A CAS Storage System for ISR

Supiti Buranawatanachoke (sburanaw)
Kanat Tangwongsan (ktangwon)

Abstract
--------

Internet Suspend/Resume (ISR) is a client/server solution to mobility,
which allows a user to acquire his unique customization and state at
any participating machine. Various studies [1, 2] have shown that many
disk pages of the same computer are identical, suggesting that page
redundancies among a collection of machines are also high. The current
version of ISR stores disk state on both the content server and the
local cache in a separate 2-level directory tree for each parcel,
where the leaves of the tree are encrypted 128 KB chunks. We introduce
Content-Addressable Storage (CAS) as an alternative, where storage
location of each data block is identified by its content. With CAS, we
can eliminate the possibility of storing identical blocks twice, which
immediately lead to a huge space saving. Users should observe
performance improvement, as with CAS, we only download unique
blocks. This project involves working in 2 phases:


In phase 1, we aim at implementing CAS at the servers, whereby
maintaining compatibility with the existing clients. Each 128KB block
in the system is uniquely addressed by a unique ID. We plan to keep
track of blocks' information -- unique ID, SHA-1, usage statistics --
in a simple database. A working version of phase 1 is a foundation for
our phase 2. In phase 2, the goal is to enhance user's experience with
the ISR. Modification to the client will make it fetch only necessary
blocks. We believe that this is a step to speeding up the ISR. This
phase also involves optimization and testing to ensure correctness.
The evaluation of this project will answer how much space a CAS system
can save in a real setting.