Re: [xsl] [ANNOUNCE] UWOBO 1.2.0 released

Subject: Re: [xsl] [ANNOUNCE] UWOBO 1.2.0 released
From: Claudio Sacerdoti Coen <sacerdot@xxxxxxxxxxx>
Date: Thu, 8 Nov 2001 10:37:20 +0100
> The last time I tried using Java serialization for a complex data structure,
> I found that writing it out as XML and then parsing the XML to reload it was
> much faster, mainly because the XML was so much more compact. But I'm
> interested if other people have a different experience.

Some results I get on the stylesheets used in project HELM (11917 lines
of XSLT) with UWOBO embedding Xalan-J_2.2.D3:

               Reloading:        De-serializing:     Gain:
Fast machine:   10s                  7s               30%
Slow machine:   30s                 24s               20%

The gain is not much, but it is there.
By the way, I can explain why we have choosed to implement the feature.
UWOBO is meant to provide on-line XSLT-processing to third parties
(in our case contributors to HELM library). This means that anyone
can just load its own stylesheet (in our case adding it to the default
stylesheet chain) and after that everyone can use it. We want to reach
a minimal degree of fault-tolerance: when tomcat crashes or the machine
reboots we want to provide the same set of stylesheets as before. To do
that, we can't just reload them, because the original stylesheet were
identified using an URL and, in the meantime, they can have been deleted 
or modified. Hence we must implement a cache, either of the XML source
or using Java serialization. We suppose the number of stylesheets to
become dramatically bigger and bigger in the near future. Hence even
small gains (in the order of half a second per stylesheet) are useful.


Real name: Claudio Sacerdoti Coen
PhD Student in Computer Science at University of Bologna
E-mail: sacerdot@xxxxxxxxxxx

 XSL-List info and archive:

Current Thread