[ 
http://issues.apache.org/jira/browse/AXIS2-533?page=comments#action_12375063 ] 

Sanjiva Weerawarana commented on AXIS2-533:
-------------------------------------------

I think we should close this: its impossible for one implementation to optimize 
both time and space .. that's the classic CS tradeoff!!!!!

I'd like to (over time, not for 1.0 ;-)) resurrect the table impl of Axiom. The 
table impl is the one that trades off speed for space. The LLOM impl does the 
opposite. Once we have both working nicely we can have a fight over which one 
to make default .. or probably compromise by having a property in axis2.xml to 
set the default impl so that the user can pick it based on their predominant 
scenarios: big or small.

> Axiom is heavy-weight and slow
> ------------------------------
>
>          Key: AXIS2-533
>          URL: http://issues.apache.org/jira/browse/AXIS2-533
>      Project: Apache Axis 2.0 (Axis2)
>         Type: Bug

>   Components: om
>     Versions: 0.95
>  Environment: n/a
>     Reporter: Dennis Sosnoski
>  Attachments: ombench.zip, perf.xls, sizes1.ods, sizes1.xls, sizes2.ods, 
> sizes2.xls, times1.ods, times1.xls, times2.ods, times2.xls, try.sh
>
> In performance tests comparing Axiom with JDOM, dom4j, and Xerces DOM 
> document models Axiom was consistently both slower and bulkier than the other 
> OMs when the tree was expanded. Axiom delivered good performance for the 
> initial "parse" operation only by virtue of not building the tree. Here are 
> some time comparison figures for Axiom vs. dom4j on a fairly large SOAP 
> response (from the MS interop suite):
> Running Axiom with 10 passes on file docs/xmlformatter.xml (274920 bytes):
>  Build mn=563    Build av=5110   Walk mn=33360   Walk av=60885   Build-Walk 
> mn=33923
>  Build-Walk av=65995 Write mn=31774 Write av=52782 Mod mn=5      Mod av=6
> Running dom4j with 10 passes on file docs/xmlformatter.xml (274920 bytes):
>  Build mn=24816  Build av=28843  Walk mn=2839    Walk av=3055    Build-Walk 
> mn=27655
>  Build-Walk av=31898 Write mn=18866 Write av=19369 Mod mn=5      Mod av=6
> The most interesting figures here are the Build-Walk time (sum of 
> microseconds for the initial parse operation plus walking the document tree, 
> which forces Axiom expansion) and the Write time. The average times are 
> especially bad by comparison with dom4j, which I'd suspect is caused by a lot 
> of temporary object creation. The write times may be at least partially 
> caused by the XMLStreamWriter API, since only Axiom uses this approach for 
> output. Here are memory comparisons:
> Running Axiom memory test with 4 passes on file docs/xmlformatter.xml (274920 
> bytes):
>  Init mem=451872 Chg mem=928     First sz=15192  Avg sz=15261    Walked 
> sz=2071960
>  Avg+Walked sz=2087221
> Running dom4j memory test with 4 passes on file docs/xmlformatter.xml (274920 
> bytes):
>  Init mem=829968 Chg mem=3120    First sz=1031760 Avg sz=971976  Walked sz=0
>  Avg+Walked sz=971976
> The interesting figures here are the Avg+Walked sz values, which give the 
> total bytes of memory in use after parsing and walking the document 
> representation.
> Here are the same time and memory test results for a collection of smaller 
> SOAP documents:
> Running Axiom with 10 passes on directory docs/soaps (30 files totaling 19407 
> bytes):
>  Build mn=13610  Build av=16897  Walk mn=2332    Walk av=10523   Build-Walk 
> mn=15942
>  Build-Walk av=27420 Write mn=16079 Write av=22565 Mod mn=9      Mod av=9
> Running dom4j with 10 passes on directory docs/soaps (30 files totaling 19407 
> bytes):
>  Build mn=7507   Build av=12354  Walk mn=121     Walk av=134     Build-Walk 
> mn=7628
>  Build-Walk av=12488 Write mn=4226 Write av=5012 Mod mn=10       Mod av=10
> Running Axiom memory test with 4 passes on directory docs/soaps (30 files 
> totaling 19407 bytes):
>  Init mem=456104 Chg mem=1640    First sz=451960 Avg sz=449760   Walked 
> sz=103520
>  Avg+Walked sz=553280
> Running dom4j memory test with 4 passes on directory docs/soaps (30 files 
> totaling 19407 bytes):
>  Init mem=836392 Chg mem=7944    First sz=103824 Avg sz=33768    Walked sz=0
>  Avg+Walked sz=33768
> Note the huge memory usage in this case. It appears that Axiom has a high 
> per-document memory overhead, perhaps caused by holding on to the 
> XMLStreamReader (which suggests the references to the reader should be 
> cleared as the tree is constructed, so that once the tree is complete the 
> reader can be garbage collected).

-- 
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

Reply via email to