Asynchronous verifiable information dispersal

Abstract

We consider the distribution of data by a client among a set of n storage servers, of which up to t might be faulty exhibiting arbitrary, i.e., Byzantine, behavior. The goal is to ensure that clients can always recover the stored data correctly, independently from the behavior of faulty servers or other, faulty clients. An inefficient solution is based on replication such that every server keeps a copy of the data. The classic alternative is information dispersal (IDA): using an erasure code, the data is split into blocks such that each server holds exactly one block and only a subset of the blocks is needed in order to reconstruct the data. A protocol for information dispersal based on IDA tolerating Byzantine servers has been proposed by Garay et al. [4]. It relies on synchronous networks, which are adequate for tightly coupled clusters but unrealistic for geographically distributed or heterogeneous systems. An asynchronous system model is more appropriate for such settings. In this paper, we introduce the notion of verifiability for information dispersal in asynchronous networks with a computationally bounded adversary. Intuitively, verifiability means that whenever the honest servers accept to store some data, then the data is also consistent and no two distinct honest clients can reconstruct different data. This notion of verifiability originates in the related context of secret sharing [3,1]. We propose a new scheme for asynchronous verifiable information dispersal that is also storage- and communication-efficient and achieves optimal resilience t