Hi Stephen,
Stephen Burke wrote:
> On Mon, 10 May 2004, Ian Stokes-Rees wrote:
>
>>Which points are appropriate for the GDA meeting (which I just missed
>>for today)? I think there were a few issues I raised:
>>
>>* queue configuration
>>* software installation
>>* need for per-VO RB,BDII,UI
>
> What are your issues with the second and third points? Software
> installation has been discussed a lot, and I thought the current scheme
> (installation into a shared area) was broadly accepted by the experiments,
> although not necessarily all the sysadmins.
I think this crosses the boundary between two meanings of the word
"issues". In terms of issues=problems, I don't have any with software
installation. In terms of issue=comment, it is tied in to the idea that
I am wondering "Wouldn't it just be easier to have everyone use gsi-ssh?".
> On the last point, you don't *need* VO-specific services, but CERN is in
> fact running brokers and bdiis for the LHC VOs and you can add
> non-standard sites to the bdii if you want. UIs are generally a site-level
> or user-level thing, you should be able to put the UI software on your
> laptop or desktop machine.
Again, this is a matter of the amount of extra work it takes to set-up
these per-VO pieces, compared to just providing a web-page with a list
of VO-aware (e.g. LHCb aware) sites and then just gsi-sshing to them to
use whatever local batch system they provide. I know tracking jobs
across sites is terrible, and one which the RB, UI, and "grid" paradigm
are meant to resolve, however in practice the difficulty of tracking LCG
jobs at the moment makes this option appear perhaps not so aweful.
I would love to install the LCG UI software on my desktop machine. I
didn't realise this was now possible. Do you know where I can get
details regarding it? My earlier attempts to do this were time
consuming and unsuccessful.
>>And then the meta-issue of the difference between the emerging LCG usage
>>pattern, and just being given (gsi-)ssh access to each CE to install
>>software and run jobs there directly.
>
> You can submit jobs through globus-job-submit, but what would be the
> advantage?
Avoiding per-VO configuration and an entire layer of software, I suppose.
Overall, I suppose I am just trying to think about the bigger picture
where the LCG is aiming towards something easy to use and well
integrated that manages to hide heterogeneity and complexity, both from
the user, and from VOs.
Ian.
--
Ian Stokes-Rees [log in to unmask]
Particle Physics, Oxford http://www-pnp.physics.ox.ac.uk/~stokes
|