I'd be doing it and would provide update. BTW, since the query/sec rate is too low, I doubt if it would show anything. but would try and update.
Both clients aggregate at a switch and come into my server as a single 10G. So it shouldn't matter from poll() system call perspective. I'm also thinking of using ftrace to debug it...any other ideas? Thanks a lot! On Thu, Feb 5, 2015 at 2:22 PM, Stewart Smith <stew...@linux.vnet.ibm.com> wrote: > Learner Study <learner.st...@gmail.com> writes: >> Setup: >> MySQL 5.6.16 + thread pool extensions running on 3.18 kernel >> >> Two sysbench clients running from different PCs with total of 14k >> connections, each connection doing 2 select queries per sec. >> >> When I specify 7k connections from each client, sysbench shows latency >> of .3 msec (for both sides) >> >> If I run with 13k from one client and 1k from the other, latency comes >> to .6 msec (for both sides) >> >> I get that we are load balancing with 7k so latency improves but any >> ideas which part of MySQL server I should profile to see which area is >> messing it up with 13k+1k connection split- could it be poll() call? >> Or some other kernel area? > > does perf show any interesting difference between the two setups? > -- MySQL General Mailing List For list archives: http://lists.mysql.com/mysql To unsubscribe: http://lists.mysql.com/mysql