[Beowulf] Rackable / SGI

Kilian CAVALOTTI kilian.cavalotti.work at gmail.com
Thu Apr 2 03:21:50 PDT 2009

On Thursday 02 April 2009 11:13:28 Jan Heichler wrote:
> KC> My bad here for not being explicit enough. By "their own hardware", I
> meant KC> "hardware they put a lot of thought in, that they carefully
> conceived and KC> designed, and have manufactured according to their own
> needs and guidelines",
> Let me ask: why is that important?

Maybe not important, but at least different.

Although it may be important anyhow, since it probably means a bunch of people 
got together to think about how to design a hardware platform that would be 
better than the existing ones. Engineering is supposed to drive innovation, 
and I guess it's a healthy practice that smaller companies can play in that 
game too, Big names shouldn't be the only ones to decide how a server should 
be made. A lot of good ideas come from smaller teams, and sometimes large 
companies are too constrained to dare thinking outside the box (although it's 
not a general rule either, as IBM demonstrated with their iDataPlex).
Anyway, diversity and innovation are good, and the customer should benefit 
from it at the end, right?

> What is driving HPC is the commodity approach. Use what is used a
> million-times all over the world. Where is the need to engineer something
> different?

True enough, but the downside of the commodity approach is that it tends to 
reduce the apparition of new ideas. See how long it took IBM to realize that 
cooling a 1U box could be done more efficiently by rotating the airflow 90°? 
Everybody was doing it along the long side, so everybody continued to it that 

> And if a small company does something new you won't sell a huge quantity.
> So the components get (usually) more expensive. The engineering isn't quite
> as advanced (because it is expensive) and many errors just show when you
> have a larger install-base.

That's right. But maybe new ideas developed by smaller companies can be picked 
up by larger ones, to pursue the engineering and improve the idea. Ok, I only 
see the customer's interest here, but hell, I'm a customer. :)

> Mhh... a Relion server does not look very different from Intel-Servers (i
> mean the intel-oem products). A Jackrabbit seems to be a standard
> Chenbro/ICP/Whatever chassis with commodity Boards/Controllers/Disks. The
> real trick is probably the software.

I don't know the details, but the very people who know are reading this list. 
They'll probably have more relevant things to say on that matter than anything 
I could.

> That is true (for ClusterVision). We (i'm with ClusterVision) are investing
> in our Software - because there we see a lot of potential to do better than
> what you see on the market. Hardware is not that different - except for the
> brand-name.

So Clustervision does with software what I was describing for hardware, and 
that's good, because it adds diversity to the landscape, it makes my options 
more interesting as a buyer, and it drives innovation. I totally understand 
your point that engineering hardware is much more expensive than engineering 
software, and that it can't be profitable unless you're already big enough to 
assume the investment costs.

> I don't think it is that different.

Maybe not, but different enough that we don't see companies in the EU selling 
their own hardware. :)

> Even a company like SGI doesn't seem big enough to really have own hardware
> (and live with what you earn for it on the market). And even if ICE is a
> nice system - it is not that much different from what other BladeSystems
> are doing. So why should a customer bother?

Not having seen one first-hand, I'm probably not the best person to describe 
the upsides of the ICE platform, but from what I've seen, the cable-free, DC 
current enclosures, the diskless blades, the dual Infiniband planes and the 
water cooling doors are somehow unique. Those days, improvements are mainly 
seen on the energy efficiency front. But customer who cares about her annual 
electricity bill should surely care. shouldn't she?

> Maybe i'm too young to really see SGI going down as a big loss - in my time
> they were just another Intel-Selling company with a strange attitude and
> really expensive products. But i never understood the excitement about Sun
> either ;-)

A shrinking market is never a good sign, and I think that both SGI and Sun 
contributed so many ideas, products or software in our IT world that it's 
became part of our culture, consciously or not. :)


More information about the Beowulf mailing list