From: Steve Arnold (steve.arnold4_at_verizon.net)
Date: Tue Mar 02 2004 - 21:40:41 PST
CFS is a peer-to-peer read-only file system using the internet as its
network. It uses a layered approach where 1) FS is the file system
interface, 2) DHash stores the data, and 3) Chord locates the servers. CFS
uses a hash to distribute a file onto several servers (depending on the
block size). This is different from PAST in that PAST uses whole files.
There is no server in CFS; the control is distributed. The Chord layer is
used to lookup blocks. It uses a ring system where each node has r
successors. In order for the ring to fail, all of the successors would have
to fail. It does look for servers that are nearby in the network. Each node
must authenticated.
The system is reliable by providing load balancing and replication.
Replication is configurable and is performed by DHash. The load balancing is
inherent in the distributed system. Administrators can configure a node to
act as several virtual servers, however, depending on the resources of that
machine. (It's too bad it's not dynamic.)
It seemed to me that they still had some work to do to really make this
useful. Currently only publishers can write and update the system. You can
only delete files via the expiration. This is a potential drawback. Also the
authors were still working on an index system. Nodes often enter and leave
peer-to-peer systems; it seems to me that r should be sufficiently large to
allow for this.
This archive was generated by hypermail 2.1.6 : Tue Mar 02 2004 - 21:39:50 PST