AdaptiveSLA: A Two-Stage Scheduling Framework for SLA Profit Maximization in Multi-tenant Database

Abstract

The requirement of multi-tenant applications is continuously increasing in cloud computing epoch, and they usually have larger data volume compared with traditional applications. Considering the quality of managing these data, SLAs (Service Level Agreement) are usually defined between multi-tenant database service providers and tenants. The providers will get revenues if SLAs are met, otherwise they will pay penalties. In order to maximize the profit, this paper presents AdaptiveSLA, a two-stage scheduling policy for multi-tenant database. In the first stage, AdaptiveSLA detects performance crises and leverages sliding window algorithm to mitigate crises in distributed multitenant database. The crises may derive from heterogeneity of requests or performance degradation of nodes. In the second stage, AdaptiveSLA makes requests’ execution sequence. This stage aims at completing requests as many as possible. The executing process is constrained by SLA, which is modeled as slack time. Extensive experimental results demonstrate the effectiveness and efficiency of AdaptiveSLA scheduling policy.