J
Jeff Owen
Guest
The duration of a lock of the pica queue is less than a millisecond. Processes that flush AI buffers add entries to the pica queue and the replication server reads entries from the queue. I should note that the pica queue is not used during the startup synchronization process. During synchronization, the repl server reads directly from the AI extent(s). The pica queue is then “turned on” once is gets close to the most recently written AI blocks. The pica queue is then used during normal processing. I would assume your contention is very low because you are running an AIW on the source. The only contention in this case would only between the AIW and the repl server. Generally, you should see very little contention for the pica queue with an AI configuration that is well tuned. Otherwise, we’d like to hear about an environment where a high level of pica queue is seen. -jeff
Continue reading...
Continue reading...