A decentralized algorithm for erasure-coded virtual disks

Abstract

A Federated Array of Bricks is a scalable distributed storage system composed from inexpensive storage bricks. It achieves high reliability with low cost by using erasure coding across the bricks to maintain data reliability in the face of brick failures. Erasure coding generates n encoded blocks from m data blocks (n > m) and permits the data blocks to be reconstructed from any m of these encoded blocks. We present a new fully decentralized erasurecoding algorithm for an asynchronous distributed system. Our algorithm provides fully linearizable read-write access to erasure-coded data and supports concurrent I/O controllers that may crash and recover. Our algorithm relies on a novel quorum construction where any two quorums intersect in m processes.

Optimistic Erasure-Coded Distributed Storage

“…In addition, note that the size of the storage at all servers is always bounded by ∆. This is in contrast to the implementation in [6] and ORCAS-B implementation that we describe later, where the worst-case storage size is dependent on the maximum number of concurrent writes.…”

“…Recently there has been lot of work on erasure-coded distributed storage [6,7,2,5,10,9]. We discuss below three representative papers that are close to our work.…”

“…Frolund et al [6] describe an erasure-coded distributed storage (called FAB) in the same system model as this paper, i.e., an asynchronous crash-recovery model. The primary algorithm in FAB implements an atomic register.…”

“…Recently however, it has been argued that erasure coding is a better alternative to data replication since it reduces the cost of ensuring fault tolerance [6,7]. In erasure-coded storage systems, instead of keeping an identical version of a data V on each server, V is encoded into n fragments such that V can be reconstructed from any set of at least k fragments (called k-of-n encoding), where the size of each fragment is roughly |V |/k.…”

“…In Appendix B, we show lower bounds on storage space usage in atomic register implementation in synchronous and asynchronous systems, for a specific class of implementations (that include both ORCAS-A and ORCAS-B, and the implementation in [6]). Roughly speaking, we show that implementations − (1) which at the end of a write store equal number of encoded fragments in the stable storage of the servers, and (2) do not use different encoding schemes in the same operation − cannot have a stable storage usage better than ORCAS-A (and ORCAS-B) in either synchronous or asynchronous periods.…”