More CPU Utilization

[mvpratap]

Hi all,

     I am running infinispan in embedded with distribution mode(3 nodes).With infinispan integration i am getting more cpu utilization(nearly 20 times more) than the without Infinispan.

hprof reference

===========

CPU TIME (ms) BEGIN (total = 206851744)

rank   self  accum   count trace method

   1 61.48% 61.48%  124327 343962 java.util.concurrent.locks.LockSupport.parkNanos

   2 18.06% 79.54%   13118 336943 java.util.concurrent.locks.LockSupport.park

   3  2.66% 82.20%    1395 372873 java.util.concurrent.locks.LockSupport.park

   4  2.31% 84.51%    3219 366886 java.net.SocketInputStream.read

   5  2.31% 86.83%    3603 367136 java.util.concurrent.locks.LockSupport.park

   6  1.24% 88.07%    3137 345644 java.lang.Object.wait

   7  1.24% 89.30%   12735 334908 java.util.concurrent.locks.LockSupport.parkNanos

   8  1.22% 90.52%    2575 360649 java.lang.Object.wait

   9  1.22% 91.74%    2576 360762 java.net.SocketInputStream.read

  10  1.22% 92.96%    2528 363747 org.jboss.mq.SpyMessageConsumer.receive

  11  1.22% 94.18%     253 363102 java.net.PlainSocketImpl.accept

  12  1.22% 95.40%      43 360924 EDU.oswego.cs.dl.util.concurrent.ClockDaemon.nextTask

  13  1.22% 96.62%      42 369602 EDU.oswego.cs.dl.util.concurrent.SynchronousChannel.poll

  14  1.16% 97.78%      16 359239 java.lang.ref.ReferenceQueue.remove

  15  1.09% 98.87%       8 374938 java.util.concurrent.locks.LockSupport.park

  16  0.17% 99.04%     115 337048 java.util.concurrent.locks.LockSupport.parkNanos

  17  0.15% 99.19%     802 336976 java.net.PlainSocketImpl.doConnect

  18  0.13% 99.32%     615 336398 java.util.concurrent.locks.LockSupport.park

  19  0.13% 99.45%       2 369061 java.net.PlainSocketImpl.accept

  20  0.13% 99.57%       2 366588 java.net.PlainSocketImpl.accept

  21  0.07% 99.64%       1 374060 java.net.SocketInputStream.read

  22  0.06% 99.71%      36 377230 java.util.concurrent.locks.LockSupport.parkNanos

  23  0.02% 99.73%       2 345400 java.lang.Object.wait

CPU TIME (ms) END

TRACE 343962:

        java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:Unknown line)

        java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.awaitNanos(AbstractQueuedSynchronizer.java:Unknown line)

        java.util.concurrent.DelayQueue.take(DelayQueue.java:Unknown line)

        java.util.concurrent.ScheduledThreadPoolExecutor$DelayedWorkQueue.take(ScheduledThreadPoolExecutor.java:Unknown line)

TRACE 336943:

        java.util.concurrent.locks.LockSupport.park(LockSupport.java:Unknown line)

        java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.await(AbstractQueuedSynchronizer.java:Unknown line)

        java.util.concurrent.LinkedBlockingQueue.take(LinkedBlockingQueue.java:Unknown line)

        java.util.concurrent.ThreadPoolExecutor.getTask(ThreadPoolExecutor.java:Unknown line)

TRACE 372873:

        java.util.concurrent.locks.LockSupport.park(LockSupport.java:Unknown line)

        java.util.concurrent.SynchronousQueue$TransferStack.awaitFulfill(SynchronousQueue.java:Unknown line)

        java.util.concurrent.SynchronousQueue$TransferStack.transfer(SynchronousQueue.java:Unknown line)

        java.util.concurrent.SynchronousQueue.take(SynchronousQueue.java:Unknown line)

Can any one provide help on this.

Kindly see the attached configuration documents.

Thanks.

[rvansa]

The hprof output says just that there are many threads just waiting for some job/lock: these do not account for CPU utilization. What are you doing with Infinispan exactly? I don't understand the difference 'before' and 'after'.

[mvpratap]

Before means  project without using infinispan..No infinispan at all.Simply i am using ConcurrentHashMap's for storage.

After means project Merged with infinispan..using distributed cache I am storing in Cache.

[sannegrinovero]

I wouldn't be too surprised, serializing and transmitting each entry as it's updated has a cost.

You can get a significant benefit from using Infinispan but it depends how you use it, just dropping it in as a replacement for a ConcurrentHashMap might not be a good idea. Generally though if CHM works fine why would you use Infinispan?

The people move to Infinispan is because the data size gets too large that CHM is no longer an option, or because you need multiple servers to read the same data efficiently - but writes will of course be slower than a CHM.

[mvpratap]

Yes,

     Storing/retrieving data in  concurrentHashMap is faster because it is available in local..

But storing/retreiving data in Infinispan Cache is some what costlier Yes I agree...But it should not take 100% cpu right...With local  ConcurrentHashMap it is not taking more than 5%cpu..

Please provide some solution..