[SunRescue] SS4/SS5/SS10 compatibility?

[Dave McGuire]

On April 4, Bj=F6rn Ramqvist wrote:
> I've had experience with some of these machines and from what I can
> tell, I can't agree totally with Dave McGuire on all points in this
> discussion.
> Looking at the official peak integer and floating point calculation r=
ate
> specs:
...

  While I'm sure we all appreciate the actual benchmark numbers (at
least I do) I was talking about pure real-world application
performance experience.

  About two years ago I was in a situation where my team needed to find=

a new machine to use as our field deployment platform.  The application=

was basically a satellite-based high-speed real-time streaming
protocol converter (SkyCache, now Cidera).  We were a real-world
development group with no suits, and were free to use pretty much
anything we wanted...we needed the lowest-costs solution that would
solve the problem reliably...so we started all the way at the bottom
with a stack of SS2 boxes (yes, several years old at the time, but SO
WHAT?  They still move bytes, and there were a practically unlimited
number of them available to us via surplus channels very cheaply).
The SS2 actually did just fine in our first-generation 4MBPS system,
but folded up quickly when the bit rates went up (it's now a 45MBPS
system).  We went all the way up through the SPARC family, skipping
all the Ultras because this was NetBSD-based and NetBSD/sparc64 is
USELESS, despite its presence in the distribution for a solid two
years. *grumble*

  We tried Cobalt RAQs, various Alphas, Corel (now Rebel) NetWinders.
All good machines, most too expensive for our application.  I really
liked the Cobalt RAQ.  We tried various stupid PeeCees (the management
types suggested them, how predictable) which all blew up spectacularly
under a constant heavy workload.  That's the LAST time I will try to
use a PeeCee in a production environment where my rent money is
hanging in the balance.

  While in the end, we did settle on the SS5 (a mix of 110MHz and 170MH=
z
due to there being fewer 170MHz machines available, they only shipped
for a few months before being discontinued as I recall), the SS10 with
an SM61 would have been our choice if they'd been available more
cheaply.  They performed MUCH MUCH better, but at the time they were
about twice the cost through our acquisition channels.

  This wasn't benchmarking, this wasn't a bunch of college "engineers"
standing around with ties and powerpoint presentations...this was a
group of very serious "it'll compile cleanly with -Wall or it won't
get committed to the tree" geeks in the wee hours of the morning,
turning the bit rates up until the machines smoked.

  We all learned a great deal from that experience.

  Sorry for being long-winded, but I had a little time and felt like
relating the experience in case anyone was interested.


               -Dave McGuire