Dear All,
Just to disprove the statement that nobody ever reads the notes taken
from our meetings, Roger has replied with a note on Atlas policy (see
the email below). As our first priority is supporting the experiments
this has quite some consequences for the small to medium sized sites
that are primarily Atlas sites. If Atlas is actively encouraging these
sites not to have storage but to concentrate (and spend their money) on
CPU then this is what they should do. This would have significant
implications for the next round of purchasing for which the money is
about to land into our bank accounts.
I don't know the Atlas computing model as well many of you out there so
I don't know if (like CMS) it has the concept of local grid storage for
users (to write out their ntuples etc). If it does is this something
where the bigger sites could help the smaller sites as well? Ie that
there would be local storage at the bigger sites for the user
communities at the smaller sites?
Clearly, moving a number of sites to being CPU only would reduce the
management burden at those sites. It would also reduce the variety of
resource management tools that we would need to support.
Thoughts and comments would be very welcome.
Best,
david
-------- Forwarded Message --------
Subject: ATLAS storage policy
Date: Mon, 28 Nov 2016 13:45:04 +0000
From: Roger Jones <[log in to unmask]>
To: D J Colling <[log in to unmask]>
Hi Dave,
Reading the Technical Board minutes, I note it makes a statement of
ATLAS policy that is at odds with the policy as stated by the ATLAS
Computing Co-ordinator, who confirmed
“We encourage small sites to be diskless and invest in CPU; we are also
exploring federated storages for medium size sites.” This is much less
strong a statement on federation that I read from those minutes, and we
certainly encourage diskless sites. I might add that the definition of
small moves with time, and I believe in the UK has a threshold around
the 1PB rather than lower values discussed a year ago.
Best
Roger
|