Quote from http://www.datastax.com/dev/blog/performance-improvements-in-cassandra-1-2
*"Murmur3Partitioner is NOT compatible with RandomPartitioner, so if you’re upgrading and using the new cassandra.yaml file, be sure to change the partitioner back to RandomPartitioner"* On Thu, Dec 12, 2013 at 10:57 PM, srmore <comom...@gmail.com> wrote: > > > > On Thu, Dec 12, 2013 at 11:15 AM, J. Ryan Earl <o...@jryanearl.us> wrote: > >> Why did you switch to RandomPartitioner away from Murmur3Partitioner? >> Have you tried with Murmur3? >> >> >> 1. # partitioner: org.apache.cassandra.dht.Murmur3Partitioner >> 2. partitioner: org.apache.cassandra.dht.RandomPartitioner >> >> >> > Since I am comparing between the two versions I am keeping all the > settings same. I see > Murmur3Partitioner has some performance improvement but then switching > back to > RandomPartitioner should not cause performance to tank, right ? or am I > missing something ? > > Also, is there an easier way to update the data from RandomPartitioner to > Murmur3 ? (upgradesstable ?) > > > > >> >> On Fri, Dec 6, 2013 at 10:36 AM, srmore <comom...@gmail.com> wrote: >> >>> >>> >>> >>> On Fri, Dec 6, 2013 at 9:59 AM, Vicky Kak <vicky....@gmail.com> wrote: >>> >>>> You have passed the JVM configurations and not the cassandra >>>> configurations which is in cassandra.yaml. >>>> >>> >>> Apologies, was tuning JVM and that's what was in my mind. >>> Here are the cassandra settings http://pastebin.com/uN42GgYT >>> >>> >>> >>>> The spikes are not that significant in our case and we are running the >>>> cluster with 1.7 gb heap. >>>> >>>> Are these spikes causing any issue at your end? >>>> >>> >>> There are no big spikes, the overall performance seems to be about 40% >>> low. >>> >>> >>>> >>>> >>>> >>>> >>>> On Fri, Dec 6, 2013 at 9:10 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 >>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>> >>>> >>> >> >