Straggler Exploitation in Distributed Computing Systems with Task Grouping
Tharindu B. Adikari, Haider Al-Lawati, Jason Lam, Zhenhua Hu, Stark C. Draper · 2023
We consider the problem of stragglers in distributed computing systems. Stragglers are compute nodes that unpredictably become slow. This often increases the completion times of tasks. One popular method for mitigating stragglers is to replicate work. Out of all replications of a task, only the first to complete is accepted. The rest are discarded. Discarding work completed by some of the workers leads to wastage of resources. In this paper, we propose a method for exploiting the work completed by stragglers rather than discarding it. We increase the granularity of the assigned work and the frequency of worker updates. We demonstrate that the proposed method reduces the completion time of tasks via experiments performed on a simulated cluster.