This is wicked cool ... hurry up and commit the darn BigInteger -> Long replacement :).
Thanks E! Alex On 3/13/07, Emmanuel Lecharny <[EMAIL PROTECTED]> wrote:
Hi guys, this is the first time I do some tests against ADS 1.5. The results are quite interesting. Here are the very first comment I can make : 1) The best result I can get is around 160 random search requests per second (sun JVM, not frankly the fastest one ..., 10K entries, 10K random searches). FYI, 1.0 was able to deliver 215 req/s back in october, on my laptop 2) 1.5 using Long instead of BigInteger is much faster, 225 req/s, which is slightly faster than ADS 1.0 Don't be afraid : a *lot* of things has been added in 1.5, without any performance improvment being introduced so far. This test was not done to see if ADS 1.5 was good or bad, but to see what can be the difference if the backend stores Long instead of BigInteger for indices. And so far, the result is impressive : 40% improvment ! I'm just begging about introducing this change in 1.0 to see how faster the server will be now ;) Ok, now, we have a lot of fat and nasty bugs to fix in 1.5, so performance tests are postponed until we release the 1.5.0 version, soon. Then we will try to improve the beast. (make it work first, then make it go faster) Emmanuel
