Sasha Pachev wrote:
>Patrick Greenwell wrote:
>> On Wed, 23 Jun 1999, Sasha Pachev wrote:
>> > I can understand when you get 64 processor Sun because you are
>> > processing so much data that you actually have to have 64 CPU's, but
>> > why would you get a low end Sun or HP when an equivalent Intel box will
>> > cost much less and will do the job just as well?
>> One word: Scalability. You don't always know the load over time in
>> advance and where it is going to peak. You could throw more boxes at the
>> problem, however that becomes more difficult to manage. And for management
>> purposes, having as few O/S's and boxes to deal with is generally
>> preferable to a hodgepodge.
>> And just for the record, the vast majority of my admin experience is with
>> BSDi/FreeBSD. :-)
>So I buy a low-end Sun for $5000 because I do not want do spend $100,000
>to throw it away and buy one for $100,000 when the $5000 guy cannot
>handle the load?
Something like that, but you buy the scalability. I probably wouldn't get
an Ultra-10 (for $5000) to be a server, but I would get an E-3000 with only
a couple of processors and a gig of memory to start out a project. Then, if
really do hit a processor/memory ceiling, you have the option to add 4 more
cpus and 5 gig of memory, and you're still in under the $100,000 -- and you
haven't thrown anything away.
Intel supposedly has the same capabilities, but it's scarier. Buying a 4-way
capable system with one cpu and adding more cpus later doesn't always
work. (Never, in my experience.)
Also, having an SMP system is a big win when you're using threaded
such as mysql -- a win that you lose when you split the database out over
several different intel boxes. And 'round here, network latency is a bigger
than I would have thought, so adding traffic to the <spit> token ring isn't