Coded Multicasting for Content Delivery and Distributed Networks

This direction is split into other directions: zero-padding overhead, XOR-based network coding. Its last version stays readable and citable here.

Version 1 of 192 papers124 conceptsfrozen 2026-10-04

In traditional network design, storage and communication are treated as separate resources. We use storage to keep data close to the user, and we use communication to fetch what isn't there. However, a transformative line of research pioneered by Mohammad Ali Maddah-Ali and Salman Avestimehr suggests that these two resources are fundamentally interchangeable. By cleverly coordinating what we store in local caches, we can create "multicast opportunities" that allow a single transmission to satisfy multiple users simultaneously, even if they are requesting entirely different files.

The Intuition: Turning Storage into Side Information

The core mechanism of this research is the use of local storage to provide side information. To understand why this works, imagine two users, Alice and Bob. Alice has file $A$ in her cache and wants file $B$. Bob has file $B$ and wants file $A$.

In a traditional system, the server would send $B$ to Alice and $A$ to Bob—two separate transmissions. In a coded caching system, the server broadcasts a single packet: $A \oplus B$ (the bitwise XOR of the two files).

  • Alice takes the received $A \oplus B$, XORs it with the $A$ she already has, and extracts $B$.
  • Bob XORs the same packet with his $B$ and extracts $A$.

A single transmission has satisfied two distinct requests. As shown in Multi-Server Coded Caching by Abolfazl Motahari and Seyed Pooya Shariatpanahi, this gain scales with the number of users and servers, effectively multiplying the network's capacity without adding new spectrum.

The Two-Phase Strategy: Placement and Delivery

This "magic trick" requires careful coordination, typically split into two phases:

  1. Placement Phase: This occurs during off-peak hours when bandwidth is cheap. The goal is not just to store popular content, but to distribute fragments of files across the network such that every possible combination of future user requests creates a multicast opportunity. Saeid Sahraei and Mohammad Mohammadi Amiri have even demonstrated that coded placement—storing XORed combinations of files before they are requested—can outperform storing raw fragments in certain regimes.
  2. Delivery Phase: Once requests are made, the server identifies which users have which fragments as side information. It then crafts a series of XORed transmissions to satisfy everyone. Yi Ouyang and Amir Salman Avestimehr developed algorithms like SACM to handle "arbitrary" placement, ensuring that even if users don't follow the caching rules perfectly, the server can still find efficient coded combinations.

From Caching to Computing: Coded Shuffling

The same logic applies to distributed computing. In frameworks like MapReduce, the "shuffling" phase—where intermediate results are moved between servers—is often the primary bottleneck.

Songze Li and colleagues introduced Coded MapReduce and Coded TeraSort to solve this. By intentionally repeating "Map" tasks across different nodes, the system creates redundant data. This redundancy acts exactly like a cache. During the shuffle, nodes can broadcast XORed intermediate results, allowing multiple destination nodes to decode their required data using their own local computation results as side information. This approach, often called Coded Shuffling, can reduce communication load by a factor that scales linearly with the number of nodes.

Bridging Theory and Reality

While the initial theory assumed "perfect" conditions, recent work has adapted these concepts to the complexities of real-world networks:

  • Heterogeneity: In real networks, users have different cache sizes and files have different lengths. Hui Liu 0011 and Wenxin Li proposed using zero-padding to align XOR operations, while Alexander Michael Daniel and Wei Yu developed optimization frameworks to handle these variations efficiently.
  • Asynchrony: Users don't all request videos at the same time. Qianqian Yang and Deniz Gunduz showed that by using a slotted-time delivery system, servers can still find overlapping data requirements among asynchronous requests to maintain coded gains.
  • Wireless Constraints: In wireless environments, interference and fading are constant threats. Seyed Pooya Shariatpanahi and Giuseppe Caire have explored combining coded caching with multi-antenna beamforming (MIMO), showing that storage can effectively increase the "degrees of freedom" of the wireless channel.
  • Stragglers: In distributed learning, slow nodes (stragglers) can delay the entire process. Kangwook Lee and Dimitris Papailiopoulos demonstrated that "Coded Computation" uses redundancy to allow the server to reconstruct the final result even if some nodes fail or lag.

By treating local memory as a distributed resource for interference management rather than just a private silo for data, this research direction transforms the fundamental limits of how much information a network can carry.

Papers in this version

Showing 60 of 92 papers.

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_coded_multicasting_for_content_delivery__f2d20d11,
  title = {Coded Multicasting for Content Delivery and Distributed Networks},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/coded-multicasting-for-content-delivery-and-distributed-networks/rdr_f2d20d11fa004c6da283e0eefed89e2d},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_coded_multicasting_for_content_delivery__f2d20d11_v1,
  title = {Coded Multicasting for Content Delivery and Distributed Networks},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/coded-multicasting-for-content-delivery-and-distributed-networks/rdr_f2d20d11fa004c6da283e0eefed89e2d/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}