Re: Are Sunblade 1000s slow?
On Fri, 3 Oct 2003 16:34:18 +0200
"Matthew French" <mfrench@telkomsa.net> wrote:
> More seriously, Sun's server prices have become far more competitive, and
> now that the prices are similar I would much rather have an 8 processor V880
> than an 8 processor Intel box. Same applies to servers that need more than
> 4Gb of RAM.
I'd get an Opteron if I needed 4GB+ of ram.
> I do think that a lot of the "coolness" factor of Sun systems is related to
> their mystique and their relative rarity. Very much like Apple hardware. I
> support Sun because it I fear a hardware monoculture as much as I worry
> about the existing Windows software monoculture. And I do believe their
> hardware performs better in the data center, but this is normally not very
> exciting stuff. (Solaris is also not a bad operating system. Although it
> does not make a very nice _desktop_ OS.)
I in fact like Apple hardware, and I think Apple is viable because they
make a computer that is very fun to use. They give something of significant
value to a large group of people. A very small (and ever shrinking) group
of people see this when they look at what Sun is offering.
Let me put it this way, if you think 4-way is the limit of scaling for
an x86 box what do you need a many-cpu system for? If it's database work
get a bunch of 4-cpu x86 boxes and setup an oracle cluster or something
similar. Clusting allows a ton of big end problems to be solved and
nothing beats the cost effectiveness of a rack of tiny x86 blades.
If you still absolutely need so many cpus, fine go get the box from
Sun and be their support headache. As their market share shrinks,
it's you the customer that is going to have to help Sun defray that
situation by paying more and more to get their systems and services.
And meanwhile, the gap between Sun's kit and what the x86 folks have
to offer on the high end steadily shrinks. Why do you think all of
these folks are working on NUMA stuff? Where do you think Opteron
kind of stuff is going?
> > That's the facts of the market today, everthing is commoditized
> > to the tilt.
>
> Although commodity components also suffer from the lowest-common-denominator
> effect. The performance of Intel servers suffer because it is much harder
> for the vendors to build tightly integrated backplanes and systems.
I don't agree with this either. Why did we have that "race to the moon"
the other year between AMD and Intel to get to 1GHZ? It's because the
volumes pay for all the R&D much better than Sun's model does.
Who has the best silicon etching and processor design technology out
there? It's not Sun and Texas Instruments, I can assure you of that :)
> > the only reason I have so many boxes is that I've gotten
> > them all for free during my years of development.
>
> Aw, can I have some. Please?!? <tries best doe eyed look>
Sorry :)
> And you can get good graphics cards for very little extra on the Athlon.
Yes, that's true too.
> So to Kent I would suggest rather set up a comparison using something
> mathematically intensive like Matlab - and then ask five people to run
> simulations on the same system at the same time. This should give you a much
> better basis for comparison.
Well, not against Linux/Sparc since there's no version of Matlab
you can get for that.
Reply to: