🚀 go-pugleaf

Christopher Browne wrote:

> Quoth George Bell <MasterofTimeADim@netscape.net>:
> 
>>While I generally agree that dual CPU may be inappropriate, this is
>>an inaccurate statement.  The allocation of CPU resources is a
>>function of the os, not the tasks running(and thus the software).
>>Since Linux can support SMP - symetric multiprocessor( each
>>processor gets as much "work" as the other ), then twice as much CPU
>>time is available to the system and the performance is twice as
>>great, period.
>>
> 
> This isn't a fair assessment of things.  
> 
> Having a second CPU /may/ allow the kernel to push tasks onto one CPU
> or another, sharing the load.  If you have a whole bunch of processes
> that /can/ run in parallel, that can be useful.
> 
> On the other hand, the two CPUs have to share:
>  a) The same memory bus;
>  b) The same memory caches;
>  c) The same I/O buses.
> 
> If the CPUs are working on totally different sorts of things, b) can
> lead in particular to "sucky performance."
> 
> Consider: on a single CPU machine, any given task is going to get a
> bunch of cycles to work on whatever is the work of the task.  When a
> context switch occurs, different activity will lead to different cache
> usage, but at least you get locality of access within the process
> context.
> 
> On a dual CPU machine, two processes really are running concurrently,
> and if their accesses to memory are really quite different, they'll be
> stumbling over each other as they concurrently try to access very
> different things in memory.
> 
> Both CPUs have to coordinate their access to memory lest they get out
> of sync and start trampling on each others' memory.  That coordination
> effort has a cost, and it is a cost that a single CPU machine does not
> need to pay.
> 
> If both CPUs are /reading/ the same code, or /reading/ the same memory
> locations, there could conceivably be a big /WIN/ as a result; the
> first CPU already read the data, so it's in cache so that the second
> CPU doesn't have to read it in.  But that is not the typical case; it
> is more likely that you'll have XFree86 on one CPU, doing its thing,
> and Mozilla on the other, doing something different, so that
> maintaining cache coherency is a net /cost/.
> 

I'm actually on your side, but for different reasons.  I doubt a dual 
processor machine brings much advantage *unless* you are in a situation 
where you need lots of processing power - and some of the posters bear 
out with their testimonies the superior performance of dual processor 
systems.  I think dual processor system gives a noticable difference if 
you a) have a lot of users logged on at once b) run a busy webserver 
that has to spawn a whole lot of connections at once c) you tend to run 
lots of applications at once.  At what point this advantage becomes 
apparent - I'm not sure....

The problems you mention I think are pretty well addressed by dedicated 
circuitry in the cpu when it is in SMP mode and advanced algorithms.  I 
am no expert on this so perhaps someone else can contribute to this 
thread, but this is why I have found that some CPUs can operate in SMP 
while others cannot.  Until up to a few years ago I think AMD didn't 
even make processors that could work in SMP - only Intel.

George