Ruby Performance in the Rails Development Environment
erik.debill.org
erik.debill.org
One problem I did have was that certain portions of my code would no longer work after reloading the system - for example I would store function references in variables and if those pages recompiled I would lose those variables that I set. A way I found around it was that each time a page was sent to the webserver and after it reloaded the system, I would have it reload those specific variables. Now if something like that could work in rails, where you always reload specific pieces of code that would no longer work if their underlying code was modified, then you might be able to make it work.
I'd also be interested in seeing REE stats.
Services complicate the debugging process. Also, parallel development means more co-ordination overhead. Deployments become a little trickier -- with a monolithic app, you can easily roll out one version while the other version is removed from the load balancers. With services, you start getting into supporting different versions of services (through forwards or backwards compatibility) or requiring downtime.
With both class caching turned off (development) and on (production), jruby 1.5 performed about 2x worse than REE+Passenger. And used more memory. Not sure why.