DEMB: Cache-aware scheduling for distributed query processing

Abstract

Leveraging data in distributed caches for large scale query processing applications is becoming more important, given current trends toward building large scalable distributed systems by connecting multiple heterogeneous less powerful machines rather than purchasing expensive homogeneous and very powerful machines. As more servers are added to such clusters, more memory is available for caching data objects across the distributed machines. However the cached objects are dispersed and traditional query scheduling policies that take into account only load balancing do not effectively utilize the increased cache space. We propose a new multi-dimensional range query scheduling policy for distributed query processing frameworks, called DEMB, that employs a probability distribution estimation derived from recent queries. DEMB accounts for both load balancing and the availability of distributed cached objects to both improve the cache hit rate for queries and thereby decrease query turnaround time and throughput. We experimentally demonstrate that DEMB produces better query plans and lower query response times than other query scheduling policies.