Terracotta BigMemory & ColdFusion @ Adobe MAX 2011
Posted by Mike Brunt at 6:31 AM
87 comments - Categories: Caching | ColdFusion | JRun-J2EE
We just presented the use of Terracotta BigMemory and Adobe ColdFusion on a Sun JVM at Adobe MAX 2011. The demonstration went well despite some challenges we faced, the main one being how to show that BigMemory was actually doing anything. This was largely due to the fact that we could not get a probe, supplied by Terracotta, to work whilst on-site at MAX, this was mainly due to network inconsistencies.
Rex Vincent did most of the heavy lifting for the demo, having created a "mini application from hell" with lot's of numbing queries, looping and session-scope nightmares. The main point being to do a lot of work in the JVM memory generations. We set the start and max memory settings down to 256MB (very low of course). After about 2 minutes of the first test, ColdFusion hung throwing out of memory errors "gc overhead limit exceeded" and the maximum number of busy threads peaked at 98 and hung there. This test had no caching and no BigMemory. We then enabled caching and BigMemory and ran the same test again, this time there was no hanging and we actually hit over 200 busy threads so we were able to show that BigMemory is beneficial.
This was the first in no doubt a good number of lab tests with BigMemory on ever larger JVM heaps which I where we expect to see even greater benefits.
Vladimir Rodionov wrote on 10/15/11 2:16 PM
Mike,I am a developer of Koda, which is off heap caching solution for Java similar to BigMemory. Its approx 7-8 times faster (I am working on a performance optimizations and will make it 10-15 faster than BigMemory) and has much lower latency than BigMemory (99.99% mixed get/put 90/10 request latencies are < 100 us, maximum latency due to OS interference is ~10-20 ms and can be decreased significantly by optimizing hardware and OS for low latencies applications ). It produces virtually no single Java object on a get/put query path, thus making it suitable for really low latency applications. I have test results, which are published on my blog site:
http://designedtoscale.blogspot.com
These results are quite outdated and do not reflect latest performance and latency improvements.
I do not have access to the expensive boxes and the maximum Koda cache size I have ever tested was 40GB on 48GB server. It worked flawlessly. Koda is not generally available yet but will be by the end of this year. I am looking for use cases and potential customers (clients) and it seems that your Adobe ColdFusion integration would be very interesting use case for Koda (and I hope for ColdFusion as well). Drop me note if you are interested in trying Koda as a caching solution for ColdFusion.
Best Regards
Vladimir Rodionov