understanding availability

This paper presents empirical data showing the churn characteristics of a
DHT-based P2P system. Unlike other studies which assume a steady-state
churn model, this paper provides evidence for the diurnal pattern of host
count, dynamic IPs exaggerating the host count, and general trend of
decreased availability over longer measured time intervals.

The graph which I found interesting in this paper is Figure 4, which shows
that the number of available hosts tends to increase during the day and
drop at night. I suspect this is a symptom of the type of P2P system
(Overnet is a P2P filesystem). My experiences with P2P filesharing
systems is that the number of hosts dramatically increases at night when
other peers leave their systems on to leech files (and thus also serve)
files. I think this suggests that trying to characterize P2P systems,
especially using empirical data, requires a different taxonomy -- perhaps
one that also takes into account the p2p overlay and its applications'
objectives.
Received on Mon Nov 14 2005 - 10:42:47 EST