = η − ( ℓ − k ) X i = δ q ℓ,k β ( G Λ ≥ i Λ | X k +1 = i Λ) q ℓ,k β ( X k +1 = i Λ) + + Λ − 1 X λ =1 q ℓ,k β ( G Λ ≥ i Λ − λ | X k +1 = i Λ − λ ) q ℓ,k β ( X k +1 = i Λ − λ ) ≤  1 − 1 ℓ − k  q ℓ,k β ( G Λ ≥ n | X k +1 = n ) + + 1 ℓ − k q ℓ,k β ( G Λ ≥ S | ⌈ S ⌉ Λ = n ) . (C.19) F or the last inequalit y , w e used that for i ∈ { δ, . . . , η − 1 } and λ ∈ { 0 , . . . , Λ − 1 } q ℓ,k β ( G Λ ≥ i Λ − λ | X k +1 = i Λ − λ ) ≤ q ℓ,k β ( G Λ ≥ n − λ | X k +1 = n − λ ) (C.20) MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 35 and, for reasons of symmetry , (C.21) q ℓ,k β ( X k +1 ∈ F Λ ) = 1 − 1 ℓ − k . The last step w e need is a omparison of ( C.18 ) and (C.19 ). The only dierene arises, when b oth of the ev en ts, { n ∈ F Λ } and {⌈ S ⌉ Λ = n } , tak e plae. Then, (C.22) q ℓ,k β ( G Λ ≥ S | ⌈ S ⌉ Λ = n ∈ F Λ ) = q ℓ,k β ( G Λ ≥ n | ⌈ S ⌉ Λ = n ∈ F Λ ) . This o urs with probabilit y q ℓ,k β ( n ∈ F Λ | ⌈ S ⌉ Λ = n ) = ℓ − k − 1 η − δ and explains the additional fator in the deomp osition (C.18 ).  A kno wledgement W e are grateful to Martin Herzog, formerly of EMT, INRS, and Ra vi Sesha hala of Arizona State Univ ersit y for assistane with the n umerial and sim ulation ev aluations. Referenes [1℄ F. Da vik, M. Yilmaz, S. Gjessing, and N. Uzun,  IEEE 802.17 Resilien t P a k et Ring T utorial, IEEE Communi-  ations Magazine , v ol. 42, no. 3, pp. 112118, Mar. 2004. [2℄ S. Spadaro, J. Solé-P areta, D. Careglio, K. W a jda, and A. Szyma«ski,  P ositioning of the RPR Standard in Con temp orary Op erator En vironmen ts, IEEE Network , v ol. 18, no. 2, pp. 3540, Mar h/April 2004. [3℄ P . Y uan, V. Gam biroza, and E. Knigh tly ,  The IEEE 802.17 Media A ess Proto ol for High-Sp eed Metrop olitan- Area Resilien t P a k et Rings, IEEE Network , v ol. 18, no. 3, pp. 815, Ma y/June 2004. [4℄ N. Ghani, J.-Y. P an, and X. Cheng,  Metrop olitan Optial Net w orks, Opti al Fib er T ele  ommuni ations , v ol. IVB, pp. 329403, 2002. [5℄ J. F ransson, M. Johansson, M. Roughan, L. Andrew, and M. A. Summereld,  Design of a Medium A ess Con trol Proto ol for a WDMA/TDMA Photoni Ring Net w ork, in Pr o ., IEEE GLOBECOM , v ol. 1, No v. 1998, pp. 307312. [6℄ M. A. Marsan, A. Biano, E. Leonardi, M. Meo, and F. Neri,  MA C Proto ols and F airness Con trol in WDM Multirings with T unable T ransmitters and Fixed Reeiv ers, IEEE Journal of Lightwave T e hnolo gy , v ol. 14, no. 6, pp. 12301244, June 1996. [7℄ M. A. Marsan, A. Biano, E. Leonardi, A. Morabito, and F. Neri,  AllOptial WDM MultiRings with Dier- en tiated QoS, IEEE Communi ations Magazine , v ol. 37, no. 2, pp. 5866, F eb. 1999. [8℄ M. A. Marsan, E. Leonardi, M. Meo, and F. Neri,  Mo deling slotted WDM rings with disretetime Mark o vian mo dels, Computer Networks , v ol. 32, no. 5, pp. 599615, Ma y 2000. [9℄ K. Bengi and H. R. v an As, Eien t QoS Supp ort in a Slotted Multihop WDM Metro Ring, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 20, no. 1, pp. 216227, Jan. 2002. [10℄ M. Herzog, M. Maier, and M. Reisslein,  Metrop olitan Area P a k et-Swit hed WDM Net w orks: A Surv ey on Ring Systems, IEEE Comm. Surveys and T ut. , v ol. 6, no. 2, pp. 220, Seond Quarter 2004. [11℄ C. S. Jelger and J. M. H. Elmirghani,  Photoni P a k et WDM Ring Net w orks Ar hiteture and P erformane, IEEE Communi ations Magazine , v ol. 40, no. 11, pp. 110115, No v. 2002. [12℄ M. J. Sp ener and M. A. Summereld,  WRAP: A Medium A ess Con trol Proto ol for W a v elengthRouted P assiv e Optial Net w orks, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 18, no. 12, pp. 16571676, De. 2000. [13℄ I. M. White, E. S.-T. Hu, Y.-L. Hsueh, K. V. Shrikhande, M. S. Rogge, and L. G. Kazo vsky , Demonstration and system analysis of the HORNET, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 21, no. 11, pp. 24892498, No v. 2003. [14℄ G. F eng, C. K. Siew, and T.-S. P . Y um,  Ar hitetural design and bandwidth demand analysis for m ultipart y video onferening on SONET/A TM rings, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 20, no. 8, pp. 15801588, Ot. 2002. [15℄ R. Bev erly and K. Clay , Wide-area IP m ultiast tra  haraterization, IEEE Network , v ol. 17, no. 1, pp. 815, Jan.-F eb. 2003. MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 36 [16℄ K. Sara and K. Almeroth,  Monitoring IP Multiast in the In ternet: Reen t adv anes and ongoing  hallenges, IEEE Communi ations Magazine , v ol. 43, no. 10, pp. 8591, Ot. 2005. [17℄ G. N. Rousk as, Optial la y er m ultiast: rationale, building blo  ks, and  hallenges, IEEE Network , v ol. 17, no. 1, pp. 6065, Jan.-F eb. 2003. [18℄ L. Sahasrabuddhe and B. Mukherjee, Multiast routing algorithms and proto ols: A tutorial, IEEE Network , v ol. 14, no. 1, pp. 90102, Jan.-F eb. 2000. [19℄ N. Singhal, L. Sahasrabuddhe, and B. Mukherjee,  Optimal m ultiasting of m ultiple ligh t-trees of dieren t band- width gran ularities in a WDM mesh net w ork with sparse splitting apabilities, IEEE/A CM T r ansations on Networking , v ol. 14, no. 5, pp. 11041117, Ot. 2006. [20℄ R. Ul-Mustafa and A. Kamal,  Design and pro visioning of WDM net w orks with m ultiast tra gro oming, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 24, no. 4, pp. 3753, Apr. 2006. [21℄ G.-S. P o o and Y. Zhou,  A new m ultiast w a v elength assignmen t algorithm in w a v elength-routed WDM net w orks, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 24, no. 4, pp. 212, Apr. 2006. [22℄ S. Sank aranara y anan and S. Subramaniam,  Comprehensiv e p erformane mo deling and analysis of m ultiasting in optial net w orks, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 21, no. 11, pp. 13991413, No v. 2003. [23℄ J. W ang, X. Qi, and B. Chen,  W a v elength assignmen t for m ultiast in all-optial WDM net w orks with splitting onstrain ts, IEEE/A CM T r ansations on Networking , v ol. 14, no. 1, pp. 169182, F eb. 2006. [24℄ X. Qin and Y. Y ang,  Multiast onnetion apait y of WDM swit hing net w orks with limited w a v elength on- v ersion, IEEE/A CM T r ansations on Networking , v ol. 12, no. 3, pp. 526538, June 2004. [25℄ A. M. Hamad and A. E. Kamal, A surv ey of m ultiasting proto ols for broadast-and-selet single-hop net w orks, IEEE Network , v ol. 16, no. 4, pp. 3648, July/August 2002. [26℄ C.-F. Hsu, T.-L. Liu, and N.-F. Huang,  Multiast tra s heduling in single-hop WDM net w orks with arbitrary tuning latenies, IEEE T r ansations on Communi ations , v ol. 52, no. 10, pp. 17471757, Ot. 2004. [27℄ H.-C. Lin and C.-H. W ang, A h ybrid m ultiast s heduling algorithm for singlehop WDM net w orks, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 19, no. 11, pp. 16541664, No v. 2001. [28℄ K. Naik, D. W ei, D. Krizan, and S.-Y. Kuo,  A reserv ation-based m ultiast proto ol for WDM optial star net w orks, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 22, no. 9, pp. 16701680, No v. 2004. [29℄ X. Zhang and C. Qiao,  On S heduling All-to-All P ersonalized Connetions and Cost-Eetiv e Designs in WDM Rings, IEEE/A CM T r ansations on Networking , v ol. 7, no. 3, pp. 435445, June 1999. [30℄ O. Gerstel, R. Ramasw ami, and G. H. Sasaki,  Cost-Eetiv e T ra Gro oming in WDM Rings, IEEE/A CM T r ansations on Networking , v ol. 8, no. 5, pp. 618630, Ot. 2000. [31℄ J. W ang, W. Cho, V. R. V em uri, and B. Mukherjee,  Impro v ed Approa hes for Cost-Eetiv e T ra Gro oming in WDM Ring Net w orks: ILP F orm ulations and Single-Hop and Multihop Connetions, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 19, no. 11, pp. 16451653, No v. 2001. [32℄ L.-W. Chen and E. Mo diano,  Eien t Routing and W a v elength Assignmen t for Reongurable WDM Ring Net w orks With W a v elength Con v erters, IEEE/A CM T r ansations on Networking , v ol. 13, no. 1, pp. 173186, F eb. 2005. [33℄ D.-R. Din,  Geneti algorithms for m ultiple m ultiast on WDM ring net w ork, Computer Communi ations , v ol. 27, no. 9, pp. 840856, June 2004. [34℄ I. F errel, A. Mettler, E. Miller, and R. Lib eskind-Hadas,  Virtual top ologies for m ultiasting with m ultiple originators in WDM net w orks, IEEE/A CM T r ansations on Networking , v ol. 14, no. 1, pp. 183190, F eb. 2006. [35℄ X. Jia, X.-D. Hu, L. Ruan, and J. Sun, Multiast routing, load balaning, and w a v elength assignmen t on tree of rings, IEEE Communi ations L etters , v ol. 6, no. 2, pp. 7981, F eb. 2002. [36℄ R. Lib eskind-Hadas, J. R. Hartline, K. Dresner, E. Dru k er, and K. Ra y , Multiast virtual top ologies in WDM paths and rings with splitting loss, in Pr o . Int. Conf. on Computer Communi ations and Networks , Ot. 2002, pp. 318321. [37℄ J. W ang, B. Chen, and R. Uma, Dynami w a v elength assignmen t for m ultiast in all-optial WDM net w orks to maximize the net w ork apait y , IEEE Journal on Sele te d A r e as in Communi ations , v ol. 21, no. 8, pp. 12741284, Ot. 2003. [38℄ C. Zhou and Y. Y ang,  Widesense Non blo  king Multiast in a Class of Regular Optial WDM Net w orks, IEEE T r ansations on Communi ations , v ol. 50, no. 1, pp. 126134, Jan. 2002. [39℄ M. V eeraragha v an, H. Lee, J. Anderson, and K. Y. Eng, A net w ork throughput omparison of optial metro ring ar hitetures, in Pr o ., OF C , Mar. 2002, pp. 763765. [40℄ A. Carena, V. D. F eo, J. M. Fino  hietto, R. Gaudino, F. Neri, C. Piglione, and P . P oggiolini,  RingO: An Exp erimen tal WDM Optial P a k et Net w ork for Metro Appliations, IEEE Journal on Sele te d A r e as in Com- muni ations , v ol. 22, no. 8, pp. 15611571, Ot. 2004. [41℄ M. A. Marsan, A. Biano, E. Leonardi, M. Meo, and F. Neri,  On the Capait y of MA C Proto ols for AllOptial WDM MultiRings with T unable T ransmitters and Fixed Reeiv ers, in Pr o ., IEEE INF OCOM , v ol. 3, Mar. 1996, pp. 12061216. MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 37 [42℄ J. He, S.-H. G. Chan, and D. H. K. T sang,  Multiasting in WDM Net w orks, IEEE Communi ations Surveys and T utorials , v ol. 4, no. 1, Third Quarter 2002. [43℄ M. Boro ditsky , C. F. Lam, S. L. W o o dw ard, N. J. F rigo, and M. D. F euer, P o w er managemen t for enhaned system p erformane of passiv e optial rings, IEE Pr o  e e dings Opto ele tr onis , v ol. 150, no. 3, pp. 229234, June 2003. [44℄ S. Aleksi and K. Bengi, Multiast-apable aess no des for slotted photoni ring net w orks, in Pr o . Eur op e an Confer en e on Opti al Communi ations (ECOC) , Sept. 2000, pp. 8384. [45℄ E. Shimada, S. F ujiw ara, K. Ok azaki, and I. Sasase, A transmission system for m ultiast tra with preallo ation s heme in WDM ring net w orks, in Pr o . IEEE Pai R im Conf. on Comm., Computers and Signal Pr o . , Aug. 2003, pp. 490493. [46℄ C. Piglione, M. Reisslein, and F. Neri, F air uni- and m ultiasting in a ring metro WDM net w ork, OSA Journal of Opti al Networking (JON) , v ol. 3, no. 8, pp. 601622, 2004. [47℄ M. Chaitou, G. Hebuterne, and H. Castel,  P erformane of m ultiast in WDM slotted ring net w orks, Computer Communi ations , v ol. 30, no. 2, pp. 219232, Jan. 2007. [48℄ M. S heutzo w, P . Seeling, M. Maier, and M. Reisslein,  Multiasting in a WDM-upgraded Resilien t P a k et Ring, OSA Journal of Opti al Networks , v ol. 6, no. 5, pp. 415421, Ma y 2007. [49℄ M. S heutzo w, M. Reisslein, M. Maier, and P . Seeling,  Multiast Capait y of P a k et-Swit hed Ring WDM Net w orks, IEEE T r ansations on Information The ory , v ol. 54, no. 2, pp. 623644, F eb. 2008. [50℄ H. Zähle and M. S heutzo w and M. Reisslein and M. Maier,  On the m ultiast apait y of unidiretional and bidiretional pa k et-swit hed WDM ring net w orks, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 25, no. 4, pp. 105119, Apr. 2007. [51℄ A. A. M. Saleh and J. M. Simmons,  Ar hitetural Priniples of Optial Regional and Metrop olitan A ess Net w orks, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 17, no. 12, pp. 24312448, De. 1999. [52℄ R. Berry and E. Mo diano,  Optimal T ranseiv er S heduling in WDM/TDM Net w orks, IEEE Journal on Sele te d A r e as in Communi ations , v ol. 23, no. 8, pp. 14791495, Aug. 2005. [53℄ H. Liu and F. A. T obagi,  T ra gro oming in WDM SONET UPSR rings with m ultiple line sp eeds, in Pr o ., IEEE INF OCOM , Mar. 2005, pp. 718729. [54℄ X.-Y. Li, P .-J. W an, and L. Liu,  Selet line sp eeds for single-h ub SONET/WDM ring net w orks, in Pr o ., IEEE Int. Conf. on Communi ations , June 2000, pp. 495499. [55℄ S. Subramaniam, M. Azizoglu, and A. Somani,  On the optimal plaemen t of w a v elength on v erters in w a v elength- routed net w orks, in Pr o ., IEEE INF OCOM , Mar. 1998, pp. 902909. [56℄ J. W ang, W. Cho, V. R. V em uri, and B. Mukherjee,  Impro v ed approa hes for ost-eetiv e tra gro oming in WDM ring net w orks: ILP form ulations and single-hop and m ultihop onnetions, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 19, no. 11, pp. 16451653, No v. 2001. [57℄ Y. Xu and X. Y ao,  Lo w er b ound on n um b er of ADMs in WDM rings with non uniform tra demands, Ele - tr onis L etters , v ol. 40, no. 13, pp. 824825, June 2004. [58℄ X. Zhang and C. Qiao,  An Eetiv e and Comprehensiv e Approa h for T ra Gro oming and W a v elength As- signmen t in SONET/WDM Rings, IEEE/A CM T r ansations on Networking , v ol. 8, no. 5, pp. 608617, Ot. 2000. [59℄ A. Zapata, M. Duser, J. Sp ener, P . Ba yv el, I. deMiguel, D. Breuer, N. Hanik, and A. Gladis h,  Next-generation 100-gigabit metro ethernet (100 GbME) using m ultiw a v elength optial rings, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 22, no. 11, pp. 24202434, No v. 2004. [60℄ I. Rubin and H. Hua,  Syn thesis and throughput b eha viour of WDM meshed-ring net w orks under non uniform tra loading, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 15, no. 8, pp. 15131521, Aug. 1997. [61℄ M. Chaitou, G. Hebuterne, and H. Castel,  Mo delling m ulti- hannel optial slotted rings with xed transmitter and xed reeiv ers, in Pr o ., A dvan e d Industrial Confer en e on T ele  ommuni ations/Servi e Assur an e with Partial and Intermittent R esour  es Confer en e/ E-L e arning on T ele  ommuni ations W orkshop , July 2005, pp. 115120. [62℄ M. an der Heiden, M. Sortais, M. S heutzo w, M. Reisslein, P . Seeling, M. Herzog, and M. Maier,  Multiast Capait y of Optial P a k et Ring for Hotsp ot T ra, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 25, no. 9, pp. 26382652, Sept. 2007. [63℄ M. Maier, M. S heutzo w, M. Herzog, and M. Reisslein,  Multiasting in IEEE 802.17 Resilien t P a k et Ring, OSA Journal of Opti al Networks , v ol. 5, no. 11, pp. 841857, No v. 2006. [64℄ A. Biano, J. M. Fino  hietto, G. Giarratana, F. Neri, and C. Piglione,  Measuremen t-based reonguration in optial ring metro net w orks, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 23, no. 10, pp. 31563166, Ot. 2005. [65℄ H. Elbiaze and O. Cherk aoui,  Exploiting self-similar tra analysis in net w ork resoure on trol: the IP o v er WDM net w orks ase, in Pr o ., IEEE Int. Conf. on A utonomi and A utonomous Systems and Int. Conf. on Networking and Servi es (ICAS-ICNS) , Ot. 2005, pp. 6571. MUL TICAST CAP A CITY OF OPTICAL WDM P A CKET RING F OR HOTSPOT TRAFFIC 38 [66℄ A. Elw alid, D. Mitra, I. Saniee, and I. Widja ja,  Routing and protetion in GMPLS net w orks: from shortest paths to optimized designs, IEEE/OSA Journal of Lightwave T e hnolo gy , v ol. 21, no. 11, pp. 28282838, No v. 2003. [67℄ A. Genata and B. Mukherjee,  Virtual-top ology adaptation for WDM mesh net w orks under dynami tra, IEEE/A CM T r ansations on Networking , v ol. 11, no. 2, pp. 236247, Apr. 2003. [68℄ E. Oki, K. Shiomoto, S. Ok amoto, W. Ima juku, and N. Y amanak a,  Heuristi m ulti-la y er optim um top ology design s heme based on tra measuremen t for IP + photoni net w orks, in Pr o ., Opti al Fib er Communi ation Confer en e and Exhibit (OF C) , Mar. 2002, pp. 104105.

Original Paper

Loading high-quality paper...

Comments & Academic Discussion

Loading comments...

Leave a Comment