MultiZone PoolSize about evcache

[smadappa]

1K RPS Gets or Sets.
If it is set then you can improve the through put using multiple connections and in a multi memcached cluster.
If it is get then you may want to give getBulk a try and see the performance improves. Also looking at thread dumps and seeing where the threads are waiting.
how big is your memcached cluster?

[pat-humphreys]

It's mostly gets, we can't do bulk gets as it's caching request based data. The cache is 3 primary & 3 secondary, so I am wondering if there would be any need to increase it over the default of 1?

[smadappa]

You can increase the poolSize. Assuming good data distribution across the 3 primary instances you can go to pool size of 3 and see what you get. If you are doing 1K RPS how many are concurrent? Do you have too many threads and the JVM is spending time context switching between them?

[pat-humphreys]

There is even a bottleneck on the cache, I am replacing a xmemcache implementation with evcache and its set to a pool of 5. I was thinking of just going with a pool of 1, but I was wondering what the recommendation was