Surely the optimum case is to have all your machines pegged at 100% CPU
True, but given the typical difference between average load and peak load, you wouldn't want to have your systems at 100% most of the time anyway. Of course, what you really ought to do is scale down during non-peak periods and scale up when traffic increases so that your utilization is always high, but before EC2/etc. that wasn't feasible for most people to do.
Re:Flickr: PHP4 vs PHP5 drop in CPU comparison
by mbrubeck on Friday May 08, 2009 @04:06PM
(#600174)
An online retailer I worked for usually aimed for 66% peak utilization, so any one of three datacenters could fail without affecting availability.
Definitely not. If your boxes are pegged at 100%, you have no spare capacity!
The real question is, are you CPU constrained at peak capacity? And in the case of dynamic web apps, the answer is almost always yes. So CPU improvements are a huge deal. Double your CPU efficiency, double your server efficiency.
Yes, I agree - in terms of extra capacity. But for your current users, they may see no difference.
Re:Flickr: PHP4 vs PHP5 drop in CPU comparison
by zackattack on Friday May 08, 2009 @03:13PM
(#600091)
This is a very insightful comment.
What are the biggest server-side bottlenecks with respect to user experience, and what practical steps can be taken to remedy them?
CPU usage is a silly stat to measure, unless it's more than 100%. Granted, this will allow PHP5 to scale more before it gets to 100%, so that's a good thing. Surely the optimum case is to have all your machines pegged at 100% CPU, using them to their maximum capacity.
It'd be useful to compare page latency across PHP4->PHP5 or some metric the user can see.