incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From srmore <comom...@gmail.com>
Subject Re: Write performance with 1.2.12
Date Fri, 06 Dec 2013 15:40:23 GMT
On Fri, Dec 6, 2013 at 9:32 AM, Vicky Kak <vicky.kak@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 <comomore@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
>>
>>
>>
>

Mime
View raw message