Hi srmore, Perhaps if you use jconsole and connect to the jvm using jmx. Then uner MBeans tab, start inspecting the GC metrics.
/Jason On Fri, Dec 6, 2013 at 11:40 PM, srmore <comom...@gmail.com> wrote: > > > > On Fri, Dec 6, 2013 at 9:32 AM, Vicky Kak <vicky....@gmail.com> wrote: > >> Hard to say much without knowing about the cassandra configurations. >> > > The cassandra configuration is > -Xms8G > -Xmx8G > -Xmn800m > -XX:+UseParNewGC > -XX:+UseConcMarkSweepGC > -XX:+CMSParallelRemarkEnabled > -XX:SurvivorRatio=4 > -XX:MaxTenuringThreshold=2 > -XX:CMSInitiatingOccupancyFraction=75 > -XX:+UseCMSInitiatingOccupancyOnly > > > >> Yes compactions/GC's could skipe the CPU, I had similar behavior with my >> setup. >> > > Were you able to get around it ? > > >> >> -VK >> >> >> On Fri, Dec 6, 2013 at 7:40 PM, srmore <comom...@gmail.com> wrote: >> >>> We have a 3 node cluster running cassandra 1.2.12, they are pretty big >>> machines 64G ram with 16 cores, cassandra heap is 8G. >>> >>> The interesting observation is that, when I send traffic to one node its >>> performance is 2x more than when I send traffic to all the nodes. We ran >>> 1.0.11 on the same box and we observed a slight dip but not half as seen >>> with 1.2.12. In both the cases we were writing with LOCAL_QUORUM. Changing >>> CL to ONE make a slight improvement but not much. >>> >>> The read_Repair_chance is 0.1. We see some compactions running. >>> >>> following is my iostat -x output, sda is the ssd (for commit log) and >>> sdb is the spinner. >>> >>> avg-cpu: %user %nice %system %iowait %steal %idle >>> 66.46 0.00 8.95 0.01 0.00 24.58 >>> >>> Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s avgrq-sz >>> avgqu-sz await svctm %util >>> sda 0.00 27.60 0.00 4.40 0.00 256.00 >>> 58.18 0.01 2.55 1.32 0.58 >>> sda1 0.00 0.00 0.00 0.00 0.00 0.00 >>> 0.00 0.00 0.00 0.00 0.00 >>> sda2 0.00 27.60 0.00 4.40 0.00 256.00 >>> 58.18 0.01 2.55 1.32 0.58 >>> sdb 0.00 0.00 0.00 0.00 0.00 0.00 >>> 0.00 0.00 0.00 0.00 0.00 >>> sdb1 0.00 0.00 0.00 0.00 0.00 0.00 >>> 0.00 0.00 0.00 0.00 0.00 >>> dm-0 0.00 0.00 0.00 0.00 0.00 0.00 >>> 0.00 0.00 0.00 0.00 0.00 >>> dm-1 0.00 0.00 0.00 0.60 0.00 4.80 >>> 8.00 0.00 5.33 2.67 0.16 >>> dm-2 0.00 0.00 0.00 0.00 0.00 0.00 >>> 0.00 0.00 0.00 0.00 0.00 >>> dm-3 0.00 0.00 0.00 24.80 0.00 198.40 >>> 8.00 0.24 9.80 0.13 0.32 >>> dm-4 0.00 0.00 0.00 6.60 0.00 52.80 >>> 8.00 0.01 1.36 0.55 0.36 >>> dm-5 0.00 0.00 0.00 0.00 0.00 0.00 >>> 0.00 0.00 0.00 0.00 0.00 >>> dm-6 0.00 0.00 0.00 24.80 0.00 198.40 >>> 8.00 0.29 11.60 0.13 0.32 >>> >>> >>> >>> I can see I am cpu bound here but couldn't figure out exactly what is >>> causing it, is this caused by GC or Compaction ? I am thinking it is >>> compaction, I see a lot of context switches and interrupts in my vmstat >>> output. >>> >>> I don't see GC activity in the logs but see some compaction activity. >>> Has anyone seen this ? or know what can be done to free up the CPU. >>> >>> Thanks, >>> Sandeep >>> >>> >>> >> >