Sean, as for the setting of "max simultaneous requests", and whether it's connected to number of CPUs, I'd say don't stress about it. I'm firmly convinced that that's an old notion (based on an old Adobe technote) that people just keep parroting. Sure, when CF was interpreting your CFML as pcode, it was CPU-intensive. These days, CFML runs as Java classes, and more important, most long-running CF requests spends far more time waiting for other things (database queries, cfhttp calls, web service calls, etc.) which burn no CPU while waiting.
So really, people can likely run more threads than that traditional recommendation. How many? No one can tell you. If you don't have enough, then you may get enough load that you're using all of them all that time, which leads to queuing. The use of threads (and queuing) can be monitored via CFSTAT (except on the Multiserver edition), or JRun metrics, or the CF Server monitor, or tools like FusionReactor and SeeFusion, to name a few.
If you set too many, you may stress your box (asking it to do too much), or you may stress a connected box (like a database server, which may not be up to you pushing many more requests at once).
It's a balancing act, which you can only set by watching measures like the above, while increasing the number, whether in load testing or in production (for those that feel they have no choice).
Hope that helps.
/charlie
/Charlie (troubleshooter, carehart. org)
29
Replies
AdChoices