REVIEW: High Availability, Scalable Storage, Dynamic Peer Networks: Pick Two

REVIEW: High Availability, Scalable Storage, Dynamic Peer Networks: Pick Two

The paper discusses reliable storage systems in peer-to-peer networks and
the challenges they face. It conducts a study of the behaviour of users in
the Gnutella network, estimates requirements of a reliable storage system
with guarantees and concludes that large scale reliable file storage in a
peer to peer network is limited by the bandwidth accross hosts and not the
local storage available at hosts. The bandwidth required is unreasonable.

The paper argues that just the cost of maintaining a highly available
system with guarantees by means of replication and coding alone consumes
unreasonable amounts of bandwidth. For example, a million cable modem
users would be required to provide a continuous month of service to
maintain 1000TB of data even if no one actually used this data.
Conversely, it argues that disk space is not a concern as, for example, at
a monthly turnover rate, each cable modem must contribute less than 1 GB
of unique data, or 20 GB of total storage, the current idle capacity of
hard disks.

The auhors take into cosideration the effects of different participation
levels by nodes as well as their availability, which may result in less
demanding maintenace costs. For example 5% of Gnutella nodes provide 40%
of service at 40% availability. Also, means to reduce this bandwidth
overhead are considered such as admission control and load shifting.

However, the methodology used to produce the figures presented is not
discussed. It would be appropriate to show how Figure 1, for example was
derived.
Received on Mon Nov 14 2005 - 10:30:14 EST