Bertrand Delacretaz wrote:
Le 27 sept. 04, à 09:12, Andreas Hartmann a écrit :
...Congratulations, it's a nice site and really fast! Are you allowed to talk about the implementation?
Thanks - sure I can talk, in fact I hope to be able to donate some components to Cocoon. 'll know more when the project is actually closed, which will take a few weeks: as you can see we had to cut a few corners to meet Saturdays deadline:
$ find . -type f | grep -v CVS | xargs grep TODO | wc -l 289
The performance part comes mainly from the front-end apache2 mod_cache. Simply adding the right HTTP headers and making sure the content-length header is generated as well (by setting the buffering flag on the HTML serializer) allows the front -end cache to do its job very nicely.
uh, I might have missed that part!!! would be cool if you could document that.
This is what allows us to reach the 800 pages/second mark (tested from localhost, so theoretical, but still a nice figure) on this fairly fast Dell dual 3.6GHz server. It's got 2 GB of RAM and the JVM is configured to use 512MB, but runs nicely with much less on test servers.
On a cache miss, Cocoon needs about 500msec to generate a page, not bad considering that there are several aggregations, at least two CInclude stages and maybe 10 (mostly simple) XSLT transforms on the way.
What pipeline flavor are you using? what XSLT transformer?
Currently we have put a fairly low caching lifetime on pages (30 seconds only) as it was more convenient for initial testing, but we can easily raise it without stopping the system, should the need arise.
The database is a plain MySQL, out of the box, and on that side the OJB object cache handles the typical "write seldom, read often" usage pattern very nicely.
The only thing I don't like on the first glance is the URI scheme :)
Why? You don't like it flat?
me neither, actually.
Document names like 50-2 deserve an explanation though: the TV show is organized by show number, so 50-2 means "the second subject of show 50".
yes, got that, but I would have liked 50/2 better, it would have allowed me to ask for 50 and maybe get a summary of the episode.
Having short names is also a plus when you need to put URLs on TV, as in that case the show is very short (about 8 minutes) and its main goal is to bring people on the website, so they put the URLs on screen during the show.
having / instead of - wouldn't have increased your URL size ;-)
-- Stefano, the URI PITA ;-)
smime.p7s
Description: S/MIME Cryptographic Signature
