Improving Server Utilization in a Distributed Computing Set-up with Independent Clients

TL;DR

This paper analyzes a distributed system with multiple servers and clients, proposing strategies that improve server utilization by addressing the inefficiencies of existing client request behaviors.

Contribution

It introduces novel client strategies that enhance server utilization in a decentralized setting without coordination.

Findings

Proposed strategies outperform existing ones in server utilization

Certain client behaviors lead to significant idle server times

New strategies reduce request collisions and improve throughput

Abstract

We consider a set-up in which there are multiple servers and multiple clients in a large distributed computing system. Clients request servers to process jobs. Servers can only process one job in unit time. There is no coordinating agent to route client requests to servers, and clients choose servers independently and simultaneously, and only have access to the outcomes of their own past requests. If more than one clients choose the same server, then only one randomly chosen client's requests will be fulfilled. If some servers do not receive any request, they remain idle. In this paper, we show that a large category of strategies are not effective in terms of server utilization. We devise strategies for clients that improve server utilization of such systems over those of strategies known in the current literature.