[BioBrew Users] separate biobrew install?

[Glen Otero]

On Thursday, August 28, 2003, at 11:25  PM, Bill Barnard wrote:

> On Thu, 2003-08-28 at 13:01, Glen Otero wrote:
>> I actually intend to create other 'Brews' for computational chemistry,
>> climate modeling, etc. A domain agnostic package structure makes sense
>> for distribution of applications, and the Rocks folks are 
>> experimenting
>> with this idea in their upcoming 3.0 release. So I may adopt that
>> structure in the future. But for right now, since the bio packages 
>> that
>> are included in BioBrew are just rpms that I've built, it's trivial
>> (for me) to integrate them with a Rocks release to create BioBrew. I
>> save people a lot of time and frustration by doing this. Installing
>> BioBrew saves one the trouble of installing Rocks, and then 
>> downloading
>> all the individual bio rpms, building and installing them on the
>> Frontend, and then installing them on the compute nodes. I understand
>> that if the cluster were to be used for more than just bio, like bio
>> and physics, that doing a regular Rocks install and then grabbing the
>> desired bio and physics packages would make the cluster a little more
>> customizable. I have been kicking around the idea of making the
>> individual bio SRPMS available in the biopackage repository so that
>> folks can pick just the ones that they want. Is that something you
>> would be interested in?
>
> As we get our cluster working I'm interested in learning how to add new
> rpms for various packages I'm interested in which are not currently 
> part
> of BioBrew, R and some R software like Bioconductor. It seems likely
> we'll also want to add other non-bio software, though I don't know what
> at this time.

R would be a great package to add. The instructions to add rpms to make 
your own customized distro are in the Rocks User's Guide.  But I can go 
over them with you if you have any questions.
>
> I would probably most like to be able to install the Rocks 
> distribution,
> then pick and choose other rpms as needed to build a particular 
> cluster.
> I don't yet know what work is required to add rpms to a frontend node 
> to
> enable shoot-node of a new BioChemoDynoSimYouNameItBrew, but I'm
> guessing I'll be learning fairly soon. Seems to me the most flexible
> approach would be most useful to the most cluster admins.

This is what I have planned. I plan to release a vanilla base distro, 
Callident Rx, that people can customize to their hearts' content. But 
I'm holding off until I have a few more features ready.  Should only be 
another month or so.

> I like having
> all the packages you've selected for the BioBrew distro, but I already
> know I want to roll my own. Possibly some tools can be developed to 
> make
> this easy (easier?) Maybe it's really easy already and I just don't
> know...

It's easier than you think. Just take a look at the Rocks User's Guide.
>
> That said, there is a big advantage to someone starting out to know 
> that
> some community of users has already used a set of packages together 
> (mpi
> plus blast, plus ??) and that the set is available with a sensible
> starting configuration. Simply providing the srpms is not going to
> address that.
>
> Maybe srpms plus scripts to assemble a set of srpms into a particular
> Brew would be a useful way to structure things. It could get 
> complicated
> if you try to then combine two or more Brews into one distro; 
> dependency
> checking could be difficult, or you could end up with incompatible
> configurations somewhere. I don't know...

I will probably write a script that will pull down, build, and install 
software that I'm not allowed to redistribute, like TCoffee, from 
remote sites.

Glen