|--05---10---15---20---25---30---35---40---45---50---55---60---65---70---75---8
0
>> 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.)
>>
>Is it really scarier? How so? We're having the same arguments right now
>as we struggle to choose to pay more for UltraSPARC when with the same
>money we could purchase more x86. Using the $5k example, you could
>purchase an Ultra10 class SPARC or $6k for an UltraAXi class server w/ 1GB
>RAM and a few SCSI disks whereas with the same money, you could buy two
>2xPIII-450 servers with the same RAM and disk and do physical/logical
>clustering or just have a hot spare.
>
>In this example, why would you still go SPARC?
What I meant is that I can take a SPARC box, say an E3000, buy it with
only two CPUs, and upgrade it to 6 CPUs if I needed it later. I wouldn't
feel comfortable buying an Intel 4-way box with only 2 CPUs and buying
two more CPUs later when I found I needed the horsepower. That almost
never works -- the n-way Intel boxes seem to want to have all n of the
CPUs from the same production batch.
I'm big on Intel (okay, actually big on AMD), so I WOULD buy the two
2xK7 systems you mention, if the situation warranted it. If I thought I
was going to be pushing the load up past about 4 on the boxes*, I'd go
with a SPARC, 'cause I can easily upgrade to 16 (or 64!) processors
out of one box. Yes, I have worked on systems that really did need
that kind of horsepower for the database alone (although they were
all running Oracle, so who knows how much of that was eaten up by
Oracle Bloat.)
To restate: I would buy the 2 Intel/AMD/Cyrix boxes for most projects.
However, if I needed a system that had a much higher ceiling when it
came to scalability, SPARC systems still win, hands down.
* I try to keep the sustained load at less than 2x(number of cpus).
--
derick
