CSC2231 - Understanding Availability review

Understanding Availability
------------------------------------------------------------
R. Bhagwan et al.

This paper studies the availability concept in P2P systems. By
looking at the Overnet P2P system, the authors analyze IP
aliasing effects, host availability, time of day effects, host
availability interdependence and turnover (arrivals and departures).

The paper strength is this different approach of considering
IDs instead of IPs when looking at availability. They actually
emphasize the IP aliasing effect due to private networks behind
NAT or due to DHCP.

As paper weaknesses, the authors do not consider in depth the
case of multiple users sharing the same machine. The only mention
this as an effect of IP aliasing but I think this case makes
sense when computing host availability interdependence (4.5).
In the case of multiple users on a single machine or multiple
OSes on a single machine there is host availability dependence.
This because those cases imply different IDs on one machine but
in many cases only one can be available. Let's say that I have
a WindowsXP and a Linux distro on my machine. When booting in XP
I will have ID1; when booting in Linux I will have ID2.
But doesn't this mean that if ID1 is available ID2 is not and
viceversa? I don't know to what extent this is true in the system
analyzed but the authors should have considered this.

One other thing is the terms arrive and depart; actually the way
the authors determine a departing host. I don't think it's meaningful
the way they considered this short term turnover.

I think the steady decrease in Figure 4 is interesting. I mean
where does it converge? If it's steady and it's about 32 hosts
per day will it reach 0 eventually? I don't think so, but a
longer experiment would be helpful.
Received on Mon Nov 14 2005 - 09:58:31 EST