Peabody: the time travelling disk

摘要

Disk drives are now available with capacities on the order of hundreds of gigabytes. What has not become available is an easy way to manage storage. With installed machines located across the enterprise, the backup, management of application installation, and maintenance of systems have become a nightmare. An increasing trend in the storage industry is to virtualize storage resources, maintaining a central repository that can be accessed across the network. We have designed a network block storage device, Peabody, that exposes virtual disks. These virtual disks provide mechanisms to: recover any previous state of their sectors and share backend storage to improve cache utilization and reduce the total amount of storage needed. Peabody is exposed as an iSCSI target, and is mountable by any iSCSI compatible initiator Using our implementation of Peabody, we show that for our workloads, up to 84% of disk sectors written, contain identical content to previously written sectors, motivating the need for content-based coalescing. The overhead for writing in a simple implementation is only 20 percent of the total write speed. This paper describes our early experiences with the Peabody implementation. We quantify how rapidly storage is consumed, examine optimizations, such as content-based coalescing and describe how recovery is currently implemented. We conclude with future plans based on these measurements.