Coding-Based Distributed Congestion-Aware Packet Spraying to Avoid Reordering in Data Center Networks

Jinbin Hu,Lei Wang,Osama Alfarraj,Amr Tolba,Chang Ruan

doi:10.1109/access.2021.3062866

Abstract

Modern Data Center Networks (DCNs) are commonly based on Clos topologies with a large number of equal-cost multiple paths to provide high bisection bandwidth. The existing Random Packet Spraying (RPS) scheme spreads each flow of packets to all available parallel paths in order to achieve good load balancing under symmetric topologies. However, under asymmetric topologies caused by traffic dynamics or link failures, RPS potentially suffers from serious out-of-order problem. Therefore, to avoid packet reordering, we propose a Coding-based Distributed Congestion-aware Packet Spraying mechanism called CDCPS. At the sender end, CDCPS encodes packets using forward error correction (FEC) technology and adaptively adjusts the coding redundancy according to the asymmetric degree of multiple equal-cost paths. To make full use of link bandwidth, CDCPS randomly spreads encoded packets to all available paths at the switches. The original packets can be recovered immediately once enough encoded packets from uncongested paths arrive at the receiver, even if some encoded packets are blocked on congested paths. The test results of NS2 simulation showed that CDCPS eliminates out-of-order packets completely and effectively reduces the average and $99^{th}$ flow completion time by up to 73% and 78% over the state-of-the-art load balancing scheme.

Highlights

To support the increasing traffic demands of large-scale distributed applications such as Web search, Data Mining and Machine Learning, modern Data Center Networks are typically based on Clos topologies, which provide high bisection bandwidth via a large number of equal-cost multiple paths between any pair of end-hosts [1], [2]
To adapt to asymmetry in the network topology, we propose a coding-based distributed congestion-aware packet spraying mechanism called CDCPS, which successfully integrates coding into packet spraying and completely avoids packet reordering
CDCPS encodes the packets of a flow at the sender, randomly spreads the encoded packets to all available multiple paths at switches, and recovers the original source packets at the receiver once receiving enough encoded packets

Summary

Introduction

To support the increasing traffic demands of large-scale distributed applications such as Web search, Data Mining and Machine Learning, modern Data Center Networks are typically based on Clos topologies, which provide high bisection bandwidth via a large number of equal-cost multiple paths between any pair of end-hosts [1], [2]. Even some other paths are under-utilized, leading to topology asymmetric [5]–[7]. As a packet-level load balancing mechanism, Random Packet Spraying [8] is proposed to make full use of the multiple parallel paths in data centers. RPS randomly splits and spreads each flow into packets to one of the available paths to the destination. RPS obtains high link utilization and achieves the best load balancing effect. RPS has already been deployed on the commodity switches due to its simplicity [9]

Objectives

Methods

Findings

Conclusion