Burke, S (Stephen) wrote:
> Testbed Support for GridPP member institutes
>
>>[mailto:[log in to unmask]] On Behalf Of owen maroney said:
>>I am somewhat alarmed by this statement. Up until now I have
>>understood
>>that, while it is required that sites support gsiftp, rfio
>>was optional.
>
>
> It's optional in the sense that it isn't widely used by applications
> except for access to castor.
I'd say it's optional because, at the moment, a site can pass all the
SFT without running it. My concern is that if new datamanagement
middleware doesn't work without rfio installed, then this will clearly
mean a new service has been added to the list of services a site must
install to pass the SFT (and therefore be judged a 'functioning' site).
Of course, this has happened before (RGMA), and is happening now
(SRM-SEs) but at least in these cases there was a long period of
development and discussion of the new service.
Actually the vast majority of sites have
> deployed it, in both EDG and LCG, and it has always been a supported
> protocol.
"Supported" is not the same as "Mandatory"! PBS/torque is supported as
the default batch system - but this does not mean all sites have to run
pbs as their batch system!
I suspect the vast majority of sites deploy rfio as it has always been
in the default installation, rather than through any active decision. I
also suspect if it was not in the default configuration and sites had to
make an active choice to include it, few sites would do so!
Sites which do not install rfio and do not advertise it as a supported
protocol in MDS, don't seem to break anything - so clearly it is not
currently a required service!
As things stand it's the only way to get posix-like access to
> files, which experiments always claimed was a high priority, but which
> doesn't seem to get used in practice.
So far no SFT was broken by not having rfio and no experiment has
complained they cannot run their software at sites without rfio.
If anything currently required rfio, an SFT for it would almost
certainly have been added...
>>My concern is if new datamanagement code has been written *presuming*
>>rfio is available, then will these sites now be *forced* to
>>install rfio?
>
>
> No-one is ever forced to install anything ...
Only in the sense that no-one is forced to join the grid (apart from any
MoU commitments, of course!). I think anyone would accept that a site
that wants to be part of the grid would need to support gridftp,
globus-jobmanager, suitable monitoring and information reporting etc.
It is far from obvious to me that rfio should fall into the same category!
>>And so a previously optional protocol becomes mandatory... when was
>>this discussed/decided?
>
>
> As you well know, virtually nothing is ever discussed or decided in a
> public way!
Indeed!
But sites in the UK are only MoU committed to install software that
GridPP decides on. So we have do have a process here: if GridPP demands
rfio, sites need to install it or they renege on their MoU. If GridPP
doesn't demand it, UK sites shouldn't need to install it!
Perhaps GridPP can raise the issue with LCG: will rfio become an LCG
required service in future middleware releases?
(Follow up question: If so, why is this new service necessary?)
A lot of the new LCG software (DPM, GFAL, LFC) has been
> written by Jean-Philippe Baud and hence has re-used the castor
> code-base, including rfio.
In other words, middleware creep.
If the principle that any service that is not required should not be
installed, had been followed by the default installation, then there
would not have been the assumption that rfio would be present and this
new datamanagement middleware would either have had to use something
different, or at least justify why the introduction of a new (possibly
quite insecure!) protocol was required.
And I assume there is client software to be installed on the WN for
these? (which they should always be trying to *reduce* not *increase*!)
That's not to say that the security hasn't
> been improved, I hope it has, but I can't say anything definite because
> there's hardly any documentation. In any case, since LCG always
> maintains that it doesn't do any development there is no mechanism to
> review the development it does ...
;-)
--
=======================================================
Dr O J E Maroney # London Tier 2 Technical Co-ordinator
Tel. (+44)20 759 47802
Imperial College London
High Energy Physics Department
The Blackett Laboratory
Prince Consort Road, London, SW7 2BW
====================================
|