Allocation Failure in std_server0.out log

[Former Member]

I am running EP Sp 14 and when I look ing the std_server0.out log under the /usr/sap/SID..../work...I noticed the following:

AF Allocation Failure cannot allocate 4300 bytes etc

EP stays up but when clicking thourhg an error will appear the dispatcher is up ut server is down....5 minutes later server is back up.

I have adjusted the server max param to 1500 M which reduces the error occruence....running on Solaris 64 bit oracle...

Any seen Allocation failure in std_server0.out u

[Former Member]

Hi,

I am facing the same problem with my portal. I have tried everything possible.(notes 716927,893149), however I face exactly the same problem.

We cannot proceed with the development due to this problem.

Can someone please help urgently?

Thanks.

[Former Member]

I set the Xmx and Xms on server to 2048 according to SAP AIX JVM note. and also set the parameter to limit MGC full GC per Eric's response. But I am still getting the following message:

<AF[2]: Allocation Failure. need 536 bytes, 65182 ms since last AF or CON>

<AF[2]: managing allocation failure, action=1 (0/1518126872) (79901416/79901416)

>

<GC(2): Bytes Traced =91425912 (Foreground: 53206137+ Background: 38219775) Stat

e = 0 >

<GC(2): GC cycle started Wed Nov 2 16:48:15 2005

<GC(2): freed 1312646336 bytes, 87% free (1392547752/1598028288), in 276 ms>

<GC(2): mark: 223 ms, sweep: 53 ms, compact: 0 ms>

<GC(2): refs: soft 0 (age >= 32), weak 180, final 9378, phantom 0>

<AF[2]: completed in 279 ms>

Any other ideas to remove Allocation Failure. AIX box has 8 GB and it 64 BIT aix....should I use 4000 for MAx heap

Thanks

Mike

[Former Member]

Mike,

I'm not sure if you can eliminate them completly. You may be able to reduce the frequency of them occuring. The note gives you a good guideline but everyones portal is different so you will have to start playing with the differet options to see how they impact the way your portal runs. One of the major trade-offs in GC is that as you increase the size of the VM, which tends to decrease the frequency with which GC occurs, when a full GC does occur it takes longer. As a start I would up the size of the VM and see how that impacts things.

[Former Member]

Hi Mike

Make sure your IBM JDK version is the latest [Preferably SR9].

SAP Note: 893149 gives some information on such Allocation failures

Might work !!!! Just try

Ashish

[Former Member]

Eric,

How would I set parallel sweep up in Sun or AIX?

Thanks

Mike

[Former Member]

Try this link for IBM.

http://www-128.ibm.com/developerworks/ibm/library/i-gctroub/index.html

And this link

http://www-128.ibm.com/developerworks/ibm/library/i-garbage3.html

The specific option for IBM is

-Xgcpolicy:optavgpause

Be aware that Sun and IBM use different VMs so the options may be different. I would assume that the Sun website has something similar.

[Former Member]

Allocation failures are a normal part of operation. The reason that it appears that the engine is not running is that the JVM has gone into a stop the world garbage collection. This means that the only thing the VM will do is perform a garbage collection, so your portal requests won't get served. Tuning of the engine is still a bit of an art with SAP as you have seen. The downside to increasing the heap size, which means fewer allocation failures, is that the full garbage collection takes longer. You may want to set your garbage collection to do a paralell sweep phase (I think this is possible with Sun). Check the Sun documentation on how to do this. Garbage collection is divided into three phases. Sweep, Mark, and Compaction. If you set the vm for a parallel sweep phase the engine will keep running while the vm trys to do the sweep phase of the gc. Provided it doesn't have another AF during that time then you won't have an issue where the engine becomes unresponsive.