[Biodevelopers] Re: [Bioclusters] how are the Redhat product changes affecting existing and future plans?

[Joe Landman]

Cluster OSes range from pre-packaged/pre-bundled to roll-your-own.  The 
roll-your-own crowd will continue to roll their own if they have the 
time/inclination.  Commercial and production sites tend to prefer the 
packaged ("who can we call for support") variety due to the inherent 
risks in one-off type distributions.

So I see several possible directions on the pre-packaged side:

1) Open source cluster OS distros
2) Customized "Consumer" distros
3) Commercial distros

On the open source distro side, we have ROCKS going to RHEL 3 recompiled 
without RedHat logos.  Not sure about support for this, but Glen Otero 
of Callident might be able to talk about commercial support of this.  
OSCAR is available, and should work across a number of distributions 
(RH, SUSE, etc).  I see Warewulf as another quite intriguing 
possibility.  Then there are the Clustermatic tools from Sandia and 
other similar "one-offs" to handle specific problems for their end 
users.  Not usually broadly applicable.

On the customized distro side, most folks seem likely to use SUSE/RH as 
these have the largest installed base, someone to complain to if stuff 
doesnt work.  Commercial codes are usually tested against RH and SUSE, 
so this gives you some assurances that it will work.  The customized 
route is somewhat more dangerous/daunting in that the support model is 
ill defined except for the xCAT distro.  There are few big companies 
backing the distribution method, installation method, etc.

On the commercial side, you have things like Scyld and CLIC.  Scyld is 
probably not the best for a biocluster, as it makes some assumptions on 
the interconnect.  CLIC is based upon Mandrake, and past experience with 
that distro personally drove me to other distros.  My problems had been 
with devfs, which is simply not stable enough to be used in a production 
environment.  Crashes with devfs go from merely annoying to spectacular 
in the amount of collateral damage.  Happily devfs is out (deprecated) 
as of 2.6 kernel.  Sure, some will point out that the problem that it 
solves is an important one.  Agreed.  The implementation was terrible 
though, and the damage experienced was breathtaking.

Many choices, little time.

Given their good support of the x86_64, and the seemingly higher quality 
of  beta experience, I am leaning towards the SUSE system at this time.  
Probably between RH WS and SUSE for most installs.  Customer dependent 
of course, but I don't like support suprises.  Given RH's ambivalence 
towards XFS, this will likely be biased in favor of  SUSE.  On top of 
SUSE, I will be trying OSCAR and Warewulf.

Joe

Chris Dagdigian wrote:

>
> Another item that has been on my mind recently...
>
> What are people doing about RedHat deciding to kill off their consumer 
> product line? Are people going to pay the freight for Redhat 
> Enterprise Linux or are people just going to use Suse/Debian/Gentoo etc.
>
> My needs are pretty simple but I'm having a hard time placing myself 
> into Redhat's current product plans.
>
> I need:
>
> 1. A stable OS with a product lifetime of at least 1 year (ideally 2+)
> 2. Product errata, updates and security patches for full lifespan
> 3. No OS or product phone/email support or SLA
>
> The RHL transition to Fedora Linux is fine but it sounds as if the OS 
> is going to change very fast (major updates 2-3 times per year). On 
> the plus side it is still free and the leaders seem committed to fast 
> errata and security updates. Still I can't see using this on a 
> production cluster due to the pace of change and the chance that I'd 
> be left without updates if I froze on a particular Fedora release.
>
> I can justify (maybe) the cost for the $125 product (Redhat WS) that 
> they are pitching towards compute clusters. The update services and 
> 5-year product lifespan is worth paying for. The big question for me 
> is what do I have to pay _after_ the initial $125 purchase. I can't 
> seem to find any info on the Redhat website telling me how much I'll 
> have to pay  for updates after my intial 1-year RedHat Network service 
> runs out.
>
> This also leaves the question of what RHEL flavor to run on cluster 
> head nodes, fileservers and database machines. $349 for RH ES could be 
> justified for a critical node but damn what if I want to run that 
> stuff on Opteron or Itanium or a node with 4CPUs? The cost for RH AS 
> (starting at $1400) is not justifiable to me. Putting a 'cheap' RHEL 
> flavor on a head node and manually compiling/updating/supporting 
> additional network services built by hand from source or .srpms may be 
> more of an operational headache than the cost savings justify.
>
> I'm torn right now between diving back into Gentoo/Debian or possibly 
> jumping on the Suse bandwagon given their existing support for Opteron 
> etc. Novell just bought Suse today so who knows what that is going to do.
>
> I'd be interested in knowing how current RHL users are planning the 
> transition and how future cluster buyers are changing their plans. 
> Personally I think I'm going to need to stay on top of RHEL for 
> project that demand it while also maintaining some sort of deep 
> familiarity with one or more alternatives.
>
> -Chris
>
>
>
>
>
>
> _______________________________________________
> Bioclusters maillist  -  Bioclusters at bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters


-- 
Joseph Landman, Ph.D
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
phone: +1 734 612 4615