[glue-wg] How to Save Glue! WRT OGF

[owen.synge at desy.de]

Dear all,

Thankyou Laurence and Sergio, I am pleased that you both share my
concerns, although less alarmist than I it has been clear from your
steadying hands that you are also concerned about the creeping scope
from discovery to modeling and monitoring within Glue.

How do I think it could all be resolved so that the experiments and the
people who want Glue to be a standard for more than HEP can coexist
happily?

I feel that Glue and HEP should acknowledge that we are working on a
schema, and the great thing about schema's is they can be extended, or
in OO speak specialised.

I acknowledge that we have some use cases that require publishing and
subscribing, and that HEP grids have little choice but to use
established Grid infrastructure for there short term production needs.
This is not Glue, just the infrastructure for distributing
data.

So I propose that HEP dominated grids such as wLCG, OSG and NorduGrid
should have a separate standardisation process for there extensions to
the markup of the there Glue as a separate paper and standard
specification.

What I see currently particularly in storage is a specialisation of the
Tier 1 and Tier 0 markup for wLCG et al Grids. In the context of these
grids schema I would find it much more acceptable to add monitoring and
Authorization information and even tightly couple to existing Grid
infrastructures. Even if I feel that a relational database is a more
natural fitting for some fields.

In and ideal world I would also allow VO's to be able to publish VO
specific data within the publish subscribe infrastructure that GLue
also uses. As I know form previous and continuing standardisation
efforts its hard work to get what you need in a standard, allowing a
free hand to add and remove markup of sites for experiments may be
hugely beneficial, particularly as it would remove an imposition to a
liberty.

What I object to is the hard work of building an abstract description
of a site being specialised to the point where it is no longer
universal. I see great danger that Glue will only work in its current
form in Tier 0 and Tier 1 sites on HEP community grids, I am even
concerned that it will be too heavy weight for tier 2/3 sites. I see
Glue as a higher purpose, sharing our experience of building a grid
with communities that want to build new potentially better grids or
cloud computing infrastructures. We may be holding back its adoption
with other grids, and the infrastructure that experiments could
effectively use with their own markup. The solution maybe to partition
the schema effort into multiple views. Universal, HEP Grid Community
(wLCG, OSG and NorduGrid), and VO based schema.

Since wLCG and VO's want additional markup on top of Glue they are the
ones who should mark up the sites, and quite rightly they should be
able to demand from their middle ware additional requirements that
would not be universal for all grids.

For me Glue is something we should be proud of, as Grid pioneers in
wLCG, OSG, NorduGrid and the assorted experimental communities we can
give the world of grid computing a basis upon which to share or
experience of what is common and clearly separate what is not meant to
be universal.

This would make me happier

Regards

Owen Synge