From: Chuck Reeves (creeves_at_windows.microsoft.com)
Date: Wed Mar 03 2004 - 17:12:46 PST
The paper, "Wide Area Cooperative Storage with CFS" was written by a
number of researchers at MIT in
2001. The paper describes the design and measurements of a distributed
file system built on top of the Chord peer-to-peer protocol. The design
of CFS takes a very pedictable next step in the development of
peer-to-peer file systems. The storage of block (or page) level
information instead of whole file (or object) level data. The Dhash
layer is introduced to fill the gaps between the Chord protocol and the
requirements of the file system. This includes support for replicating
content to multiple nodes, caching of popular content as well as a cool
technique for properly weighting disimiliar sized stores called virtual
servers.
Like Pastry the Chord algorithm identifies node storage by hashing
content. In the case of CFS a unique hash is created for each block of
data stored into the system. The hash identifies a position on a virtual
circle to which get and save requests are forwarded using a finger
table. The finger table stores node addresses for successors at each
power of 2 away from that node. The implementors also added an
optimization similiar to the proximity metric in Pastry called server
selection that utilized network timing to prioritize the selection of
forwarding requests.
Since it stores files in a distributed way, this system has some
advantages in how it stores large files. I thought the idea that nodes
could adjust their virtual server profile was a uniquely adaptive
characteristic.
Chuck Reeves, creeves_at_microsoft.com
Microsoft | Windows | Directory Services
This archive was generated by hypermail 2.1.6 : Wed Mar 03 2004 - 17:12:31 PST