Here you can find some performance figures:

http://infinispan.blogspot.com/2010/02/infinispan-as-local-cache.html

An open source benchmark framework was used for that, so you should be able to reproduce the results yourself: http://sourceforge.net/apps/trac/cachebenchfwk/wiki

Hi,

I'm getting similar figures. They partly have to do with the transactional mode. Using nodeLockingScheme="pessimistic" is better by roughly a factor of three, which is similar to not using any transaction at all.

Pessimistic and optimistic schemes are deprecated in favour of mvcc, however, and non-transactional isn't a real alternative either.

In any case, the figures are still way off from what I'd expected them to be. Would be nice to hear from the developers whether this is about it or whether we/they can tweak the performance noticeably.

Regards,

     Gerret

Gerret, and the rest of the audience, you're gonna have to be more precise and explain what the test that you're running, your numbers, percentage of reads vs writes, whether you're using any cache store...etc. mvcc is much better than PL and OL, particuarly because reads are non blocking. Maybe the reason of your poor performance is due to transactions lasting for too long?

Regardless, you should definitely focus on Infinispan since JBoss Cache is in maintenance mode and Infinispan's performance is much better.

on my laptop write performance was 2500 oper/sec but read performance was around 500 000 oper/sec !

     for (int i = 0; i < 100000; i++) {
            String test = (String)appCache.get(tokenNodeFqn, "" + i);
            if (i%1000 == 0 ) {
                log.info("result=" + test);
            }
        }