harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Xiao-Feng Li (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HARMONY-3906) [drlvm][gc_gen]Significant performance degradation after switch on GC v5
Date Mon, 21 May 2007 07:16:16 GMT

    [ https://issues.apache.org/jira/browse/HARMONY-3906?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12497367
] 

Xiao-Feng Li commented on HARMONY-3906:
---------------------------------------

Pavel, I tried your test case on my desktop, the result didn't show big time difference´╝î
as given below:

========================GCv4.1=============================
C:\temp>java.exe -XX:vm.dlls=gc_cc.dll -classpath . ExceptionsWeakRefBenchmark
threads quantity = 1
        633     414     301     243     183
        3559    2032    1218    826     440
        707     471     348     290     230
        3820    2230    1456    1068    681

threads quantity = 2
        378     241     179     144     109
        2186    1215    733     468     253
        414     277     214     180     141
        2258    1328    865     622     404

threads quantity = 4
        385     247     182     151     117
        2151    1219    732     472     255
        423     291     219     198     143
        2280    1347    855     617     397

threads quantity = 8
        387     250     178     146     114
        2080    1206    709     474     256
        420     284     212     178     144
        2242    1308    843     620     398

threads quantity = 16
        405     247     186     149     115
        2115    1215    753     500     269
        418     284     220     181     146
        2256    1367    850     609     404

PASSED

========================GCv5=============================
C:\temp>java.exe  -classpath . ExceptionsWeakRefBenchmark
threads quantity = 1
        639     415     298     242     182
        3522    1997    1228    838     448
        702     470     351     292     236
        3782    2248    1467    1092    699

threads quantity = 2
        387     263     189     147     117
        2241    1214    703     501     275
        419     280     212     178     147
        2313    1379    853     663     407

threads quantity = 4
        395     247     185     152     113
        2074    1178    715     489     271
        417     290     214     186     149
        2283    1325    864     616     408

threads quantity = 8
        382     253     187     154     118
        2077    1193    716     500     268
        417     281     215     183     148
        2226    1307    844     611     425

threads quantity = 16
        388     257     183     151     118
        2062    1154    712     498     270
        420     292     220     186     156
        2217    1312    865     642     451

PASSED
==================================================================

I wonder if the data above are reasonable´╝î and please advise how I can reproduce it. Thanks.

> [drlvm][gc_gen]Significant performance degradation after switch on GC v5
> ------------------------------------------------------------------------
>
>                 Key: HARMONY-3906
>                 URL: https://issues.apache.org/jira/browse/HARMONY-3906
>             Project: Harmony
>          Issue Type: Bug
>          Components: DRLVM
>            Reporter: Pavel Afremov
>         Assigned To: Xiao-Feng Li
>         Attachments: ExceptionsWeakRefBenchmark.java
>
>
> I found out significant performance degradation of class unloading support after switch
on GC v5. On 1 processor machine performance degradation in micro benchmark from HARMONY-3845
 can reach 10 times, on 4 HT processors degradation reaches up 80 times. It is 1000% and 8000%
correspondently!
>  I developed additional benchmark, which shows this problem without any changes in VM.
You can find it in attachment.
>  Of cause it's a synthetic benchmark. I didn't find any performance issues on specJVM
and specJBB.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message