On Sat, 2003-09-20 at 12:04, Eric Huff wrote: > > not really... first, we must define what busy is :-) I have an > > impatient daughter standing at my elbow > > speaking of busy... > > > Processes aren't always bottlenecked by CPU, and the well behaved > > ones will let go of CPU when they're waiting on something else. > > Say that it's a database startup-- program comes up, reads config, > > makes sure that all is cool, then starts the process of opening > > disk files. For the next few jiffies, it's basically waiting on > > the disk bus and doesn't need the CPU, so it's only using 2 or 3 > > percent. > > Ok, that's what i was looking for, that it's often not the cpu the > bottlenecks. > > Thanks for the info, > eric
Eric, The number one bottleneck in things like this is actually interupt requests. This is why 2 800mhz Xeons will usually out benchmark a single 2.6 ghz processor. While one is waiting the other is working. Currently the init process waits for each "process" to completely finish. If for example you have started process foo. Foo needs to go out on the net and gather some data, bring it back and then take action accordingly (like say ntpd) currently the entire process is on hold waiting for foo to finish. The idea here is that while foo is waiting bar could be started and finished while foo is waiting. (1 second to a comp is a long time). MDK is already moving this way. If you notice X starts up even though the entire init process is not complete and in fact you can actually start logging in before the init is finished. James >
Want to buy your Pack or Services from MandrakeSoft? Go to http://www.mandrakestore.com
