My biggest pitfalls when writing a large-scale C2 site: 1) Using <xsl:for-each>. This XSL construct forces the XSL transformer to what for the whole XML document before moving to the next instruction. 2) Not using DB connection-pooling (yeah, yeah, supid mistake ;-) 3) Having the object pools so large that the server ran out of physical memory and started swapping.
Hope it helps On Fri, 2001-11-23 at 19:50, Michael Homeijer wrote: > ----- Original Message ----- > From: "Michael Homeijer" <[EMAIL PROTECTED]> > To: <[EMAIL PROTECTED]> > Sent: Friday, November 23, 2001 8:38 PM > Subject: Cocoon 2 RC2 performance disappointment > > > > Hi, > > > > We built a site using Cocoon 1. Whe are now in the process of scaling the > > hardware so it can handle the requested amount of users. > > > > Next we started a project to perform migration to Cocoon 2. I think we use > > all mechanisms (at least the ones I know) to create a site that performs > > well. But... under heavy load, the performance of the Cocoon 1 site > remains > > stable (slow but stable) and the Cocoon 2 site collapses. The initial > > performance of the Cocoon 2 site is higher, but under "heavy load", it > will > > not do very much anymore. > > > > Mechanisms we used: > > - Split pages in multiple parts based on caching posibilities. Ie. parts > > that can use the same cache strategies (ie same key and validity) are > > grouped together. > > - The parts that can be cached are also translated to HTML using XSL as > much > > as possible. > > - The XSL files are kept as small as possible. We tried to minimize the > > amount of templates (initially there were about 60 templates). > > > > This site should be our showcase and one of the most complex ones we will > > build because of the very dynamic nature of the site and high amount of > > personalisation based on countries, companies and user profiles. Most > pages > > contain at least three to four calls to session beans. > > > > A quote from a report about the migration: > > "The expected performance improvement was not confirmed by the obtained > > results. Performance of the c2 site was dramatically lower than of the c1 > > site. In 10 minutes, the c1 site was able to process 16283 catalogue > > requests and send out 6MB of pages, the c2 site managed 1011 requests and > > 400 kB." > > > > I could use some help tuning this site. Any suggestions are welcome. > > > > TIA, > > Michael Homeijer > > > > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, email: [EMAIL PROTECTED] > -- Sergio Carvalho --------------- [EMAIL PROTECTED] If at first you don't succeed, skydiving is not for you --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, email: [EMAIL PROTECTED]
