General store placement for response time minimization in parallel disks

摘要: We investigate the placement of N enterprise data-stores (e.g., database tables, application data) across an array disks with aim minimizing response time averaged over all served requests, while balancing load evenly in parallel disk array. Incorporating non-FCFS serving discipline and non-work-conserving nature drives formulation problem is difficult current strategies do not take them into account. present a novel to incorporate these crucial features identify runlength requests accessing store as most important criterion for placing stores. use insights design fast (running NlogN) algorithm that optimal under assumption transfer times are small. Further, we develop polynomial-time extensions minimize even if large, loads disks. Comprehensive experimental studies establish efficacy proposed wide variety workloads reducing real storage traces by more than factor 2 heterogeneous workload scenarios.