> -----Original Message-----
> From: Christopher J.Walker [mailto:[log in to unmask]]
> Sent: 31 May 2012 17:07
> To: Testbed Support for GridPP member institutes
> Cc: Gordon, John (STFC,RAL,ESC)
> Subject: Re: glexec tests for ATLAS
>
> On 31/05/12 16:41, John Gordon wrote:
> > Jeremy, why do you interpret that GDB agendum to be that glexec is
> > up for discussion? This sounds as if you still do not believe that
> > it is a requirement. The discussion will be about getting it
> deployed,
> > not whether. The UK has done a fair job of deployment so far so let's
> > get it finished. Your language may deter some sites who haven't yet
> > deployed or make them feel they have a loophole to avoid changing
> anything.
> >
> > Similarly, WLCG is a collaboration between sites, experiments, and
> middleware
> > providers. It is not separate from the experiments. For six months
> the output
> > from the TEGs has been that identity switching is crucial for
> identity tracing
> > and separation of workload. If there is an alternative to glexec then
> make it
> > known; otherwise let's just get it deployed. And let's forget the
> 'waiting
> > for experiments'. We know the experiments are waiting for the sites.
>
> And a number of UK sites are waiting for middleware developers to
> release a tarball release. QMUL certainly is. Why has this not
> happened?
Jeremy knows the full story but I think it has to be compiled knowing the location. The middleware developers won't impose a location. We could build one for the UK. I thought someone had been talking to the French who did that.
>
>
> > They are not going to make their tests critical until it has been
> working
> > at a large range of sites.
>
> Chris
>
> PS In QMUL's case, we could deploy an rpm that didn't depend on the
> rest
> of glite, but I'm not aware that's available either.
>
> PPS If the model is (and it seems to be) all sites to deploy it, then
> experiments test it, that's broken. What should happen is that some
> sites deploy it and are tested by the experiments, the bugs are fixed,
> and then the rest of the sites deploy it.
I didn't mean every last site but it has to be a reasonable number to cover a multitude of batch systems etc.
|