I have a bi-directionally published database with 5 subscribers. Generally the amount of data transferring is pretty small (sometimes 1 record). There is a service that runs on each subscriber's workstation that executes a pull merge replication between itself and the publisher every 5 minutes. On 3 of the workstations, the durations are fairly consistent (2-10 seconds). On the other two, much of the time the durations are similar to the other 3, but every few times, it takes 30+ seconds to execute, often only sending a record from 1 table.
I have been capturing performance counters on two of the stations, one consistent and one not, for many different types including SQL Server: Buffer Manager, Replication Merge, General Statistics and others. In comparing the two, I don't notice much difference, certainly not anything striking, especially at the times that the 30+ second durations occur.
Has anyone seen anything like this? If so, what did you do to debug the issue? If not, any suggestions of debug methodologies, including recommended performance counters and/or other debugging statistics?
Thanks!