[Beowulf] Clusters and Distro Lifespans

[Gary Stiehr]

Gerald Davies wrote:

> I'm new to posting, however, i do read the list :)
>
> John's comment about the short support lifetime of FC raises one of my
> concerns about distros and cluster set-ups in general.  In my
> department we have RH/FC based clusters.  When purchased they came
> with a pre-installed distro and have pxeboot/images.  I then spent
> time tuning them to our needs.
>
> My questions relating to this are:
>
> i)  Is the practice of buying clusters with pre-installed distros 
> popular?
>
> ii) Would it be better to develop our own installation process for
> clusters so that upgrades, in terms of distros, can be rolled out
> easily?  I feel like i'm tied in some way to the supplier of our
> cluster for upgrades.


Hi Gerald:

I do think you should have your own installation process.  However, this 
should be a general process based off of your requirements (e.g., you 
want to install over a network and have nightly security updates).  Of 
course, you will benefit if the process is implemented using tools that 
are not distribution/vendor-specific since you will then be free to (or 
more likely to be able to) use that infrastructure in the future without 
being tied to any specific supplier.  A common example is using 
PXE/tftp/dhcp for installs and yum for updates.  Ideally, you would find 
a vendor who understands the value of that approach and will provide 
that infrastructure as an added value to the customer.  Otherwise, if 
you are able (time-wise), it certainly is a good learning experience to 
set these things up (if you don't already know how).


>
> iii) Do people regularly upgrade their clusters in relation to
> distros?  I guess this is like asking how long is a piece of string
> because everyone's needs are different.


Of course we have our nightly security/errata updates on all of our 
nodes (using yum).  As for actually upgrading our distribution (we use 
Scientific Linux), we do not mind updating minor revisions since this 
can actually be done from yum.  When  a major new version comes out, it 
is sometimes the case that there is no "upgrade" path so we must 
reinstall systems completely.  However, before we do that, all of the 
physics software that our users rely upon (and there is a lot of it) 
undergoes a certification process to be sure it works on the latest 
version (this is a major event).  Critical software on our storage 
clusters undergoes a similar process.  As you can see our upgrade 
timelines are governed mostly by security and compatibility with our 
critical applications.  Typically, when we buy new compute nodes, we run 
our latest certified distribution on it to check for hardware 
compatibility before we make a purchase; therefore, our distro upgrade 
timeline is not typically tied to kernel hardware support.

Hope this was helpful,

Gary



>
> Apologies if this sounds like a strange first post :)
>
> Cheers,
>
> Gerald
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit 
> http://www.beowulf.org/mailman/listinfo/beowulf
>