Review of Understanding Availability

This paper argues that the availability in p2p systems is quite different
with traditional host availability. The authors found some interesting
results by measuring Overnet. Peers join and leave the system
very frequently since the measured data shows each day over 20% new users
join or leave the system. This indicates p2p system has a very large but
unstable membership. It also demonstrates the host availability changes
with the long term duration and varies during a day. These results pose a
big challenge for p2p storage networks since the overhead of maintaining
replicas is high under p2p churn.

However, there are some factors missing in this paper. It is important to
know the life time distribution of hosts in the system, and the active and
inactive period distribution of hosts. We can consider inactive status as
a failure, but it is mainly due to that the host leaves the system, quite
different with traditional machine failures. These factors can influence
the replication strategy. In addition, this paper shows the hosts failings
are independent. But if hosts are classified by geographical regions,
most of them may be up during the day, and inactive during the mid night.

The authors claim that host availability is not well modeled as a single
stationary distribution, but instead is a combination of a number of
time-varying functions. But it does not further illustrate how to define
availability explicitly, and how to compare the availability among
different p2p systems. It is not clear whether host availability, its
time variation, and diurnal patterns are enough to evaluate the
availability of a p2p system or not, and whether we should take the number
of total hosts and other factors into account.
Received on Mon Nov 14 2005 - 00:33:43 EST