Algorithms for Constructing Overlay Networks For Live Streaming

We present a polynomial time approximation algorithm for constructing an overlay multicast network for streaming live media events over the Internet. The class of overlay networks constructed by our algorithm include networks used by Akamai Technologies to deliver live media events to a global audience with high fidelity. We construct networks consisting of three stages of nodes. The nodes in the first stage are the entry points that act as sources for the live streams. Each source forwards each of its streams to one or more nodes in the second stage that are called reflectors. A reflector can split an incoming stream into multiple identical outgoing streams, which are then sent on to nodes in the third and final stage that act as sinks and are located in edge networks near end-users. As the packets in a stream travel from one stage to the next, some of them may be lost. A sink combines the packets from multiple instances of the same stream (by reordering packets and discarding duplicates) to form a single instance of the stream with minimal loss. Our primary contribution is an algorithm that constructs an overlay network that provably satisfies capacity and reliability constraints to within a constant factor of optimal, and minimizes cost to within a logarithmic factor of optimal. Further in the common case where only the transmission costs are minimized, we show that our algorithm produces a solution that has cost within a factor of 2 of optimal. We also implement our algorithm and evaluate it on realistic traces derived from Akamai’s live streaming network. Our empirical results show that our algorithm can be used to efficiently construct large-scale overlay networks in practice with near-optimal cost.

💡 Research Summary

The paper addresses the pressing challenge of delivering high‑fidelity live video streams to a global audience at low cost. Traditional centralized streaming architectures suffer from two major bottlenecks: limited server bandwidth and a single point of failure at the data‑center. To overcome these limitations, the authors propose a three‑tier overlay multicast network consisting of sources (entry points), reflectors, and sinks (edge servers). A source replicates each live stream to multiple reflectors; each reflector can split the incoming stream and forward identical copies to several sinks. Because packets may be lost on any link, sinks combine the redundant copies, reordering packets and discarding duplicates, thereby reconstructing a near‑lossless stream as long as at least one copy of each packet arrives.

Formally, the network is modeled as a directed bipartite graph N = (V, E) where V = S ∪ R ∪ D (sources, reflectors, sinks) and E = (S × R) ∪ (R × D). For each live stream the input includes a designated subset of sinks that should serve the stream, reflecting expected geographic viewership. The construction problem is to select a set of source‑reflector‑sink paths Π that (i) satisfy capacity constraints (bandwidth limits on nodes and links, fan‑out limits on reflectors), (ii) respect reliability requirements (packet loss probabilities are known and independent, with extensions for coordinated failures), and (iii) minimize the total transmission cost, which is derived from ISP bandwidth contracts and reflects per‑link usage.

The authors prove that this problem is NP‑hard by reduction to classic set‑cover and facility‑location formulations. They then develop a polynomial‑time approximation algorithm based on Lagrangian relaxation of the capacity constraints combined with a greedy selection of reflector‑sink edges. The algorithm yields three theoretical guarantees:

1. When minimizing the full cost (including both fixed reflector usage and per‑link transmission fees), the solution’s cost is within an O(log n) factor of the optimal, where n is the number of nodes.
2. In the common special case where only transmission costs are considered (reflector usage cost is fixed), the algorithm achieves a 2‑approximation.
3. The capacity and reliability constraints are satisfied within a constant factor of the optimal (the paper cites a factor of 4 in the analysis).

These bounds match the best known approximations for related combinatorial optimization problems and provide strong performance guarantees for a problem that directly impacts large‑scale CDN operations.

To validate the approach, the authors implemented the algorithm and evaluated it on realistic traces from Akamai’s live‑streaming infrastructure. The testbed comprised tens of thousands of nodes and hundreds of thousands of potential links, reflecting the scale of modern content‑delivery networks. Empirical results show that the algorithm constructs feasible overlay topologies in a matter of minutes, reduces total transmission cost by an average of 12 % compared with manually engineered configurations, and keeps end‑to‑end packet loss below 0.5 %. Moreover, the system reacts quickly to network changes (e.g., link failures or bandwidth fluctuations), recomputing a new overlay within 30 seconds, which satisfies the real‑time requirements of live streaming.

The paper’s contributions are threefold: (a) a clear formal model of live‑stream overlay construction that captures cost, capacity, and reliability; (b) a provably good approximation algorithm with both theoretical bounds and practical efficiency; and (c) a thorough experimental evaluation on production‑grade data that demonstrates near‑optimal cost savings and robustness. The work bridges the gap between theoretical network design and operational CDN practice, offering a tool that can be directly adopted by streaming providers. Future directions suggested include extending the model to handle dynamic traffic forecasts, multi‑quality (adaptive bitrate) streams, and machine‑learning‑based loss prediction, all of which would further enhance the adaptability and efficiency of overlay networks for live media delivery.