Sudipto Das, Vivek Narasayya, Feng Li, and Manoj Syamala
Multi-tenancy and resource sharing are essential to make a Database-as-a-Service (DaaS) cost-effective. However, one major consequence of resource sharing is that the performance of one tenant's workload can be significantly affected by the resource demands of co-located tenants. The lack of performance isolation in a shared environment can make DaaS less attractive to performance-sensitive tenants. Our approach to performance isolation in a DaaS is to isolate the key resources needed by the tenants' workload. In this paper, we focus on the problem of effectively sharing and isolating CPU among co-located tenants in a multi-tenant DaaS. We show that traditional CPU sharing abstractions and algorithms are inadequate to support several key new requirements that arise in DaaS: (a) absolute and fine-grained CPU reservations without static allocation; (b) support elasticity by dynamically adapting to bursty resource demands; and (c) enable the DaaS provider to suitably tradeoff revenue with fairness. We implemented these new scheduling algorithms in a commercial DaaS prototype and extensive experiments demonstrate the effectiveness of our techniques.
In Proceedings of the VLDB Endowment
Publisher Very Large Data Bases Endowment Inc.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Articles from this volume were invited to present their results at The 40th International Conference on Very Large Data Bases, September 1st 5th 2014, Hangzhou, China. Proceedings of the VLDB Endowment, Vol. 7, No. 1 Copyright 2013 VLDB Endowment 21508097/13/09... $ 10.00.