Improving Adaptivity in Mobile File Systems

Benjamin Atkin and Kenneth P. Birman
Reliable Distributed Systems Group
Department of Computer Science, Cornell University

Applications which communicate extensively over a network can be highly sensitive to network performance. This is especially true in a wireless network, where mobile hosts must cope with large variations in available bandwidth. While extensive research has been done on adapting applications to work well in low-bandwidth environments, our research focuses on supporting mobile hosts operating in an "insufficient bandwidth" environment. This is motivated by the fact that an application that expects its network operations to complete relatively quickly could be adversely affected even if it receives 90% of its ideal bandwidth.

Our research in network-aware adaptation has been particularly focused on mobile file systems, since that's an area where there's a lot of scope for improving application performance by adapting to bandwidth variations. Now that it's possible to envisage people walking around with laptops or PDAs at a workplace, accessing and modifying shared documents as they do their jobs, mobile file systems have an opportunity to make a big impact in a new application environment. Allowing an architect sitting in an office to immediately see the changes to the construction project being made by an engineer walking around the building site isn't well-supported by existing collaborative project management software.

Adapting mobile file system behaviour to changes in bandwidth means having establishing rules for how a mobile client reacts to the current bandwidth availability, such as "don't write back changes unless bandwidth is above this threshold". Unfortunately, that kind of static behaviour isn't very reactive to what traffic the file system is generating (maybe it's not doing anything else right now) or could waste bandwidth (maybe there's enough bandwidth to write back some changes, just not all of them). Instead of using static rules, we've explored doing fine-grained adaptation to bandwidth variation by attaching priorities to different classes of client-server interactions. This allows us to make good use of the available bandwidth, whatever the particular value maybe, and whatever traffic is being generated by the file system.

We've implemented a mobile file system, MFS, to test out these ideas in mobile file system adaptation and cache consistency for collaborative work environments. Our experiments using MFS have demonstrated that it's a promising platform for supporting new kinds of data-sharing applications.


This abstract has an accompanying poster. MFS is part of the Modeless Adaptation project.

This work was supported in part by DARPA under AFRL grant RADC F30602-99-1-0532, and by AFOSR under MURI grant F49620-02-1-0233, with additional support from Microsoft Research and from the Intel Corporation.