This paper studies the availability of peer-to-peer storage systems
using Overnet as a testbed. With Overnet's ability to identify clients
with unique IDs, the experiment crawls for random IDs to discover hosts
and probes them periodically. A quick conclusion is that frequent and
periodic file refreshes is necessary due to the fact that clients are
independent and their availability varies from time to time.
This paper gives a good idea to readers on how the availability of P2P
clients affects the overall availability of the system. Nevertheless, I
think it lacks the tweaks on data that also affect the availability,
such as replication and distribution of files.
I think the host availability interdependence results may be interpreted
as the availability of a file given one of the servers and the client.
By looking at this results alone, it seems like a client has a slight
chance to get hold of the file it wants. However, because P2P storage
systems usually break files into parts and store them on separate
clients, I would suggest to have PDF plots with more servers, such as
P(Y=1 | X1=1, X2=1) and P(Y=1 | X1=1, X2=1, X3=1), to determine how many
parts a file should be divided in order to achieve reasonable availability.
Received on Mon Nov 14 2005 - 09:51:57 EST
This archive was generated by hypermail 2.2.0 : Mon Nov 14 2005 - 09:58:32 EST