A Combined LIFO-Priority Scheme for Overload Control of E-commerce Web Servers

TL;DR

This paper introduces a LIFO-Pri overload control scheme for e-commerce web servers that prioritizes revenue-generating requests, significantly improving throughput and request completion rates during overload conditions.

Contribution

It proposes a novel LIFO-Pri scheduling discipline combining workload separation and dynamic prioritization to enhance server performance under overload.

Findings

Server operated at 60% capacity under 1.5x load

Seven-fold increase in completed revenue requests over FIFO

Significant throughput improvement during overload

Abstract

E-commerce Web-servers often face overload conditions during which revenue-generating requests may be dropped or abandoned due to an increase in the browsing requests. In this paper we present a simple, yet effective, mechanism for overload control of E-commerce Web-servers. We develop an E-commerce workload model that separates the browsing requests from revenue-generating transaction requests. During overload, we apply LIFO discipline in the browsing queues and use a dynamic priority model to service them. The transaction queues are given absolute priority over the browsing queues. This is called the LIFO-Pri scheduling discipline. Experimental results show that LIFO-Pri dramatically improves the overall Web-server throughput while also increasing the completion rate of revenue-generating requests. The Web-server was able to operate at nearly 60% of its maximum capacity even when offered load was 1.5 times its capacity. Further, when compared to a single queue FIFO system, there was a seven-fold increase in the number of completed revenue-generating requests during overload.