Roger Jones couldn't reply to this list. In another mail he confirmed to
me:-
Indeed, no permanent storage at a Tier-2 - but certainly not all
volatile.
Durable (at least in the common English sense) covers it!
Cheers,
Roger
-----Original Message-----
From: Roger Jones [mailto:[log in to unmask]]
Sent: 06 October 2005 14:26
To: Gordon, JC (John); 'Testbed Support for GridPP member institutes'
Cc: 'Dario Barberis'
Subject: RE: Reminder of the next UKI meeting - Wednesday 5th October
11:00-12:00 via VRVS
Hi,
Speed reading here to try to gather the thread. (I could come back to
this next week if this needs more discussion.)
The ATLAS eventual usage of the T2 storage is indeed to make various
datasets (primarily for analysis) available. This does not imply a
custodial role (the data will either be available elsewhere or
reproducible without too much pain), but should not be deleted on spec.
The last time I heard a discussion of the SE flag, I thought there
were three available states: volatile (essentially SCR%WEEK), an
intermediate state and permanent. My interpretation of the last two
states in the light of the computing models would be that the
intermediate state would mean managed storage - probably with a quota
for the VO and management of quotas within the VO by a join of global
and local policy. This would have no implication of a custodial role -
if the disk crashes, the VO or user has to recover it somehow. Given the
quotas, the local sysadmin would not be deleting any files unless the VO
asked them to. Permanent should mean that the data is VO managed, but
there is a safe custodial copy.
I had noted a couple of people saying they didn't thing the middle
state meant anything; obviously from the above, I think it does.
The points John rightly highlights below would then imply that for a
T2, the bulk of the storage for ATLAS would be in the middle category,
the temporary storage would be volatile and there would be a very small
need for permanent storage. In fact the need for the latter will depend
very much on the more recent discussions of the VO access policy and
roles in ATLAS.
Hope this gives some clarity not more confusion!
Cheers,
Roger
-----Original Message-----
From: Gordon, JC (John) [mailto:[log in to unmask]]
Sent: 06 October 2005 13:57
To: Testbed Support for GridPP member institutes
Cc: Roger Jones
Subject: RE: Reminder of the next UKI meeting - Wednesday 5th October
11:00-12:00 via VRVS
I would be surprised if a knowledgible person in ATLAS had made this
statement. As I can't be bothered reading their TDR in full I am copying
in Roger Jones for his comments. I have glanced through it and find the
following requirements for a Tier2 (I include the whole list for
interest)
- Medium- or long-term storage for required samples. For analysis work
these will be mostly AOD but could include some fraction of ESD or RAW
data for more detailed studies.
- Transfer, buffering and short-term caching of relevant data from
Tier-1 and transfer of produced data to Tier-1 for storage.
- Provision and management of local temporary working space for
analysis, calibration of development tasks.
- Database services in agreement with the ATLAS DDM strategy.
- Prioritization of access to data and computing resources.
- Accounting for processing and data storage.
- Publication of the necessary information to be used by the Grid
services and the ATLAS Production System.
To me, the first three seem relevant. If the fourth implied database
backup I would expect it to be a dedicated system and not a general
purpose SE.
Medium and long-term storage seems consistent with my understanding of
'permanent' without implying any quality or guarantee of service.
In contrast the ATLAS description of a Tier1 includes the requirement -
In accepting data from Tier-2, a Tier-1 accepts to store them in a
permanent and safe way and to provide access to it in agreement with
current ATLAS policy. This is true for both simulated and derived data.
This seems fairly clear to me, the ATLAS model recognises different
levels of storage quality and identifies them in T2 and T1. Worth
talking further with them as to whether they expect these differences to
be published in the information service.
John
> -----Original Message-----
> From: Testbed Support for GridPP member institutes
> [mailto:[log in to unmask]] On Behalf Of Simon George
> Sent: 06 October 2005 13:30
> To: [log in to unmask]
> Subject: Re: Reminder of the next UKI meeting - Wednesday 5th October
> 11:00-12:00 via VRVS
>
> Hi John,
>
> I also had the impression from the discussion with ATLAS a while ago
> that they were surprised that our SEs could not be regarded as
> permanent, high quality, backed up storage. And I also drew the
> conclusion (as I think did everyone in London
> Tier2) that the recommended way to flag a SE which is not backed up
> and could go down for a long time if there was a failure (due to
> limited manpower at the site) was "volatile".
> If several of us thought this then there must be something in it, even
> if it is misguided.
>
> To answer you Q: we don't think our SEs are less reliable than
> everyone else's. Rather, we think other people are are not necessarily
> advertising the true nature of their site. A
> Tier1 SE is so different to what a Tier2 site can provide.
> Some sites my have better support and facilities than others.
>
> btw This was in addition to the discussion with ATLAS about whether
> sites could delete old files.
>
> I think we do need ways to distinguish between the differing qualities
> of SEs. I know for ATLAS, long term availability and uptime are
> important factors. Disk vs tape (access time) is also important. The
> way forward should be a quick discussion between sites and experiments
> of how they would like SEs to be described, then update the Glue
> Schema, document it and we will fill it in. If there is no forum for
> this, please set one up!
>
> Cheers,
> Simon
>
> --------------------------------------------------------------
> -------------
> Simon George, Dept of Physics, Royal Holloway college, University of
> London
> Email [log in to unmask] Tel. +44 1784 41 41 85 Fax.
> +44 1784 472794
>
> On Thu, 6 Oct 2005, Gordon, JC (John) wrote:
>
> > Would either of you care to spell out why you think your
> systems are
> > more unsafe than everyone else's?
> >
> > John
> >
> > > -----Original Message-----
> > > From: Testbed Support for GridPP member institutes
> > > [mailto:[log in to unmask]] On Behalf Of William Hay
> > > Sent: 06 October 2005 12:36
> > > To: [log in to unmask]
> > > Subject: Re: Reminder of the next UKI meeting - Wednesday 5th
> > > October 11:00-12:00 via VRVS
> > >
> > > > Hi John,
> > > >
> > > > I have declared my storage 'volatile' and I've done so not
> > > because I
> > > > delete files whenever I feel like but because I don't
> consider my
> > > > storage safe enough to be declared permanent and if
> > > something happens
> > > > I know I've done my duty warning the experiments in advance. If
> > > > experiments don't exploit the glue schema they should.
> > > That's why we
> > > > have an information system.
> > > >
> > > > cheers
> > > > alessandra
> > >
> > > Ditto.
> > >
> > > William Hay, UCL-CCC Site Administrator, Information Systems,
> > > EISD,UCL
> > >
> >
>
|