mbecker ™ wrote:
Just a simple example of P4 vs. Athlon XP. I chose
older chips simply
because they are cheap and the benchmarks were easy to find. If I knew
which P4 was being considered, I could provide more relevant
benchmarks.
Unfortunately I can't give you a particular test system example because
I don't know exactly what's on the table (though that ought to get
mentioned eventually, if someone actually has been working on an order).
[snip generic benchmarks]
I'd be more interested in seeing some MediaWiki rendering and Lucene
search benchmarks, which could represent our actual workload.
Price/performance shouldn't be calculated for the CPU alone, either;
we're buying whole machines, not bare CPUs.
Not all benchmarks favor the same candidate, and it looks like
differences are at most going to be a couple percentage points when
comparing whole system price.
Note that my own personal prejudice runs in favor of AMD; I buy Athlons
for my home PCs and hiss and boo Intel at every opportunity. But if
we're going to make purchasing decisions explicitly based on a
price/performance claim I think we need something more than warm fuzzy
feelings of fighting The Man.
We started out getting Opteron boxes for the databases specifically in
order to have something that would deal well with a big chunk of memory,
whereas the commodity P4-based 1U servers were the generic case,
considered to be reliable and reasonably cost-effective based on
(anecdotal) experience.
-- brion vibber (brion @
pobox.com)