Estimation of available bandwidth and measurement infrastructure for Russian segment of Internet

1 Estimation of av ailable bandw idth and measure ment infrastruc ture for Russian segmen t of Interne t Platonov A.P. 1) , Sidelnikov D.I. 2) , Strizhov M.V. 3) , Sukhov A.M. 4) 1) Russian Institute for Public Networks, Moscow, Russia, e-mail: plat@ripn.net 2) Institute of Organic Chemistry of RAS, Moscow, Russia, e-mail: sid@freenet.ru 3) Samara State Aerospace University, Samara, Russia, e-ma il: strizhov@ip4tv.ru 4) Internet TV ltd., Samara, Russia, e-mail: amskh@yandex.ru Abstracts In paper the me thod for esti mation of available bandwidth is supp osed which does n ot demand the advanced utilities. Our method is based on the measu rement of network delay D for packets of different sizes W . The simple expression for available band width B av = (W 2 -W 1 )/(D 2 -D 1 ) is substantiated . For the experimental t esting the measurement infrastructure for Russian segment o f Internet was installed in framework of RFBR grant 06 - 07 -89074. C 2.3 Network Oper ations, Netw ork monitoring Introduction In order to co nstruct the fullest picture of a global network, it ’s monitoring, search of bottlenecks, and also development of t he standards describing new appendices, the modern measuring infrastructure should be installed. In Russia there are some measuring points of diff erent measurement projects in the area of networking, for example PingER [ 11 ] in Institute o f Theoretical and Experimental Ph ysics (ITEP) , but full access to the colle cted data is limited for researchers . Unfortunately, the current measurement points do not reflect structure of the Russian segment of a global network. The available bandwidth [6,9] of a network path is an imp ortant performance metric and its end- to -end estimation has recently received significant attention. The available band width is the maximum throughput that the path can provide to an application, given th e path's current cross traffic load. Mea suring available bandwidth is not only for knowing the ne t work status, but also t o provide information to net work applications on h ow to control their outgoing traffic and fairly share the network bandwidth. Bulk Transport Capacity (BTC) is a measure of a network's ability to transfer significa nt quantities of data with a single congestion - aware tran sport connection (e.g., TCP). The intuitive definition of BTC is the expected long-term average data rate (bits per second) of a single ideal TCP implementation over the path in question. Measuring the available ba ndwidth is of great importance for predicting the end- to -end performance of applications, for dynamic path selection and t raffic engineering, and for selecting between a numbers of d ifferentiated classes o f service. The available bandwidth is an important metric for se veral applications, such as grid, vid eo and voi ce streaming, overlay routing, p2p file transfers, server selection, and interdomain path monito ring. 2 In o rder to measure different capacity metric s the installation of special utilities [8] is de manded at both ends of p ath. This is uncomfortable p rocess especially for usual network users who try to install modern application like videoconference. At present time powerful measuremen t system like RIPE Test B ox is expanded [5]. Unfortunately, this sy stem doesn’t measure the available bandwidth but it collects the numerical values characterized the network heals like delay, jitter, routing path, etc. This data allows us to investigate the basic interdependences of available bandwidth from b asic network parameters. Our aim is to est imate the availab le bandwidth from the delay value received from one point of path. Model J. Padhye V. Firoiu D . Towsley and J. Kuro se [1 ,10] f ind th at TCP throughput may be calculated as D W B av (1) This formula may b e used f or calculation of available bandwidth between two net work p oints that are connected immediately. Here W is the size of transmitted packet and is t he packet delay. The delay value is caused by such constant network factors as propagation dela y, transmission delay, per-packet router processing time, etc [2]. Choi et al [2] proved that the f ixed delay component (i.e., the total propagation and transmission delay) for a packet of size p is a linear (or p recisely, an aff ine ) function of it s size. To validate this assumption, they check the minimum delay of packets of t he same size for each path, and plot t he minimum delay against the packet size. But the parameters of th eir linear equation will not be a simple function of the link c apacities and the propagation delays. Fig.1. Available bandwidth On t he Fig.1 the graphic of linear d ependence between n etwork delay and p acket size constructed in paper [2] shows computed minimal delay. Slope angle concerning Y axe could be considered as available bandwidth. Prolongation of thi s line to Y axe gives the intercept val ue a. 3 Then the equation (1) for the available bandwidth which path consists of two or more hops should be modernized to the following view: a D W B av (2) where a is the unknown function depended on n and l . Here n is the n umber of hops (routers) that gives the traceroute command and n n l l is th e sum of single length of routing path. The value is re lated to the distance between the sites plus the propagation delay and per-packet router processing time at each hop along the path between the sites [3,4]. It is easy to use the linear approach for calculation of a l n l n f a ) , ( (3) In any case the equation (2) gives us the simple way for estimation of active bandwidth. Our model supposes the variat ion of packet size on the same path for calculation of available bandwidth. If the testing pro cess between two fixed poin t s is organized by packets with different sizes 1 W and 2 W and then t he D times get also two different values. Equatio n (2) gives the identical value of available bandwidth independently from packet size. Th e system from two equations is easy solved to find av B and a : 2 1 2 1 D D W W B av (4) The sizes 1 W and 2 W should d ifferent in several times, it is reasonable t o c hoose 100 and 1124 bytes correspondingly. The basic problem is the precise of measurements for delay D that is necessary for accurat e result. The d elay should be measured with micro second pre cision, such accuracy could be reached from RIPE Test Box database. Routinely the special ut ilities could be used for delay measurements; we tried to test traditional ping, t he new UDPping and other utility. In result of test the simplest utility ping was f ound to be a best choice for delay measurements. For examp le, ADSL connection at my home gives , th at corresponds to 350 Kbps of available bandwidth . During FTP session the delay grows to 300 ms and 425 ms that corresponds approximately to 60 Kbps of available bandwidth . This is very rou gh computation, but it could be made quickly and independently. It shou ld be noted that Table II fro m paper [2] gives us th ese values; calculated slope is inverse value to available bandwidth. Th e corresponding bandwidths fo r data set 1, 2, 3 (path 1 and 2) are 285 Mbps , 128 Mb ps , 222 Mbps and 205 Mbps . The second test allows calculating the value a in alternative way. (6) The further data from RIPE test box database allo w calculating the coefficients from Eq.3. 4 The additional experiments on the different d irection are necessary for verifying our model. During experiment the received data should be c ompared with results of iperf u tility from NLANR.org. We have contacted with Baek-Young Cho i from University of Minn esota. It will be interesting to t est remote direction like Australia and New Zeland. Unfortunately, there is not a simple way to receive data with delays for different packet sizes in current configuration of RIPE Test Box. Measurement infrastructure In 20 06 our team develops two stationary measuring points in the FREENet network on the basis of Institute of Organic Che mistry of the Russian Acad emy of Science and in the Samara regional net work for a science and education at the Samara St ate Aerospace Un iversity within the limits of the RFBR gran t 06 - 07 - 89074 “ Creation of a measuring infrastructure for studying quality the Internet of appendices in the Russian segment o f a global network". Each of our point has two blocks: 1. Ripe Test Box [12] 2. Application analysis server (with tools such as Wireshark, H. 323 Beacon Tool [7], etc.). In December 2007 RIPE me asuring system totaled about 80 points scat tered worldwide. Practically all the cores world the Internet the cent ers are captured by system, data f or the analysis t he In ternet of an infrastructure act from Northern and South America, Au stralia, New Zealand and Japan. But the greatest quantity of measuring points, as well as it is necessary to expect, is in a zone of responsibility RIPE: the European countries (Fig. 2 ). Fig 2. RIPE Test Box in Europe 5 Measured values includes: 1. Delay of packages 2. Variation of a delay (jitter) 3. Losses of packages 4. Availability of DNS root servers 5. Statistics of routing map 6. Availability of satellites All these data are reduced together and results of their processing are accessible in the form of various schedules and tables. Work with data is convenient fo r beginning with starting page on the Internet to th e address: http://ripe.net/projects/ttm/Plots /. Each of Test Boxes has its own RIPE identification (for example, T est bo x in Samara called tt143.ripe.net). Basis for t he analysis is the summary table where d ata on a studied measuring point are together shown all, for example, tt143 is a point in Sama ra. In the table d ata about a time delay of p ackages are presented average percent of losses of packages in entering and proceeding traffic, and also: average size, the boundary sizes cutting first and last 2,5 % o f results. The data presented in the table , can be deciphered in more detail. Following a hyperlink in a corresponding cell of the table, it is possible to receive the additional information on dynamics of any of the measured sizes resulted ab ove. So in a Fig. 3 presented data on param eter of a de lay of packages between measuring points in Samara and Zu rich, collected during between December 12th, 2007 and on January 12th, 2008 are presented. 6 Fig. 3. Month statistics for delay On the top left schedule data about two variables are presented: 1. To delay of packages depending on time 2. Routing hops The presented schedule evidently shows that during supervision the route changed. The top right schedule shows density of distribution for a delay o f packages, at change of routing time of passage of a softw are package of a network tripled. Also the b ottom right schedule showing losses of packages by tran sfer which allocates two time periods with practically full termination of communication is interesting also. The variation of a delay is size which dynamics of change measuring system RIPE allows to analyze (Fig. 4 ). Fig. 4. Day statistics for packet loss The top left schedule describes density of distribution for a variation of a delay, on right - t ime dependence of distribution for a variation of a delay. On the b ottom pair schedules the density of distribution and time dependence for a delay of packages, and also dynamics of change of number of points of routing are presente d. Especially it would be desirable to note the service showing availability root D NS o f servers of domains of the first level. Th e given service automatically traces responses from almost 200 servers for 30 domains world wide a nd conducts st atistics of their availability. T he specified 7 service can be applied to monit oring national DNS systems, for t his purpose it is enough to deduce of all data about availability of this or that server in a separate page. The statistics of routing includes not only re cords of ways of the routin g received b y means of a command t raceroute, but also and some analytical sizes, as t hat distribution of n umber of routers and numbers of ind ependent systems on a way between two measuring p oints as it is shown in a Fig. 5. Fig. 5. AS plots for tt143 During operation of measuring system we h ad been eliminated some n etwork incidents, most significant of them concerns to detection of inaccessibility of a root server of a domain name of the first level. System engineers of German national domain DE (a.nic.de), an alyzed availability of the server by means of mea suring sy stem RIPE and h ad found out, that from a point in Samara and, accordin gly, from all independent system s of network RBNet there is no opportunity to serve inquiries to the above- stated server. Summary and feature work In this work we present the simple me thod for estimation of available bandwidth. The d etailed experiment data [2] allows us to modernize the expression for end - to -end available bandwidth and to derive the new formula where the well measured values are included. 8 We hope that our results will b e incorporated in measure ment mechanisms of RIPE test b oxes [5] as well as in other measurement packets like [ 7] H.323 Beacon. An y VVo IP applications demand the knowledge of this value for tuning of transmi tting process. The additional experiments should be organized where the detailed comparison the results received with our meth od and with iperf utility [8] . It is interesting to investigate how the TCP throughput and available bandwidth depends from packet loss value. References 1. Ben Fredj S., Bonald T ., Proutiere A., Regnie G., Roberts J. Statistical Bandwidth Sharing: A Study of Congestion at Flow Level // ACM SIGCOMM, August 2001 2. Choi, B.-Y., Moon, S., Zhang, Z.-L., Papagiannaki, K. and Diot, C., Analysis of Point - To -Point Packet Delay In an Operational Network. Infocom 2004 , Hong Kong, 2004, pp. 1797 -1807 3. Cottrell L., Matthews W. and Logg C. , Tutorial on Internet Monitoring & PingER at SLAC 4. Crovella M.E. and Carter R.L., Dynamic Server Selection in the Internet, In Proc. of t he Third IEEE Workshop on the_ Architecture and Implementation of High Pe rformance Communication_ Subsystems (HPCS’95) 5 . Georgatos F., Gruber F., Karrenberg D., Santcroos M., Susanj A., Uijterwaal H. and Wilhelm R., Providing active measurements as a regular servic e fo r ISP’s, PAM2001 6 . Guojun J. , Available Bandwidth Measurement and Sampling, http://www.caida.org/workshops/isma/0312/abstracts/g uojun.pdf 7. H.323 Beacon Tool, http://www.osc.edu/networking/itecohio.net/beacon/ 8. Iperf, dast.nlanr.net/Projects/Iperf/ 9. Jain M., Dovrolis K., End- to -end Estimation of the Availa ble Bandwidth V ariation Range, SIGMETRICS’05, June 6 – 10, 2005, Banff, Alberta, Canada 10 . Padhye J., Firoiu V., Towsley D., Kurose J., Modeling TCP Through put: A Simple Model and its Empirical Validation // Proc. SIGCOMM Symp. Communications Architectures and Protocols - Aug. 1998 - pp. 304- 314 11. PingER, http://www-iepm.slac.stanford.edu/pinger/ 12. Ripe Test Box, http://ripe.net/projects/ttm/Pl ots/.