Improving access to remote storage for weakly connected users

  • Authors:
  • John Kubiatowicz;Patrick Randolph Eaton

  • Affiliations:
  • University of California, Berkeley;University of California, Berkeley

  • Venue:
  • Improving access to remote storage for weakly connected users
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Storage researchers and service providers have been developing a new type of large-scale, distributed storage system. These systems aggregate storage from a large number of components, exploiting the content-addressable interface to address and manage data stored in the system. Potentially, such systems may be able to offer managed storage services to many new user groups, but they must yet overcome many challenges. In this thesis, we consider several important challenges that these new content-addressable storage systems confront when serving weakly connected clients. First, we justify why weakly connected users are an attractive target user population for these new storage systems. We then analyze the usage patterns of these users to understand the most important issues that they face. This background leads us to focus on techniques to improve the effective bandwidth and latency of the "last-mile" network link while protecting the privacy and integrity of data from untrusted agents in the infrastructure. We present three mechanisms to address these challenges. We describe privacy-preserving delta-encoding that can improve the effective bandwidth of the network link without sacrificing data privacy. We introduce intention updates, which allow clients to implement flexible link scheduling policies to reduce the effective latency of the link without sacrificing data consistency or visibility. Finally, we propose a new interface that raises the granularity of content-addressable storage from individual blocks to collections of blocks, reducing the cost managing data for both clients and storage systems. Finally, we describe and evaluate Moxie, a prototype that demonstrates how to integrate these mechanisms into a single storage system. Moxie provides a user-level file system client that allows end users to interact with a content-address storage system through the traditional file system interface. The file system client executes requests, transparently applying delta-encoding and intention updates when appropriate. Moxie improves client-perceived performance for weakly connected users by reducing the amount of data that must be transferred across the weak link and moving the transfers that must occur off of the critical path for most operations.