Dear Mike,
EasyGrid also has problems due the middleware. See
http://www.hep.man.ac.uk/u/jamwer/#sec15
The stress test was only 800 jobs, which is nothing if compared with real
production.
I would like to study the problem with more detail, using 10 computers, but
it is not possible.
Best regards,
James
>From: Mike 'Mike' Jones <[log in to unmask]>
>Reply-To: Mike 'Mike' Jones <[log in to unmask]>
>To: [log in to unmask]
>Subject: Ganga + Alibaba, 2d worth , Was: Provoking?
>Date: Mon, 5 Sep 2005 12:23:26 +0100
>
>On Mon, 5 Sep 2005, James Werner wrote:
>
>>I was working at Imperial with Janusz and I know his work.
>>The problem about develop software for BaBar is we have to provide at
>>least the same level of reliability, flexibility and functionality of
>>SLAC. Years ago users tried to use Alibabar and Ganga, it does not work
>>and users went back to slac.
>
>Are you absolutely sure this is true James? -- as I remember it, the proof
>of concept was demonstrated.
>
>I would rather you rephrased the sentence above to something more along the
>lines: "Years ago users tried to use Alibaba and Ganga, users could not
>get it to work and went back to slac."
>
>Also the years ago you're referring, to most middleware in the arena was in
>the same state. Some users tried it and were throughly disappointed. It
>might take some time to persuade those people back when we've really tamed
>the HEP grids!
>
>Furthermore, I'm not sure if /users/ did actually use the Alibaba-Ganga
>combination. I'm not sure it was actually released.
>
>For Ganga, the only change I made to Alibabar was to allow for Ganga's
>separation of job preparation and job submission stages, something that
>added functionality to Alibabar (useful both within and without Ganga). It
>turned out that this introduced probably about 20-30 lines of code (not
>much eh?); 90% of that work work was actually making sure that the Globus
>and AFS and firewall and ... environments were up to scratch on the Plug-in
>developers' machines (something you'd probably expect to be there,
>maintained and reasonably understood these days)!
>
>Remember that Alibabar's gsub is a throw-down substitution to qsub if you
>like to view the grid as one great big batch system. If you prefer to look
>at it through Globus tainted/tinted spectacles it's a throw-down
>substitution to globus-job-submit. (It could just as easily be a substitute
>to edg-job-submit, lcg-job-submit, glite-job-submit or whatever it's called
>these days!)
>
>What ever calls Alibabar doesn't matter; however Alibabar works doesn't
>matter. Whatever calls easy-grid shouldn't matter and whatever calls the
>next generation all-singing all-dancing solution to Babar Grid shouldn't
>matter either. So long as the interfaces are well described and remain
>reasonably stable.
>
>Mike
>
>--
>http://www.sve.man.ac.uk/General/Staff/jonesM/
|