I see about 33% improvement in GET and SET throughput and 50% reduction in p99 latency if we remove the candidate from the mmContainer first.
I'm opening this PR to find the correct approach for those items that we fail to remove from the access container. My stats show there are 9 instances where this occurs.
I see about 33% improvement in GET and SET throughput and 50% reduction in p99 latency if we remove the candidate from the mmContainer first.
I'm opening this PR to find the correct approach for those items that we fail to remove from the access container. My stats show there are 9 instances where this occurs.
base-stats.txt single-stats.txt
This change is