[ http://issues.apache.org/jira/browse/EL-5?page=all ]
Matthias Ernst updated EL-5:
----------------------------
Attachment: patch-el-cache.txt
I have stumbled over this again. The synchronized map proves to be a point of
contention in an application that runs on a highly concurrent machine (Sun
T2000 w/32 cores) with JSPs that contain many EL expressions. Several hundred
hits/s times several ten expressions/JSP leads to several (ten) thousand
ELuations per second.
The simplest patch that could possibly work is as follows: instead of using a
mutable, synchronized hashmap, use an immutable, copy-on write table where only
the access to the cache field itself is protected through a monitor. A
corresponding patch is attached.
[With Java 5, this monitor-protected field could even be replaced through a
volatile field. Doing so is legal under the revised memory model and produced
even more throughput in my non-trivial testcase.]
> [el] Contention on global cache / parseExpression
> -------------------------------------------------
>
> Key: EL-5
> URL: http://issues.apache.org/jira/browse/EL-5
> Project: Commons EL
> Type: Bug
> Versions: 1.0 Final
> Environment: Operating System: other
> Platform: Other
> Reporter: Matthias Ernst
> Attachments: patch, patch-el-cache.txt
>
> The ExpresssionEvaluatorImpl maintains a static synchronized hashmap as a
> global
> parser cache. In my tests, this proves to be a contention point in highly
> concurrent web access, even if the cache is already filled:
> "tcpConnection-8080-514" daemon prio=1 tid=0x08214890 nid=0x4e39 waiting for
> monitor entry [bd9ff000..bd9ff908]
> at java.util.Collections$SynchronizedMap.get(Collections.java:1942)
> - waiting to lock <0x45586ef0> (a
> java.util.Collections$SynchronizedMap)
> at
> org.apache.commons.el.ExpressionEvaluatorImpl.parseExpressionString(ExpressionEvaluatorImpl.java:306)
> ...
> Even pre-parsing the expressions through #parseExpression doesn't help: it
> parses the expression for syntactic correctness and returns an instance of
> JSTLExpression that, however, doesn't contain the parsing result. It is
> reparsed
> on every evaluation.
> I propose
> * evaluator instance local caches that still need synchronization, but can be
> made thread-local, page-local oder whatever by the calling application /
> container.
> * to actually maintain the parsing result in JstlExpression
> I would volunteer to develop the respective patches, if desired.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
