Cloud storage with minimal trust

Опубликовано 12 августа 2016, 2:26

Cloud storage has emerged as a popular deployment model for building highly available storage services. In a typical cloud deployment, the storage service provider (SSP) owns the machines and handles all the details related to performance, reliability, and geographic replication that are needed to build a highly reliable storage. This model seems promising for both the SSPs and the clients: SSPs can better manage the cloud resources at a lower cost due to the economies of scale; clients get professionally managed and scalable storage for a fraction of the cost. In this talk, I will describe Depot, a highly available cloud storage system, in which clients do not have to trust the correctness or availability of the storage servers. The key motivation for Depot comes from the fact that cloud storage service providers (SSPs), such as S3 and Azure, are fault-prone black boxes operated by a party other than the data owner. To address this limitation, Depot minimizes the trust clients have to put in such third party storage provider. As a result, Depot clients can continue functioning despite benign or Byzantine faults of the cloud servers. Yet, Depot provides stronger availability, durability, and consistency properties than those provided by many of the existing cloud deployments, without incurring prohibitive performance cost. For example, in contrast to Amazon S3ΓÇÖs eventual consistency, Depot provides a variation of causal consistency on each volume, while tolerating Byzantine faults.