Model-Glue: Not optimal under load
Posted by Joe Rinehart at 10:08 AM
8 comments - Categories:
Model-Glue
I'm working with Mike Brunt over at Webapper to add Model-Glue to the list of CFPetmarket apps that have been tested. When I got preliminary results back, I was a little suprised: it was much slower than I thought it'd be.
I've never had an opportunity to compare it side-by-side with a Mach-II (or Fusebox) application, and because my Petmarket implementation used the same model code as the Mach-II implementation, I was able to use the Mach-II version as a reference for performance tuning.
It turns out I made two low-level mistakes in the framework: the CFC that most people know as "arguments.event" was re-instantiated for each <event-handler>, and, when it was instantiated, it re-instantiated the CFC that handles Event Bean population. I've tuned those two flaws out of the core, and JMeter (the Apache foundation's F/OSS load test tool) is showing that M2 and MG behave comparably under heavy (100-thread) load.
I'm testing this as a 1.1.01 release of Model-Glue right now, and if anyone else wants to give it a whirl, it's comitted to the Subversion BER URL.
Craig M. Rosenblum wrote on 04/17/06 11:41 AM
Just two serious questions...1. How do each of the framework's compare in terms of performance, creating the same basic app.
2. What kind of processing does each have to do, to allow the special handling that makes developing in x framework that much easier?
Just something I always wanted to know...
And what is reasonable and what is not...Where do you draw the line?