Comment 17 for bug 1405549

Plan for fixing:

1. Try with small socket_timeout. socket_timeout=1.
2. If results are still not good enough, reduce the number of processes per controller.
3. If results are still not good enough, implement decimal socket_timeout and try with socket_timeout=0.5 (or less
4. If results are still not good enough, implement info sharing across processors

If one of the steps is successful, the resulting code should be tested on scale.

QA folks, could you please try reproducing the bug with socket_timeout=1? You need to set [memcache]socket_timeout = 1 and [cache]memcache_socket_timeout = 1.