[rescue] SMP on intel wasteful?

James Lockwood james at foonly.com
Mon Jun 24 15:52:02 CDT 2002


On Mon, 24 Jun 2002, Dave McGuire wrote:

> On June 24, Chris Hedemark wrote:
> > I've always found those nebulous arguments against PeeCee hardware
> > interesting.  Usually unfounded in recent fact, but interesting
> > nonetheless.  When put to the test though I've found that dollar for
> > dollar a PeeCee running UNIX will spank a RISC box running UNIX.
>
>   *bzzzt*  Troll alert.
>   *bzzzt*  Don't-know-what-CPAI-is alert.
>   *bzzzt*  Possible remote Intel Marketing Clone alert.

Sticking your fingers in your ears and saying "la la la" won't make x86 go
away.  Believe me, I've tried.  Sorry if the following is grammatically
poor, I'm just rambling.

Bottom line, x86 is fast.  It is as ugly as sin and has some nasty
compromises in it, but it is fast.  The reason for this is dead simple,
it's a commodity architecture that discourages stratification within it.
Let me relate a little story that I feel has some relevance to this
discussion.

A number of years back, the mainframe world represented computing to a
near complete extent.  Over time, there was a significant shift away from
'frames down to minis, despite the fact that reliability, speed and
features were decreased.  Price/performance was significantly better, and
the systems were perceived as "good enough".  Mainframe companies suffered
heavily.

After that, there was another shift from the minis down to the Unix
workstations and servers from Unix-only vendors such as Sun.  These
typically offered less in the way of CPU speed, I/O and reliability (which
had gone up in the minis over the years) but had significantly better
price/performance, and they were "good enough".  Mini vendors suffered
until they could compete on equal footing.  Few did.

We are seeing a logical continuation of this trend.  The Unix boxes of
today have made significant strides, but the PCs are catching up and have
already surpassed many other systems in price/performance _for certain
classes of problems_.  To deny this is to close your eyes to an entire
class of useful tools to solve problems.

Computers are TOOLS.  They are devices to solve problems, and you must
take care in choosing the proper tool for your problem.  Blind advocacy
does no good whatsoever to those of us who are willing to understand
and use every available tool in our arsenal to solve a problem.

Do I like Suns?  Yes.  Do I prefer them for quantifiable reasons?  Yes,
I thoroughly enjoy the tight hw/sw integration, the OBP, the quality of
support and the quality of construction.  Are any of these so important to
me that they preclude ever using anything else?  Certainly not!

Chris raises valid points.  I personally have been soured by the taste of
Linux every time I have tried it, but I can certainly see the utility of
it for many tasks.  I have been burned by x86 many, many times, yet I
admit that its time is already here for applications that can use cheap
boxes with a lot of CPU power.

Reliability of x86 vs non x86 is a strawman.  _Everything_ fails
sometimes.  The only way to avoid this is to design your system so that
one (or more than one) failure does not take you out.

Rant on x86:

Low end x86 gear cuts corners on I/O and reliability.
High end x86 h/w is hard to get/justify because there is a lack of market
stratification.
Low end x86 is cheap because it is commoditized.
High end x86 is not.
Therefore, the advantages that exist for some problems (great price/cpu)
with low end x86 are not as compelling for the purchase of high end x86.

Hence: high end x86 does not receive the bulk of actual development.  This
explains the lack of big SMP x86 boxes, as well as the slow progress of
other things needed on the server side (such as fast I/O).

IMHO this will happen to _any_ dominant architecture.  Should Sun
magically succeed in trumping Intel at the low end desktop level, you will
see similar parity in terms of quality and price.  The low end drives the
market whenever it can because of volume.

I enjoy debate.  I do not enjoy advocacy.  If I wanted to hear advocacy, I
could go to Linuxworld wearing my "Microsoft: #1 in UNIX" t-shirt (yes,
they were the largest Unix installed base at one point in time).

-James



More information about the rescue mailing list