As I test more with CF2016 Enterprise 64-bit, I'm finding weird and inconsistent performance issues. At the moment, I have two queries against my Oracle database that vary from 2ms to 3000+ms from one run to the next. Both queries are against relatively small tables that have all necessary indices, and both use bind variables. In production (CF9 32-bit), I'm not seeing anything like that. My Oracle analytics show no problems with the queries, no recommendations about tuning.
I can't help but assume that this release is too fresh and will require significant updates to get it to be production-ready. It feels like nobody else is trying to use it yet, so I can't find any other posts about this. Not to mention that the stupid decision to stop numbering and adopt this new clunky name makes it even harder to get decent Google results.
Has anybody else begun testing their applications in this new version and found weird issues like this? I'm wondering if I need to start trying alternative JVMs or something.
Could you please tell me the following:
1) Which platform did you test ColdFusion 2016
2) Which bit and version of Oracle is used for CF9 and ColdFusion 2016
3) Please share the screenshot of query response time
I've been testing on Fedora Release 10 (64-bit), with Apache 2.2.29. My version of Oracle is 11g 64-bit, but that shouldn't matter since it's the same on both CF versions. I'm not going to share screenshots.
I did figure out that at least much of my issue came down to the random number generator. Apparently with the 64-bit install on Linux, actions like connecting to a database require the random number generator to create secure IDs, and the default settings using /dev/random don't provide enough entropy. I found some articles advising using /dev/./urandom, which was incorrectly configured in the JVM settings in both my new CF2016 install AND my failed CF11 install from last year. So it seems like out of the box, CF2016 is doomed on 64-bit Linux till you fix that. Once I did, my performance became normal, if not slightly better than 32-bit CF9.
I'll be doing some more testing over the course of this week to see if there are any other issues to resolve.