RE: Blade Servers

From: Matthew Zito <mzito_at_gridapp.com>
Date: Fri, 9 Apr 2004 16:39:55 -0400
Message-ID: <MEEDLFOIHOCDBCOFPFHDGEDDCAAA.mzito@gridapp.com>

> Well ... OK. So I guess that means not having the functionality at =
all is
> better? Enlighten me. I'm missing something here.

Well, no, some functionality is better than no functionality, of course. =
 But there are a lot of limitations that make domaining unsuitable for a =
number of environments that Sun has been selling it into.  I should have =

made clear that my point was just was to educate about the limitations, = since Sun is certainly not taking that step with their customers. =20
>=20
> And this is a lot of space?

<snip>
> > All this for a maximum of 106 processors.
> > In comparison, you can put 168 Intel processors

>=20

> Silly me! You have me here. I forgot that it is all about=20
> processor count.

<snip>
> but everything else is crap. And I STILL get stuck in this mode that =
the
> "everything else" part matters too ... A LOT.

"I can't see how going to blades is going to save space."

That's what I was addressing. Certainly, there is no comparison, = architecturally, between a large SMP server such as an e15k and a bunch = of individual blade servers. If I ever gave that impression, I = apologize. But, if we're going to be talking about processor and server = density, not to mention heating, power, and cooling, then the e15k is = about the worst offender I've seen outside of the mainframe world.

> > From IBM, the same config lists out at $650k in blades
>=20
> I see that IBM still sells the multi-CPU servers. Are these intended =
for
> idiots who don't know any better?

Nono, there's a bunch of reasons why you might want multi-cpu servers.  =
For example, you use Cisco switches, and don't want to use the built-in =
blade server ethernet switches.  You require a particular kind of fibre =

channel card only available on standard servers. Or you need lots of = I/O connectivity - a server with 8 fibre channel connections, for = example. =20

But for the environment he's discussing - dev instances, where there's = going to be regresssion testing, application/load instances, etc., = density becomes important, as well as keeping costs down. Spending $2m = on a single server with limited flexibility, high environmental costs, = high support costs, and poor density for a development environment seems = unsupportable to me. Perhaps an organization with deep pockets might = disagree.

> > I realize that money may not be an issue, but if you're building out
> > lots of small servers, the logic of using a huge Sun server and
> > partitioning it escapes me, I'm afraid.

>=20

> I'm afraid my e-mail client malfunctioned, and the part of the=20
> original post
> that specified these were to be small servers was lost.

Well, he said blade servers, which means a maximum of 4-processor = servers - I qualify a 4-processor server as "small". My point was only = that, if you're already looking at an environment that limits you to = four processor servers, moving to an environment that doesn't really = carve up well at that size seems ill-advised, at best. =20

On the flip side, if he had said, "I need 4 20-processor servers", it = would have been equally ill-advised for me to recommend blade servers - = at that point, an e15k-style server becomes an attractive option.

Thanks very much,
Matt

Please see the official ORACLE-L FAQ: http://www.orafaq.com

To unsubscribe send email to: oracle-l-request_at_freelists.org put 'unsubscribe' in the subject line.

--
Archives are at http://www.freelists.org/archives/oracle-l/
FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------

Received on Fri Apr 09 2004 - 15:36:18 CDT