xml-xindice-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Paul Brown" <pjb...@cs.nott.ac.uk>
Subject RE: Scaleability and transaction support
Date Wed, 14 May 2003 17:51:45 GMT
Wouldn't Tomcat/other interfere with the concurrency support? Thus
rendering the results for Xindice 1.1b biased when compared to 1.0?

-----Original Message-----
From: STONE,ROBERT (HP-SanDiego,ex1) [mailto:bob.stone@hp.com] 
Sent: 14 May 2003 17:15
To: 'xindice-users@xml.apache.org'
Subject: RE: Scaleability and transaction support 

HPUX, Tomcat 4.0.3, 1Gb of memory, JDK1.3, indexing all the fields I'm
querying against
I actually run some performance testing on 1.0 and, surprise, a single
query was faster on 1.0 comparing to 1.1b. Search the archives for
'xindice1.1b performance' you'll find the short discussion around the
degraded 1.1 performance.
Bob S.
-----Original Message-----
From: Devrim Ergel [mailto:devrim.ergel@parsera.com]
Sent: Wednesday, May 14, 2003 9:13 AM
To: xindice-users@xml.apache.org
Subject: Re: Scaleability and transaction support 

These results are very good.
We also applied similar tests on 1.0. The result was disappointing. We
made two concurrent Xpath queries,  in 3000 documents(20-30 kb size),
which returns 400 of them. As a result we had "out of memory" exception.
Single result query time was good; 20-30 ms.
Our configuration: 
Xindice 1.0 (we implemented collection pooling)
Tomcat 4.0.3
Windows 2000
Pentium 4 1.6
256 Mb JVM 
What is your hardware and JVM memory settings? Do you use Tomcat as
application server? Did you ever used Xindice 1.0, is there some timing
improvement on 1.1b?
Thank you.
Devrim Ergel
----- Original Message 

From: STONE,ROBERT  <mailto:bob.stone@hp.com> (HP-SanDiego,ex1) 
To: 'xindice-users@xml.apache.org' 
Sent: Wednesday, May 14, 2003 6:34 PM
Subject: RE: Scaleability and transaction support 

We did some load tests with 75 concurrent users accessing collection of
approximately 2500 docs, 3 - 5Mb in size. 1.1b was behaving fine. 
The setup: 
1. client access through the browser
2. XPath query (retrieves about 20 docs)
3. XSL stylesheet is applied to each to render the resulting HTML
Here are some numbers:
Average Page Load: 0.74 Seconds

Minimum Page Load: 0.30 Seconds

Maximum Page Load: 23.39 Seconds

Total HTTP Requests: 90,376

Requests/Sec: 100.41

Pages/Sec: 100.41

CPU Utilization of hpma523: 23%

Load Time Standard Deviation: 1.20

# of objects in page: 1

What was puzzling - on approximately every 30th request, response will
choke up to 25 sec, bringing the overall response time to .74 sec

And don't forget to implement some sort of object pooling for the
collections. I was using jakarta.commons.pool

good luck

Bob Stone

-----Original Message-----
From: jcplerm [mailto:]
Sent: Tuesday, May 13, 2003 8:21 PM
To: xindice-users@xml.apache.org
Subject: Scaleability and transaction support 

What is the status of Xindice with regards to scaleability and
concurrency and transaction support?
How does Xindice behave under heavy loads, with lots of users querying
and updating XML docs?
Let's assume the documents are of reasonable size, let's say 400KB the
largest and 3KB in average.
Is Xindice multithreaded, or there is a separate process per client?
I am seriously considering Xindice for a project and I would appreciate
any feedback.
Thanks a lot,
Julio Lerm
Chicago, IL

View raw message