W3C home > Mailing lists > Public > www-jigsaw@w3.org > July to August 1999

Re: benchmark tests with ZD WebBench

From: Yves Lafon <ylafon@w3.org>
Date: Wed, 11 Aug 1999 15:57:59 +0200 (MET DST)
To: "Andre Beliveau (LMC)" <lmcbeli@lmc.ericsson.se>
cc: www-jigsaw@w3.org
Message-ID: <Pine.GSO.4.10.9908111535410.21737-100000@tarantula.inria.fr>
On Tue, 10 Aug 1999, Andre Beliveau (LMC) wrote:

> Hi, 
> I am trying to make some capacity measurement of jigsaw using
> WebBench, a web server benchmark suite from ZDnet.
> I have the following configuration
> PC's running Windows NT 4, JDK 1.2.2, Jigsaw 2.0.3 (latest)
> and WebBench 3.0
> After installing Jigsaw and Webbench I can run a set of static
> test (static pages only) for a while until (without any specific
> reason) I get a result code 404 on one of my get operations.
> When I run jigsaw with "-trace" option I can see that I try to
> fetch 
> http://localhost:8001/wb30tree/zdwb_1/zdwb_1/zdwb_2/
> we get back
> http://localhost:8001/wb30tree/zdwb_1/zdwb_1/zdwb_1/zdwb_2/

The trace is not thread safe, maybe the output was mixed! It is more
reliable to see the generated logs.

> it seems to be getting mixed up with the directories.

I have no clear idea of what may happen, each child of a container which
is a container is uniquely identified with a serial number, maybe you
didn't have luck and bumped into a clash. It is calculated with the
identifier and the date.
     * Get the container Key. This key must be unique and unchanged
     * during the container life.
     * @return a String instance.
    public Integer getKey() {
	Integer key = (Integer) getValue(ATTR_KEY, null);
	if (key == null) {
	    key = new Integer(getIdentifier().hashCode() ^ 
			      (new Date().hashCode()));
	    setValue(ATTR_KEY, key);
	return key;
So the risks of a clash is quite low, but it can happen.

> I can reproduce the problem every time I start the benchmark 
> test but the web page which returns 404 is different every time.
> Can there be something screw up with the stores ?

Try to delete the whole wb30tree, and reindex it, (just point your browser

But if it happens only after a while, it is more than strange. We did a
lot of heavy use tests since 2.0alpha and we never got this kind of
problem :(
If after the delete/reindexation it happens again, can you make available
to us a zip of the stores?

A note if you are doing benches, the setup in the distribution is for
"normal" servers, you may want to increase the number of clients, threads
and such to fit your needs.

      /\          - Yves Lafon - World Wide Web Consortium - 
  /\ /  \        Architecture Domain - Jigsaw Activity Leader
 /  \    \/\    
/    \   /  \   http://www.w3.org/People/Lafon - ylafon@w3.org    
Received on Wednesday, 11 August 1999 09:59:08 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 22:41:27 UTC