Hi, As I posted a few weeks back I have been running a series of load tests and have come to the following key conslusions
- *Tapestry is fast* - the response times even under load are low. It isn't alway's fastest solution but it is sufficiently fast for me to focus my tuning attention elsewhere - *Tapestry does not leak memory *- One of the concerns I have heard raised about tapestry's approach is that it is likely to be heavy on memory. In fact I found the opposite. It was lower than struts/jsp and in the extended run the memory was pretty flat - *Tapestry is faster than struts for forms apps* - Tapestry seems to excel as the application complexity gets significant. This is probably a feature of its pooling model. - *Tapestry doesn't break easily and when it does it recovers* - Tapestry does slow down under extreme load, but it does break or hit bottlenecks. This is a really positive thing. It also recovers when the load goes away. - *More CPU's doesn't always equal better* - in a number of tests performance didn't increase linearly. 2 CPU's did double throughput but 4 didn't. This is often a feature of the OS/platform but would suggest in a virtualised environment that running more dual core instances is better than running fewer quad core instances - *64bit is faster than 32 bit* - This one surprised me a bit - I had heard and thought that 64bit was a bit slower as it pushes more data around. However it seems that using a 64bit JVM on a 64bit OS makes a positive difference on both Solaris and Linux. - *Linux is faster than Open Solaris X86* - This one also surprised me a bit - I expected them to be pretty near the same whereas in fact is was nearly 25% better on linux. I can't explain it - but the numbers speak for themselves. The full details of the load tests and a longer write up is at http://blog.gidley.co.uk/2009/05/tapestry-load-testing-round-up.html Ben Gidley www.gidley.co.uk b...@gidley.co.uk
