MyProxy High Availability Approach

MyProxy service is rated as critical. Based on the service criteria, this therefore requires a high availability automatic switch in the event of a single software or hardware failure to ensure that the service is not affected.

Proposal

The MyProxy software provides a replication functionality (myproxy-replicate). This permits replication of the myproxy data at regular intervals to a slace server.

The configuration proposed is as follows

• Replication every n seconds between master and slave
• Virtual Cluster IP address managed by Linux-HA

Impact of failover

1. Up to n seconds of credentials data will be lost. Any user who has authenticated during this period will lose their myproxy.
2. Problem -> Detection -> Takeover may take m seconds

If the loss of credentials is a show-stopper, a shared disk or shared file system solution could be considered. Shared disks would increase the takeover time since there would be a log replay.

Next Steps

• Setup a test bed with two servers

-- TimBell - 07 Sep 2005