Straggler Mitigation with Tiered Gradient Codes

TL;DR

This paper introduces tiered gradient coding techniques for distributed gradient descent, allowing staggered task initiation to reduce task sizes, improve efficiency, and mitigate stragglers in server systems.

Contribution

It proposes a novel tiered system model that starts with fewer tasks and adds more later, optimizing task sizes and reducing job completion time compared to traditional methods.

Findings

Lower task sizes achieved with tiered approach

Reduced job completion time due to staggered task start

Improved server utilization efficiency

Abstract

Coding theoretic techniques have been proposed for synchronous Gradient Descent (GD) on multiple servers to mitigate stragglers. These techniques provide the flexibility that the job is complete when any $k$ out of $n$ servers finish their assigned tasks. The task size on each server is found based on the values of $k$ and $n$. However, it is assumed that all the $n$ jobs are started when the job is requested. In contrast, we assume a tiered system, where we start with $n_1\ge k$ tasks, and on completion of $c$ tasks, we start $n_2-n_1$ more tasks. The aim is that as long as $k$ servers can execute their tasks, the job gets completed. This paper exploits the flexibility that not all servers are started at the request time to obtain the achievable task sizes on each server. The task sizes are in general lower than starting all $n_2$ tasks at the request times thus helping achieve lower task sizes which helps to reduce both the job completion time and the total server utilization.