[rescue] SMP on intel wasteful?

[Dave McGuire]

On June 24, Chris Hedemark wrote:
> >   *bzzzt*  Moron alert.
> 
> More name calling.  That says a lot about your character (or lack
> thereof)

  Oh of course.

> Before I suspected, but now I am quite sure, I'm dealing with a common
> BS artist that is either (a) regurgitating the same tripe they heard

  What's that about name calling?

> from someone else who was posting the same nebulous scat but in an
> authoritative and confident tone or (b) has convinced their employer to
> invest so much $$$ in said hardware that it would look very bad for them
> to ever concede that cheaper hardware could be better.

  I don't convince employers to spend money.  I've spent the past twelve
years of my life STARTING companies...gratuitous spending doesn't fit
within that environment very well if you're trying to do it right.
However, ease of maintenance, lack of downtime, and uncompromising
performance are of paramount importance...something you'd know if you
weren't an Intel-cpu-humping list troller. ;)

> >   Exotic hardware?  No.  Sure, the Crays might be considered "exotic",
> > I'll give you that.  But most of the machines here are SPARCs.  The
> > SPARC architecture is a modern, scalable IEEE standard.  The x86
> > architecture is an old, non-scalable, proprietary architecture dating
> > back to the 8-bit era that's completely incompatible with everything
> > else.  Now...which is "exotic", again?
> 
> How is the x86 not scalable?
>
> I've got an 8MHz x86 box here, and an 800MHz x86 box here (the one I
> check mail from in fact).  Then there are the dual 2GHz boxen.  There is
> no place to go but up.

  Wow, you really don't know much about architecture, do you.  Ok,
here's another bone.  Like it or not, a primary driving force (second
only to "mega hurts") behind the evolution of Intel's x86-architecture
implementations is backward compatibility with the 8086 at the object
code level.  The 8086 was designed as a 16-bit extension to the wildly
successful 8-bit 8080/8085 architecture (not object-level compatible
but compatible in most other respects), a design which is still
reflected in the flashiest P4 today.  The same basic register layout
(with the addition of the segment registers and a few others) remains,
forever tying the x86 architecture to the mid-1970s when it was
designed.  Changing these aspects of the design would be a change to
the *architecture*.  While on more modern designs, like SPARC for
example, adding more registers or changing the word width is a change
to the implementation, not the architecture.

  To summarize: To build it any better would result in an incompatible
implementation.  This is the very definition of "not scalable".

...
> reach for the big UltraSPARC boxen.  No argument from me there.  When I
> need reliability I reach for the smaller UltraSPARC boxen (sometimes,
> though increasingly the Compaq Proliants have been proving their stuff
> in the field).

  So Compaq Proliants are cheaper than server-class computers?  Last I
checked (which admittedly was some time ago) some of those boxes cost
many thousands of dollars.

>  But when it comes down to just getting the job done, as
> expediently and as efficiently as possible, my money is often on the
> X86.

  I have NO doubt of that whatsoever.  That doesn't, however, make it
a good decision. ;)

> I've never done beowulf so I won't speak for that.
> Mosix is more along the lines of what I've worked with.

  It's still parallel vs. vector...your experience there counts.

> 99% of data centers don't need Crays.  This is evidenced by their
> increasing difficulty to find a market for their boxen.

  And 99% of them don't need Beowulf (or Mosix) clusters either.  They
just need high-speed, high-bandwidth servers.  SER-VERS.  Not great
big PeeCees with KVM switches.

> Real world, why on God's green earth would you want to spend $100,000 on
> a piece of kit when the job can get done just as well (or better) on
> $2,000 to $10,000 worth of kit?  Why can't anyone answer that simple
> question?

  I've NEVER spent $100K on a piece of kit.  Ever.  Even when I was
running around blowing $10K at a time on a whim (in a previous life as
a financially well-to-do person).

  Have a very lovely day.

        -Dave

-- 
Dave McGuire                  "Needing a calculator indicates that
St. Petersburg, FL              your .emacs file is incomplete." -Joshua Boyd