Obtaining High Performance For Storage Outsourcing

Storage outsourcing is an emerging industry that shields storage users from the complexity of in-house storage management, while providing cost savings and reliability improvements via the aggregation of storage into large, special-purpose facilities. These distributed and replicated facilities are operated by a storage service provider, and are accessed by remote users via high-speed network connections.The viability of storage outsourcing is critically dependent on the performance of remote storage. In this paper, we measure the performance of I/O benchmarks accessing a remote block-level storage system. We use benchmarks that represent a variety of workloads, running on several operating systems and file systems. Network latencies represent distances ranging from a local neighborhood to halfway across a continent. We vary the network loss characteristics to correspond with the conditions of either dedicated fiber or shared Internet (with loss rates up to 10(-3)). We examine the effectiveness of basic latency-hiding techniques such as caching, application prefetching, and asynchronous writes. We conclude that remote storage is already viable for a wide variety of active workloads, and we point out areas where new techniques could provide significant additional performance enhancement.