promotion failed

Ken--@newsgroupstats.hk dragonken at gmail.com
Fri Jul 18 17:14:48 PDT 2008


Thanks for your respnose.

My application will long live around 500M data in memory, actually that is
preloading the whole database to memory and will daily reload the whole
database.

Now, My plan is:

1) decrease the new size to 128M, it result decrease in pause time but
increase the frequency of the pause.
2) decrease CMSInitiatingOccupancyFraction to 50 see how dose fragmentation
effect
3) increase MaxTenuringThreshold

Thanks,
Ken



Jon Masamitsu wrote:
> 
> Ken,
> 
> A promotion failure is often a sign that  the  CMS
> collection did not start soon enough.  In your case
> based on the -XX:CMSInitiatingOccupancyFraction=70
> flag, it looks like the CMS collection would have started
> soon and that was a little too late.  There is probably a
> fair amount of fragmentation in the CMS generation
> at the point of the promotion failure since it looks like
> there is 100m free in the CMS generation and from
> the log fragment you sent, it looks like the promotion
> rate is very low.
> 
> If most of the data in you application is short lived,
> try increasing the MaxTenuringThreshold to see
> if you can keep more of it in the young gen until it
> dies.
> 
> A smaller CMSInitiatingOccupancyFraction will
> help as will a larger heap, but if fragmentation
> is your real problem, then those will only delay
> the promotion failure.
> 
> If it's fragementation, take a look at
> 
> http://blogs.sun.com/jonthecollector/entry/when_the_sum_of_the
> 
> for some suggestions.
> 
> Jon
> 
> 
> 
> Ken-- at newsgroupstats.hk wrote On 07/18/08 09:43,:
> 
>>Hi Jon,
>>
>>uptime is 346136 (96 hours)
>>
>>first Big GC occur at 227781 second (63th hour)
>>
>>2nd Big GC occur at 311815 second (86th hour)
>>
>>Total count of promotion failed = 2
>>
>>Regards,
>>Ken
>>
>>
>>
>>Jon Masamitsu wrote:
>>  
>>
>>>Ken,
>>>
>>>How often do your promotion failures occur?
>>>
>>>Jon
>>>
>>>Ken-- at newsgroupstats.hk wrote On 07/18/08 05:21,:
>>>
>>>    
>>>
>>>>Dear All,
>>>>
>>>>please advise any more tuning for avoiding promotion failed
>>>>
>>>>VM Options:
>>>>"C:\Program Files\Java\jdk1.6.0_04\bin\java" -server -verbose:gc
>>>>-XX:+PrintGCDetails -Xloggc:gc.log -showversion -Xms4G -Xmx4G
>>>>-XX:NewSize=512M -XX:PermSize=64M -XX:MaxPermSize=64M
>>>>-XX:+CMSClassUnloadingEnabled -XX:+DisableExplicitGC
>>>>      
>>>>
>>-XX:+UseConcMarkSweepGC
>>  
>>
>>>>-XX:+UseParNewGC -XX:SurvivorRatio=6
-XX:CMSInitiatingOccupancyFraction=70
>>>>-XX:TargetSurvivorRatio=90 -XX:+HandlePromotionFailure
>>>>-Dcom.sun.management.jmxremote.port=7191
>>>>-Dcom.sun.management.jmxremote.authenticate=false
>>>>-Dcom.sun.management.jmxremote.ssl=false
>>>>
>>>>GC Log:
>>>>227734.958: [GC 227734.958: [ParNew: 458751K->65536K(458752K), 0.4883967
>>>>secs] 2975829K->2611470K(4128768K), 0.4885529 secs] [Times: user=1.42
>>>>sys=0.20, real=0.48 secs] 
>>>>227781.363: [GC 227781.363: [ParNew (promotion failed):
>>>>458752K->458751K(458752K), 1.1739277 secs]
>>>>227782.544: [CMS: 2566464K->903491K(3670016K), 14.2218888 secs]
>>>>3004686K->903491K(4128768K), 15.4022930 secs] [Times: user=16.16
sys=0.70,
>>>>real=15.39 secs] 
>>>>227796.767: [GC [1 CMS-initial-mark: 903491K(3670016K)]
910269K(4128768K),
>>>>0.0014926 secs] [Times: user=0.02 sys=0.00, real=0.00 secs] 
>>>>227796.769: [CMS-concurrent-mark-start]
>>>>227799.448: [GC 227799.449: [ParNew: 393216K->21851K(458752K), 0.0851390
>>>>secs] 1296707K->925343K(4128768K), 0.0853620 secs] [Times: user=0.86
>>>>sys=0.00, real=0.09 secs] 
>>>>227799.536: [CMS-concurrent-mark: 2.670/2.767 secs] [Times: user=9.72
>>>>sys=0.03, real=2.77 secs] 
>>>>227799.536: [CMS-concurrent-preclean-start]
>>>>227800.624: [GC 227800.624: [ParNew: 415067K->11565K(458752K), 0.0345770
>>>>secs] 1318559K->915056K(4128768K), 0.0347455 secs] [Times: user=0.27
>>>>sys=0.06, real=0.05 secs] 
>>>>227800.684: [CMS-concurrent-preclean: 0.995/1.148 secs] [Times:
user=4.38
>>>>sys=0.23, real=1.14 secs] 
>>>>227800.684: [CMS-concurrent-abortable-preclean-start]
>>>>227801.270: [CMS-concurrent-abortable-preclean: 0.475/0.587 secs]
[Times:
>>>>user=2.22 sys=0.09, real=0.59 secs] 
>>>>227801.273: [GC[YG occupancy: 240407 K (458752 K)]227801.273: [Rescan
>>>>(parallel) , 0.0092978 secs]227801.282: [weak refs processing, 0.0002179
>>>>secs]227801.283: [class unloading, 0.0082907 secs]227801.291: [scrub
>>>>      
>>>>
>>symbol
>>  
>>
>>>>& string tables, 0.0066056 secs] [1 CMS-remark: 903491K(3670016K)]
>>>>1143898K(4128768K), 0.0275619 secs] [Times: user=0.03 sys=0.00,
real=0.03
>>>>secs] 
>>>>227801.302: [CMS-concurrent-sweep-start]
>>>>227803.075: [CMS-concurrent-sweep: 1.773/1.773 secs] [Times: user=4.19
>>>>sys=0.03, real=1.77 secs] 
>>>>227803.075: [CMS-concurrent-reset-start]
>>>>227803.103: [CMS-concurrent-reset: 0.028/0.028 secs] [Times: user=0.06
>>>>sys=0.00, real=0.03 secs] 
>>>>227816.785: [GC 227816.786: [ParNew: 404776K->25694K(458752K), 0.2086003
>>>>secs] 1188297K->809215K(4128768K), 0.2087517 secs] [Times: user=1.13
>>>>sys=0.19, real=0.20 secs] 
>>>>
>>>>Regards,
>>>>Ken
>>>>
>>>>
>>>>
>>>> 
>>>>
>>>>      
>>>>
>>>
>>>    
>>>
>>
>>  
>>
> 
> 
> 

-- 
View this message in context: http://www.nabble.com/promotion-failed-tp18528055p18539751.html
Sent from the OpenJDK Hotspot Garbage Collection mailing list archive at Nabble.com.




More information about the hotspot-gc-dev mailing list