[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] Re: xen benchmark results on nehalem

Peter Booth <peter_booth@xxxxxx> writes:
> If, however, you have a slow application (which is sadly very common
> even today) and you want to optimize response time then straight line
> CPU power might be the variable you need to tweak. Many people don't
> realize that a 2GHz core two duo can be faster than a 3.8 GHz P4 or
> that one 2GHz core two duo can be 25% faster than a 2.33 GHz core two
> duo.

my experience has been that often people try to upgrade the CPU while the 
application is still touching disk to read data.  This is obviously
silly, because instead of being 25% faster or what have you, disk is millions
of times slower than ram, so you get this 'fall off a cliff' effect whenever
you hit disk.

Sure, once your dataset is cached to ram, then yeah, you are right, 
upgrading the cpu and/or ram speed makes a lot of sense.   And there are
some workloads that work like that;  from what I understand most HPC
applications are designed with the knowledge that disk is so many orders
of magnitude slower than ram, so it doesn't hit disk much, and you get a 
real boost out of upgrading cpu or ram speeds.

But most web applications (and *NIX applications in general)  are still
hitting disk, so it makes sense to have enough pagecache for all your 
commonly accessed data before you start worrying about cpu speed.  
As long as you are still falling off the cliff and hitting disk, a cpu
that is 2x or 3x faster isn't going to make a bit of difference.  

Xen-users mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.