Conclusion
The increasing dependence of users on distributed file systems
makes the availability of these systems a matter of concern.The
two mechanisms of Coda that address this issue, server replication and disconnected
operation,provide resiliency in the face of many failures while minimally
impacting performance .Although Coda is far from maturity,its design has been
warmly accepted.
One can sit down at a Coda client and execute Unix applications without
recompilation or relinking.Execution continues transparently when contact
is lost with a server due to a crash or network failure.In the absence of
failures, using coda client feels no different from using an AFS client.
Preliminary measurements with the Andrew benchmark shows that the degradation
due to replication is relatively small.With one replica ,Coda performs about
5%worse than with a non-replicated volume.With two and three replicas,
it performs 9% and 11% worse respectively.With three replicas, Coda
performs 28% worse than a local unix file system. It should be emphasized
that these measurements are from an untuned prototype, and significant improvements
are been made as Coda is refined.