Penalty Cost Minimization for Multi-tenant Query Deadline Employing Cache Optimization and Log Based Dispatching

摘要

Execution time is an important indicator of Service Level Agreement. Software as a Service providers expect to finish all queries before their respective deadlines and avoid penalties in violation of agreements. Hence, caching frequently occupied multi-tenant data is an efficient way of reducing execution time and penalty cost. However, it is difficult to minimize penalty cost through choosing cached data from massive multi-tenant data since it hardly finds a quantitative relationship between cached data and penalty cost. To address the challenge, we develop a mechanism of penalty cost minimization for multi-tenant queries based on cache optimization. We generate the cached data by analyzing cost space of multi-tenant data and the violation of queries of each node, and dispatch the incoming queries on the node with the shortest execution time according to multi-tenant cached data distribution. Experimental results suggest that penalty cost by employing the developed mechanism reduces 30% than that of the benchmark solution. Further, the developed mechanism is capable of limiting the dispatching time of queries within 1ms.