compiling cores > clockspeed ?

bah , all these solutions are lame :-), compiling 3.0rc2 latest kernel runs instantly on distcc., 318 node (mixture of P4, dual cores and quad cores) cluster, @ work of course! ... it would be a sad day when an individual owns this many nodes.

On a more pragmatic note... what tntcoder said, the more cores you have + multi-threading, the better.

@tntcoder, why don't you do your hash crunching on the GPU? :-)

I don't quite agree about using the -j flag wildly. its usually the number of cores + 1, at least that's what I've been taught by the school of gentoo.
 
I don't quite agree about using the -j flag wildly. its usually the number of cores + 1, at least that's what I've been taught by the school of gentoo.

I agree, my understanding is that if (j > core count + 0 or 1) then you will essentially saturate the thread scheduler and achieve a negative performance impact (unnecessary context switching etc).
 
I agree, my understanding is that if (j > core count + 0 or 1) then you will essentially saturate the thread scheduler and achieve a negative performance impact (unnecessary context switching etc).

Thats old school. I build distros /all day/ and the scheduler has evolved massively over the years.

You, don't "saturate" the thread scheduler, tasks that are CPU bound run their time slice, and their priority goes down a notch; tasks that are IO bound have a higher priority because they act "nice" and don't use their time slice.
In fact, CPU bound tasks are pretty "cheap" on the scheduler because they need to be preempted after a longer time than say, concurrent IO tasks that do a lot more context switches and therefore need more maintenance.

Therefore by having more tasks than CPUs, you allow the "spare" tasks that might be IO bound to run quicker, do in a compile job you have all the loading, lookup, and writing bit of the job that can happend while the CPU are used for other CPU bound tasks.

Michael, ARM Samsung s3c2440 kernel maintainer :D
 
Very interesting , Thanks guys

May be a lot of c++ to compile, say like web-kit ,

With a budget of £200 for a cpu what would you go for ?
 
2.6.38 has the nice scheduler improvements that are probably a worthy upgrade IMO.

I'm tempted to try out the Xen integration in 3.0, but other than that I don't think it's anything special, Linus just went to 3.0 "because he could" :p

I'm also interested to see the Xen improvements.

Michael, do you maintain the kernel in a professional environment or as a hobbyist?
 
I take it you do development and want to compile the kernel often then test your code. Anything else other than using kernel compilation as a benchmark and I have problems with your motives!
 
Back
Top Bottom