incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Vicky Kak <vicky....@gmail.com>
Subject Re: Write performance with 1.2.12
Date Fri, 06 Dec 2013 17:15:12 GMT
Since how long the server had been up, hours,days,months....?


On Fri, Dec 6, 2013 at 10:41 PM, srmore <comomore@gmail.com> wrote:

> Looks like I am spending some time in GC.
>
> java.lang:type=GarbageCollector,name=ConcurrentMarkSweep
>
> CollectionTime = 51707;
> CollectionCount = 103;
>
> java.lang:type=GarbageCollector,name=ParNew
>
>  CollectionTime = 466835;
>  CollectionCount = 21315;
>
>
> On Fri, Dec 6, 2013 at 9:58 AM, Jason Wee <peichieh@gmail.com> wrote:
>
>> 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 <comomore@gmail.com> wrote:
>>
>>>
>>>
>>>
>>> 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