Read-Write Replication in Arla

Background

In the distributed AFS file system a volume can be replicated to an other file server. The replicas are read-only to avoid complications. Only one file server has read-write permissions in the volume. Replication gives load balancing for frequently accessed volumes. It also means that users can continue accessing the file even if the read-write file server goes down. Updates to the contents of the volume must be done via the read-write replica and require that all replicas are alive and reachable.

The Arla project at KTH is a project for implementing a free, fully AFS-compatible file system. Also some improvements to the system are planned. One of these is read-write replication of volumes.

Read-write replication means that updates can be done even if some replicas are unreachable for some reason. This would mean that read-write volumes can be replicated and gain the same advantages as replicated read-only vomues. It would also introduce new problems.

The Task

The task is to study and design ways of achieving read-write replication of volumes in the Arla file system. The goal is to find a model which is effective and implementable. The work will mainly consist of studying previous work in the area of replicating file systems and evaluate the posibilities to implement similar methods in Arla.

Several solutions satisfying different requirements or suitable for different scenarios may be presented. The solution(s) should be compatible with the existing AFS clients. Also the wishes of current AFS administrators should be considered.

The thesis will not include implementing the replication in the Arla file system.

Examiner and supervisors

Examiner
Karl-Filip Faxén
Dept. of Teleinformatics, KTH
kff@it.kth.se

Supervisor, IT
Fredrik Lundevall
Dept. of Teleinformatics, KTH
f@it.kth.se

Supervisor, NADA
Ragnar Andersson
Dept. of Numerical Analysis and Computing Science, KTH
ran@nada.kth.se

Back to main page