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 >> >> >> >
