[Novalug] Server Recommendation!
Michael Stone
mstone at mathom.us
Fri Dec 1 06:15:18 EST 2006
On Thu, Nov 30, 2006 at 11:51:13PM -0500, Peter Larsen wrote:
>Michael Stone wrote:
>>Yes. Am I going to bother to dig them up? No.
>
>Interesting. First this rant, with no backing of your conclussion, and
>later you complain to Brian that he doesn't present you numbers??
No, I didn't complain that he didn't present me numbers. Please reread
what I wrote; I made a rhetorical statement indicating that the
bandwidth used in operation doesn't necessarily relate to the bandwidth
measured in a benchmark. That's different than "got any numbers to back
that up". Sorry if that wasn't clear enough.
>I wasn't interested in a religion war here - if you are happy with your
>software based raid, please continue. I prefer to use my CPU power for
>processing and not disk io related stuff.
*That* is the "religion" that I really hate to see--the repetetion of
old theories of operation like "software raid costs a lot of CPU". And,
again, this is why I keep suggesting that people run their own
benchmarks: see for yourself whether these old chestnuts really hold up.
And, more importantly, this is why I pointed out that a person should
ask *why* they're speccing things the way they are, and then evaluate
whether a given configuration is the best way to deliver that. There are
perfectly valid reasons why you might want, e.g., hardware raid over
software raid, but it's important to make sure that you reexamine those
reasons--and the conclusions--peridically, and not just spec things out
of habit. If you're determined to not find any reason to doubt that the
config you already have is optimal, fine. But for someone asking about
what to do in the future there's still time to critically examine the
options. I will freely admit that's hard to do, with technology &
purchasing cycles as they are in the real world.
Mike Stone
More information about the Novalug
mailing list