Dear All
Following recent questions on required network bandwidth for Tier-2 sites,
Pete Clarke has kindly written a brief summary of what is known within
GridPP (see below - but note that this does not tackle the exact question of
what the needs are to be part of a data challenge). It seems that few tests
have been performed in the area of network bandwidth available to Tier-2
(and Tier-1) sites (i.e. realistic sustainable data transfer rates) which is
why the idea of a series of network challenges is looking increasingly
necessary.
If any of you already have useful lessons learned from independent data
transfer tests (or have experienced data transfer problems during the data
challenges to date - rates that I understand have been quite low) please
could you let me know. Many thanks.
Kind regards,
Jeremy
---------------------------------------------
Colleagues
The question of Tier2 (and Tier-1) bandwidth requirements
has re-arisen recently in a few places. E.g. Simon George asked on
the TB support list, and it was raised at the Tier-1 network
meeting recently where Robin Tasker was present.
This mail is a brief summary of the practical state of play
w.r.t. assimilating and presenting HEP network requirements
to relevant authorities.
1. About a year ago all experiments in the UK were asked for an update
of expected flows into/out of their sites. That information which
was forthcoming is encapsulated in the appendices of the attached document.
Robin Tasker will put this up on the GridPP website.
2. There is a recent spreadsheet from David Foster with numbers which
are a bit updated from previous numbers, mainly due to a change in the
amount of raw data to be replicated, and a few duty cycles.
This has the advantage that it was gathered from experiments under
an LCG mandate, so I believe the experiments took it seriously.
This is about the most up to date thing which exists. This
should be summarised and put up, but unless Robin can find someone
to do it, it will have to wait until 3 below.
3. From as soon as possible after September 1st, I will have a half
post whose job will be to care about network support for all Tier-2.
Part of this job will be to assimilate and report on exactly
this sort of thing. So, for example, a 6 monthly "best guess update"
on bandwidth requirements could be produced. The appointee will
visit Tier 2 centres in due course.
4. However, when all is said and done, regardless of any documents we
produce, or meetings we go to, here is the practical answer:
If you are a major Tier 2 centre (meaning have significant storage
and expect to be a significant node in your experiments data
management), then you will need
AT LEAST 1 Gbit/s connectivity to your site (to allow sustained
rates <= 500 Mbits/s)
PREFERABLY 2 X 1 Gbit/s or 2.5 Gbits/s connection to your site
(to allow sustained rates ~ 1 Gbit/s)
This is a statement which is likely to be true for the next
few years. It embodies "aspirational" numbers with the "practical reality"
of what people are actually able to do with limited resources, and the
"limitations" of current servers.
You can safely use this with your local campus LAN authorities.
5. Finally, please come to the UKLight town meeting if you are
interested in doing some more "out of the box" work to use
switched lightpaths.
http://www.nesc.ac.uk/esi/events/457/
This is all over and above the production network connection.
Pete
|