Modern distributed storage systems offer large capacity to satisfy theexponentially increasing need of storage space. They often use erasure codes toprotect against disk and node failures to increase reliability, while trying tomeet the latency requirements of the applications and clients. This paperprovides an insightful upper bound on the average service delay of sucherasure-coded storage with arbitrary service time distribution and consistingof multiple heterogeneous files. Not only does the result supersede known delaybounds that only work for a single file or homogeneous files, it also enables anovel problem of joint latency and storage cost minimization over threedimensions: selecting the erasure code, placement of encoded chunks, andoptimizing scheduling policy. The problem is efficiently solved via thecomputation of a sequence of convex approximations with provable convergence.We further prototype our solution in an open-source, cloud storage deploymentover three geographically distributed data centers. Experimental resultsvalidate our theoretical delay analysis and show significant latency reduction,providing valuable insights into the proposed latency-cost tradeoff inerasure-coded storage.
展开▼
