[rescue] ID this SGI board and win, oh, I dunno, my eternal thanks

[Patrick Giagnocavo +1.717.201.3366]

On Fri, Feb 06, 2004 at 12:01:25AM -0500, Joshua Boyd wrote:
> > I don't know how to evaluate that, actually.  From the performance
> > comparisons at http://futuretech.mirror.vuurwerk.net/perfcomp.html, I get
> > the sense that a single R10k-195 is 2-3 times as fast as an R4k-200 in an
> > Indigo2, but he doesn't have much in the multi-processor area or that's
> > Challenge/Onyx-specific.  And, of course, we're talking R4k-150 in my
> > Challenge, not -200.
> 
> Well, my understanding is that a single MC3 board is capable of fully
> feeding 2 to 4 CPUs of either type, so for tasks that allow it, you
> would likely see a linear speed up in all cases.  I strongly think that

MC3 is RAM card, no?  How much interleaving do you get on that?
Challenge-R10k weakness vs. newer systems is memory bandwidth.

> On the other hand, 4 procs is cooler, even if it isn't faster.  Perhaps

I think 4 procs will be warmer :-) but I know what you mean.

> > However, I don't know how much "general computing" stuff uses more than one
> > processor, or, indeed, how that would affect things at all.
> 
> "General computing" likely won't take advantage of multiple CPUs so
> well.  However, I thought this machine was for doing non-general
> computing?  BTW, database software should scale well, but I don't know
> if Postgres or MySQL actually do scale well in this manner or not.

Postgres will have an advantage, I think, since each connection is a
new, separate process.  PG uses shared memory to communicate with the
backend server.

> Shell scripting will also take advantage of multiple CPUs, and with
> care, you can maximize this to good advantage for certain types of
> processing tasks. 

Yes, 4 CPUs would be nice for that.  Though I would go with the dual
R10ks myself.

--Patrick