Network communications are facing increasing challenges from emerging applications with low-latency and high-throughput communication demands.
Limitations of Traditional Network Communications
Traditional network communication technologies, represented by TCP/IP, are characterized by a few core features: packets are transmitted over a fixed path for each communication, retransmission provides end-to-end reliability, and congestion control is performed end to end. These features place strong requirements on the reliability and latency of network links.
Underlying these mechanisms is the principle of flow conservation: for every user, at each intermediate node on the user’s path, the total incoming flow must equal the total outgoing flow. This requirement comes from classical network flow models such as the max-flow problem, and is natural for physical commodities like water or vehicles, where material cannot be created or lost in transit. Flow conservation is a fundamental principle of existing congestion control mechanisms — departures from it are traditionally associated with congestion and packet loss, and congestion control tends to produce conserved flows.
Such approaches are far from optimal in nearly all practical scenarios:
- Using one or a set of predetermined paths prevents achieving the maximum communication rate.
- End-to-end control mechanisms respond relatively slowly to dynamic changes in the network.
- Even at the expense of communication rate, wireless communication struggles to simultaneously ensure high reliability and stable latency.
Coding-based Network Communications
To fundamentally address these issues, it is necessary to appropriately adopt coding techniques, primarily erasure codes and network coding.
- Erasure codes can reduce the number of retransmissions in end-to-end network communication, lowering the delay caused by waiting for feedback. They also adapt to different path delays and bandwidth variations, enhancing multipath communication performance. Common erasure codes include Reed–Solomon codes and fountain codes.
- Network coding allows intermediate nodes to generate and transmit new data packets, breaking the limitations of the traditional store-and-forward model. In multicast communication and lossy network environments, linear network coding achieves the highest communication efficiency under given resource conditions, and random linear combination coefficients make distributed implementation possible.
Coding-based network communication also broadens the opportunities for enhancing network resource allocation and communication media:
- Congestion control must be tailored to the unique characteristics of coding to fully leverage its benefits.
- High reliability is not always a necessity for wireless communications.
Our Work
Following this direction, we study how to build a protocol suite that natively supports erasure coding. Our Coding-based Non-conserved Communication Protocol (CNCP) relaxes flow conservation: intermediate nodes may drop packets to absorb congestion, while receivers recover the lost data from coding redundancy. CNCP is formulated in a utility maximization framework with hop-by-hop control and a two-level queueing system, fairly shares bandwidth among users, and lets each user choose its own redundancy level. In data center and multipath network evaluations, CNCP achieves lower flow completion times than state-of-the-art schemes such as DCQCN, BFC, MPTCP, and FMTCP under heavy load.
Overall, coding-based network communication can offer advantages in throughput, latency, and reliability that far surpass those attainable by traditional technologies, though much research and development is still needed for practical applications.
See also: Non-Conserved Flow Control for Erasure Coding-Based Network Communications (IEEE ToN)