[kaffe] Benchmarking kaffe using XSLTMark

[Dalibor Topic]

Hi,

I thought I should follow up on my announcement to
take a closer look
at XSLTMark. First the good news: it works. All open
source java XSL
processors I could get my hands on (Xalan, Xt, Saxon)
work, too. We
get the same correctness results like JDK 1.3.1 on
linux.

The bad news: we are slower than JDK 1.3.1. Especially
xalan seems to
suffer from class and resource loading. So I'd like to
propose a
new caching scheme for our system class loader: beside
just caching
classes, it should cache the directories (i.e.
packages) in classpath
entries. That should reduce the cost of class/resource
lookup from
linear (scanning the classpath) to constant (single
classpath entry)
for most resources.

So next time the class loader is looking for "a/b/c"
it could do a
cache lookup on "a/b", and get a List of classpath
entries x,y, and z,
in that order. Now it only needs to scan x,y, and z to
find the
resource. In most cases, there would be just one entry
in the Iterator
anyway.

What do you think?

I've also attached my first draft at a FAQ.xsltmark.

dalibor topic


__________________________________________________
Do You Yahoo!?
Yahoo! - Official partner of 2002 FIFA World Cup
http://fifaworldcup.yahoo.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: FAQ.xsltmark
Type: application/octet-stream
Size: 3222 bytes
Desc: FAQ.xsltmark
Url : http://kaffe.org/pipermail/kaffe/attachments/20020530/e567145e/attachment-0003.obj