Avoiding Bad Query Mixes to Minimize Unsuccessful Client Requests Under Heavy Loads

webserver;timeouts;Computer Science

Tozer, Sean

University of Waterloo

关键词: webserver;    timeouts;    Computer Science;   

Others  :  https://uwspace.uwaterloo.ca/bitstream/10012/4573/1/3tier.pdf

瑞士|英语

来源: UWSPACE Waterloo Institutional Repository

【 摘 要 】

In three-tiered web applications, some form of admission control is required to ensurethat throughput and response times are not significantly harmed during periods of heavyload. We propose Q-Cop, a prototype system for improving admission control decisionsthat computes measures of load on the system based on the actual mix of queries beingexecuted. This measure of load is used to estimate execution times for incoming queries,which allows Q-Cop to make control decisions with the goal of minimizing the numberof requests that are not serviced before the client, or their browser, times out.Using TPC-W queries, we show that the response times of different types of queriescan vary significantly, in excess of 50% in our experiments, depending not just on thenumber of queries being processed but on the mix of other queries that are running simultaneously.The variation implies that admission control can benefit from taking intoaccount not just the number of queries being processed, but also the mix of queries. Wedevelop a model of expected query execution times that accounts for the mix of queriesbeing executed and integrate that model into a three-tiered system to make admissioncontrol decisions. This approach makes more informed decisions about which queriesto reject, and our results show that it significantly reduces the number of unsuccessfulclient requests. Our results show that this approach makes more informed decisions aboutwhich queries to reject and as a result significantly reduces the number of unsuccessfulclient requests.For comparison, we develop several other models which represent related work inthe field, including an MPL-based approach and an approach that considers the type ofquery but not the mix of queries. We show that Q-Cop does not need to re-computeany modelling information in order to perform well, a strong advantage over most otherapproaches. Across the range of workloads examined, an average of 47% fewer requestsare denied than the next best approach.