REVIEW: "Total Recall: System Support for Automated Availability Management"

REVIEW: "Total Recall: System Support for Automated Availability
Management"

The paper presents TotalRecall, a peer-to-peer storage system. Total
recall is a DHT based storage system in which content is indexed by keys.
More specifically, TotalRecall uses Chord as its underlying
infrastructure.

 A guiding design goal for the system has been to automate the task of
managing availability. The motivation for one such automated system arises
from the realization that although many systems incorporate mechanisms to
provide high availability, they are often shipped with these mechanisms
not fine tuned, and the task is left to the system manager. Many, however,
have no knowledge or time to understand and tune these settings.
Peer-to-peer systems impose additional difficulty as as constituent hosts
join and leave the network at will. TotalRecall thus provide the
following: availability prediction using stochastic models, redundancy
management through replication or erasure coding, and dynamic repair. To
accomplish these goals TotalRecall employs a storage manager (TRSM), an
availability monitor (AM) and redundancy engine (RE).

Among its challenges, TotalRecall imposes a significant network overhead
as it transmits data for repair. It consumes an average of 6.5KB/s per
file stored. Over a period of one day this easily amounts to 500MB, a
limiting factor even to broadband users, who have limited bandwidth usage.
Furthermore, this network overhead also limits the scalability of the
system as only a limited number of users can safely run the system within
one local area network.
Received on Thu Nov 24 2005 - 10:16:24 EST