[rescue] SM100 CPUs (Re: SNMP BABY!)

[Bjorn Ramqvist]

"Loomis, Rip" wrote:
> 
> If I'm running a production system today, I want at least Solaris
> 2.5.1 (or SparcLinux, or *BSD/SPARC).  None of those support the
> SM100 in SMP mode--from the initial testing I did, Solaris 2.5.1
> was the last rev to support SM100 at all, and disabled all but
> one CPU on the module (no SMP).  Also, the fact that a CPU on
> an SM100 runs at 40MHz does not mean it's the same CPU as in an
> SS2--from my testing it was significantly "slower" in actually
> doing real work.
> 
> I'll be happy to listen to anyone who's done exhaustive recent
> performance testing of an SM100 drone on...but I stand by my
> statement about throwing away SM100s.  There are just some
> pieces of hardware that are neither elegant nor special enough
> to be retained once they're past their useful life.  (SM20
> MBus modules are in the same category...)

OK, first off, I'm not trying to be a hardass about something so small
and irrelevant as an SM100 module, but just for the debate beeing, I'll
try to squeeze some enjoyment out of this. And for the record, don't
think I am one of those ex-pee kiddies that thinks Gigahertz counts in
every situation.
I have had a 4/600-140 running for quite some time a couple of years
ago, along with Solaris 2.4, and yes, with MP support. The earlier
revisions (pre -08) was incapable of this, but luckily we had the newer
ones.

First off, let's look at the SPARCstation 2. According to the Sun-Ref it
is a CY7C601 @ 40MHz inside there, doing some hefty 517 SPECintRate92
and 541 SPECfpRate92. Compare that with the (almost) perfect Mbus-guide
(http://mbus.sunhelp.org/modules/) which states SM100 to be two CY7C605
@ 40MHz. From what I know, about the only differences between the 601
and the 605, is the addition of MP-support.
I'm not sure, but that's what I've been told and I could very likely be
wrong about this if someone else proves otherwise.

Looking at the SM100 module specs, a Sun 4/600-140 (2 x SM100) there was
a flash in Septemper 1992 SPEC news stating 1847 SPECintRate92 and 1930
SPECfpRate92. Comparing that with the SS2, that'd leave us with 3.5
times the integer performance and 3.5 times the floating point
performance ratings. That's all off the original spec-benchmarks.
Ofcourse there are countless of other factors involving the overall
system performance that may have forced your system to go sluggish.
Memory-design, buses, disks, network, operating system (and parameters).
Heck, the design of the whole system differs quite alot between the
4/600 and the 4/75.

Don't bash that module in the very first place. Infact, this was the
very first CPU offered on the 4/600, and it did perform our tasks quite
well, beeing so low-spec comparing to the later (and otherwise totally
different) Mbus-modules.
But trust me, I'd never ever waste a 4/600 chassie or a SS10 to run one
of those CPU modules in production.

/Bjorn