Pauses with the CMS

Y. S. Ramakrishna y.s.ramakrishna at oracle.com
Wed Oct 27 09:11:58 PDT 2010


There were two related CR's which may be affecting you:

(1) 6692906 CMS: parallel concurrent marking may be prone to hanging or stalling mutators for periods of time
     I do not think this is affecting you.

(2) 6919638 CMS: ExplicitGCInvokesConcurrent misinteracts with gc locker

I suspect it could be this latter CR that's affecting you. If so, it's
fixed in 6u22 and in 6u23 (and in 7 since 7bb95), so you might want to
try those. There's also a workaround documented in the CR.

-- ramki


On 10/27/10 01:50, mrk wrote:
> Hi All
> 
> Hi
> 
> We have made a webapp running in JBoss 4.2.3 on RedHat Linux 5.2 on a 12 
> core IBM machine. Lately we have seen long response times, which seems 
> to be caused be GC activities. The pattern is more or less as follows:
> All requests seem to come at normal rate, but no bytes are sent until 
> suddenly all responses are "flushed".
> At same time the gc.log has this pattern:
> 
> // Here the unresponsiveness starts
> 36788.309: [GC [1 CMS-initial-mark: 1919297K(4456448K)] 
> 1969797K(5111808K), 0.1473460 secs] [Times: user=0.15 sys=0.00, 
> real=0.15 secs]
> Total time for which application threads were stopped: 0.1530790 seconds
> 36788.457: [CMS-concurrent-mark-start]
> 36793.886: [CMS-concurrent-mark: 5.429/5.429 secs] (CMS-concurrent-mark 
> yielded 46 times)
> [Times: user=6.69 sys=0.07, real=5.43 secs]
> 36793.886: [CMS-concurrent-preclean-start]
> (cardTable: 15956 cards, re-scanned 15956 cards, 1 iterations)
> 36793.988: [CMS-concurrent-preclean: 0.089/0.101 secs] 
> (CMS-concurrent-preclean yielded 9 times)
> [Times: user=0.12 sys=0.01, real=0.10 secs]
> 36793.988: [CMS-concurrent-abortable-preclean-start]
> (cardTable: 75 cards, re-scanned 75 cards, 1 iterations)
> Application time: 5.6923460 seconds
> Total time for which application threads were stopped: 0.0170000 seconds
> (cardTable: 96 cards, re-scanned 96 cards, 1 iterations)
> (cardTable: 207 cards, re-scanned 207 cards, 1 iterations)
> (cardTable: 45 cards, re-scanned 45 cards, 1 iterations)
> (cardTable: 157 cards, re-scanned 157 cards, 1 iterations)
> (cardTable: 88 cards, re-scanned 88 cards, 1 iterations)
> (cardTable: 213 cards, re-scanned 213 cards, 1 iterations)
> (cardTable: 84 cards, re-scanned 84 cards, 1 iterations)
> (cardTable: 143 cards, re-scanned 143 cards, 1 iterations)
> (cardTable: 40 cards, re-scanned 40 cards, 1 iterations)
> (cardTable: 109 cards, re-scanned 109 cards, 1 iterations)
> (cardTable: 77 cards, re-scanned 77 cards, 1 iterations)
> (cardTable: 156 cards, re-scanned 156 cards, 1 iterations)
> (cardTable: 132 cards, re-scanned 132 cards, 1 iterations)
> (cardTable: 119 cards, re-scanned 119 cards, 1 iterations)
> (cardTable: 115 cards, re-scanned 115 cards, 1 iterations)
> (cardTable: 195 cards, re-scanned 195 cards, 1 iterations)
> (cardTable: 163 cards, re-scanned 163 cards, 1 iterations)
> (cardTable: 166 cards, re-scanned 166 cards, 1 iterations)
> (cardTable: 149 cards, re-scanned 149 cards, 1 iterations)
> (cardTable: 218 cards, re-scanned 218 cards, 1 iterations)
> (cardTable: 46 cards, re-scanned 46 cards, 1 iterations)
> (cardTable: 57 cards, re-scanned 57 cards, 1 iterations)
> (cardTable: 213 cards, re-scanned 213 cards, 1 iterations)
> (cardTable: 189 cards, re-scanned 189 cards, 1 iterations)
> (cardTable: 104 cards, re-scanned 104 cards, 1 iterations)
> (cardTable: 60 cards, re-scanned 60 cards, 1 iterations)
> (cardTable: 368 cards, re-scanned 368 cards, 1 iterations)
> (cardTable: 118 cards, re-scanned 118 cards, 1 iterations)
> (cardTable: 129 cards, re-scanned 129 cards, 1 iterations)
> (cardTable: 291 cards, re-scanned 291 cards, 1 iterations)
> (cardTable: 105 cards, re-scanned 105 cards, 1 iterations)
> (cardTable: 63 cards, re-scanned 63 cards, 1 iterations)
> (cardTable: 73 cards, re-scanned 73 cards, 1 iterations)
> (cardTable: 165 cards, re-scanned 165 cards, 1 iterations)
> (cardTable: 69 cards, re-scanned 69 cards, 1 iterations)
> (cardTable: 157 cards, re-scanned 157 cards, 1 iterations)
> CMS: abort preclean due to time [37 iterations, 13 waits, 4954 cards)] 
> 36799.161: [CMS-concurrent-abortable-preclean: 2.965/5.173 secs] 
> (CMS-concurrent-abortable-preclean yielded 166 times)
> [Times: user=5.12 sys=0.10, real=5.17 secs]
> Application time: 4.9945030 seconds
> 36799.172: [GC[YG occupancy: 572074 K (655360 K)]36799.172: [GC 
> 36799.173: [ParNew
> Desired survivor size 67108864 bytes, new threshold 2 (max 15)
> - age 1: 59654872 bytes, 59654872 total
> - age 2: 26827232 bytes, 86482104 total
> - age 3: 2613360 bytes, 89095464 total
> - age 4: 929600 bytes, 90025064 total
> - age 5: 893176 bytes, 90918240 total
> - age 6: 795736 bytes, 91713976 total
> - age 7: 809240 bytes, 92523216 total
> - age 8: 936216 bytes, 93459432 total
> - age 9: 971376 bytes, 94430808 total
> - age 10: 624472 bytes, 95055280 total
> - age 11: 883616 bytes, 95938896 total
> - age 12: 817296 bytes, 96756192 total
> - age 13: 977992 bytes, 97734184 total
> - age 14: 733816 bytes, 98468000 total
> - age 15: 654744 bytes, 99122744 total
> : 572074K->110988K(655360K), 0.1855140 secs] 
> 2491372K->2031059K(5111808K) icms_dc=0 , 0.1866380 secs] [Times: 
> user=0.72 sys=0.00, real=0.19 secs]
> 36799.359: [Rescan (parallel) (Survivor:12chunks) Finished young gen 
> rescan work in 3th thread: 0.032 sec
> Finished young gen rescan work in 2th thread: 0.035 sec
> Finished young gen rescan work in 0th thread: 0.035 sec
> Finished young gen rescan work in 1th thread: 0.035 sec
> Finished remaining root rescan work in 1th thread: 0.030 sec
> Finished remaining root rescan work in 3th thread: 0.033 sec
> Finished remaining root rescan work in 2th thread: 0.032 sec
> Finished remaining root rescan work in 0th thread: 0.042 sec
> Finished dirty card rescan work in 2th thread: 0.029 sec
> Finished dirty card rescan work in 3th thread: 0.031 sec
> Finished dirty card rescan work in 0th thread: 0.019 sec
> Finished dirty card rescan work in 1th thread: 0.031 sec
> Finished work stealing in 3th thread: 0.000 sec
> Finished work stealing in 0th thread: 0.000 sec
> Finished work stealing in 2th thread: 0.000 sec
> Finished work stealing in 1th thread: 0.000 sec
> , 0.0959640 secs]36799.455: [weak refs processing, 0.0121220 secs] [1 
> CMS-remark: 1920070K(4456448K)] 2031059K(5111808K), 0.2952350 secs] 
> [Times: user=1.11 sys=0.00, real=0.30 secs]
> Total time for which application threads were stopped: 0.3066430 seconds
> 36799.468: [CMS-concurrent-sweep-start]
> 36805.824: [CMS-concurrent-sweep: 6.355/6.356 secs] 
> (CMS-concurrent-sweep yielded 515 times)
> [Times: user=8.76 sys=0.24, real=6.35 secs]
> 36805.824: [CMS-concurrent-reset-start]
> 36805.903: [CMS-concurrent-reset: 0.079/0.079 secs] 
> (CMS-concurrent-reset yielded 0 times)
> [Times: user=0.64 sys=0.01, real=0.07 secs]
> Application time: 6.5590510 seconds
> // About here the server starts sending out responses
> 36806.044: [GC 36806.045: [ParNew
> Desired survivor size 67108864 bytes, new threshold 2 (max 15)
> - age 1: 55456640 bytes, 55456640 total
> - age 2: 54278984 bytes, 109735624 total
> : 635276K->131072K(655360K), 0.3230430 secs] 
> 2554507K->2089990K(5111808K) icms_dc=0 , 0.3242750 secs] [Times: 
> user=1.02 sys=0.02, real=0.33 secs]
> Total time for which application threads were stopped: 0.3422430 seconds
> Application time: 0.7628480 seconds
> 
> We have tried various GC setups, but we always end up more or less like 
> this.
> Our current parameters:
> -Xms4096m -Xmx5120m -XX:MaxPermSize=384m
> -XX:+UseConcMarkSweepGC -XX:+ExplicitGCInvokesConcurrent
> -XX:+PrintClassHistogram -XX:+PrintGCApplicationConcurrentTime 
> -XX:+PrintGCApplicationStoppedTime -XX:+PrintGCDetails
> -XX:ReservedCodeCacheSize=64m -XX:NewSize=768m -XX:SurvivorRatio=4
> -XX:+CMSParallelRemarkEnabled -XX:+CMSScavengeBeforeRemark 
> -XX:+CMSIncrementalMode
> -XX:+CMSIncrementalPacing -XX:CMSIncrementalDutyCycleMin=0 
> -XX:CMSIncrementalDutyCycle=10
> -XX:+UseTLAB -XX:ParallelGCThreads=4
> -XX:+UseParNewGC -XX:MaxGCPauseMillis=1000 -XX:GCTimeRatio=10 
> -XX:+CMSPermGenSweepingEnabled
> -XX:+PrintCMSInitiationStatistics -XX:PrintCMSStatistics=1 
> -XX:+PrintGCTaskTimeStamps -XX:+PrintGCTimeStamps
> -verbose:gc -XX:+TraceClassLoading -XX:+TraceClassUnloading
> -XX:+PrintTenuringDistribution
> 
> How can we avoid the unresponsiveness?
> 
> We have tried with both JDK 1.6.0_06 and JDK 1.6.0_021.
> 
> It seems that the problem have a lot in common with:
> http://old.nabble.com/long-unaccounted-pause-during-concurrent-mark-phase-of-ParNew-cycle-to27459365.html#a27463304 
> 
> But setting -XX:-CMSConcurrentMTEnabled does not seem to help?
> 
> med venlig hilsen / Best regards,
> Morten Knudsen
> 
> 
> 
> 
> 


More information about the hotspot-gc-dev mailing list