Cluster Poll Results (tangent into OS choices)

Joseph Landman landman at scalableinformatics.com
Tue Nov 4 19:29:13 EST 2003


On Tue, 2003-11-04 at 19:05, Andrew M.A. Cater wrote:
> On Tue, Nov 04, 2003 at 05:50:57PM -0500, Joe Landman wrote:

[...]

> > There are interesting bits in debian.  I am not sure it is necessarily 
> > the right choice for clusters due to the specific lack of commercial 
> > support for cluster specific items such as Myrinet, and the other high 
> > speed interconnects.  
> 
> Dan - if I build a _really big_ cluster, will you get Quadrics to do 
> Debian :) 

I think the question is, if you buy $10M in interconnects from them,
would they please port to distro X.  Likely it would be worth their
while in that case.  

Are you going to build such a big cluster?  :)

> Same goes for any other vendor - if you ask them nicely and make it 
> worth their while, they'll do it.  In many cases, it's only a recompile 
> of a device driver to account for library differences, after all.

Not always.  The issue is not simply a port, but also the support
costs.  Support in the sense of qualifying the port against a standard
load.  Coming up with the standard load, building the regression tests,
educating the staff on the new support ...

They may simply make the port and say, good luck, you are on your own.

> HP use Debian internally, IIRC.  

Depends upon who you ask.  Bruce Perens had some effect there, but as I
remember, they use SUSE, RedHat, ROCKS, etc.

> Some of the Debian developers are also
> HP folk - HP are potentially looking to support more of their products 
> under Linux? [See, for example, Debian Weekly News for today :) ]

Following on others foray into this, I am going to take a pragmatic
position.  I will believe it when I see it (.deb's from HP and others).

> > Commercial compiler support for Debian (e.g. 
> > Intel, Absoft, et al) is largely non-existant as far as I know (please 
> > do correct me if I am wrong).  
> 
> Compaq Alpha compilers work on the Alpha port or can be tweaked to IIRC.
> I have no current expertise on big commercial compilers, however.

:)

I seem to remember HP recently EOLing the Alpha in favor of some other
chip... can't remember its name ... ;-)

I can run Debian on my SGI Indy.  I am not, but I can.  Doesn't mean
much as the market for Indy's has basically dried up.

> > Few if any commercial applications are certified to work on Debian 
> > (Oracle, Legato, ....) and again, please correct me if I am wrong.
> > 
> 
> Many of these will run fine without formal certification from the 
> vendor.

Ok.  Now sell that to a CIO/CTO, or someone responsible for making the
infrastructure work.  Mike at Linux Networx (though speaking for
himself) called it the "smile test" or something like that.  The
question you will be asked is, if something breaks in our critical
business application, who are we going to call if we are using the
un-certified OS distribution?   

This is a hard sell.  

> Few, if any, current commercial apps run on Red Hat 4.2 / 5.0 - and 
> current Red Hat 7.x/8.x/9.x is now as commercially relevant.  

I respectfully disagree with the last portion of the statement.  Most of
the engineering code that I have played with recently spec out RH7.x as
their linux supported platform.  Anything else and you are on your own. 
The bio and chem codes which come pre-compiled tend to have a
"requirements" section as well, listing RH7.x.  Remember that RHAS2.1
will be supported a few more years, and it is ostensibly RH7.x. 

> The big 
> commercial apps will have to retrench their markets, potentially, to 
> (one/both) of Novell / RH Enterprise Linux at ??$ per licence.  Unless 
> it says RH/Novell on the box, they won't certify on something "less but 
> Libre" based on RH.  But this is Linux - a commercial Linux app. will run 
> on other distributions with a little thought / planning.  I'm not sure
> they'll run Oracle on Scyld / ROCKS, for example.

Some distros are more (for lack of a better term) engineered than
others.  There are some code issues which some of these which break the
"defacto" standard Linux.  As for Oracle on ROCKS, well, Oracle does run
in a supported mode on RHAS2.1 (see above), and ROCKS == RH7.3, so the
rest is left to the reader.  As the underlying OS is RedHat, with a meta
layer atop it called ROCKS, Oracle should not see any reason not to work
under this environment in a supported manner.  That said, I am not sure
that is what you want to do with Oracle though.

[...]

> 
> > ROCKS, BioBrew (and I understand Warewulf) make 
> > this ridiculously easy.  Increasing the setup/management time, or making 
> > your life harder in general, doesn't make much sense.   There is a 
> > Knoppix variant that does clustering (OpenMosix style).  Not sure it is 
> > the best solution, but I would like to hear from anyone using it.
> > 
> 
> This is fun if you want an ad-hoc StoneSouperComputer - the 512 node
> machine built in a night on a German TV show or the four node proof
> of concept idea for a show and tell in someone's office - but I'm
> not entirely sure I'd trust my most valuable data to it. But hey, like
> most things KNOPPIX based its an ultra cool demo :)

I think the irony in all of this is that the one disk I carry with me
everywhere is a Knoppix disk (Debian based).  I like it, it is
technically neat.  

> 
> Have fun - at 0015 or so Zulu time, I'd better get some rest :)
> 
> Andy
-- 
Joseph Landman, Ph.D
Scalable Informatics LLC
email: landman at scalableinformatics.com
  web: http://scalableinformatics.com
phone: +1 734 612 4615


_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list