Models for a sustainable NGS
University Resources
and fEC
Peter Clarke
Feb 22th 2007
NeSC
and Academic Director of
ECDF
(Edinburgh Compute and Data Facility)
Costs of Research Computing in
Universities
Obvious Elements to be covered
Power and Environment = 20%
Systems Staff = 40%
Depreciation = 40%
Other less obvious elements
Service evolution
Scalable and continuous requirements capture
Middleware /software services
Training and education (enabling efficient use of services)
Software development (advancing code base technology)
What university research users need
Baseline free at point of access service
This has no guarantees but is available to all researchers
Pre fEC grants
Speculative nobel prize winning work
Research without traditional research grants (e.g. humanities)
Guaranteed Resource service
This will guarantee a certain resource against agreed costs
For specific projects with specific needs and funding sources
Essential to be able to "commit" to a project
Essential to obtain fEC costs from RCs
What university research users need (2)
Agility
Capability
Service Level
What University /School /Department
finance probably believe
Computing is like any other commodity (paper clips, animal houses)
Researchers will be able to pay for everything they use from grants
at 100% (=80%) immediately now there is fEC in place
The Disjoint
Researchers don’t have mindset to pay full costs
It has to be cheaper than I think it will cost me to roll my own
My dept doesn’t charge me for power and my postdoc runs the system
I don’t want to loose flexibility
Im not sure central services will give me what I want
Research councils havn't fully assimilated this in practice
Peer bodies may still not have bought in fully to the real consequences of fEC
Owning computer means its in RC control.
Still talk of "University contribution"
Finance
Not all research use of computing can get a grant to pay costs in advance - and we
don’t even want this to be the case. It is against the Universities research mission
for this to be the case.
Finance often groups don’t understand this.
In my opinion we are in danger of an unintended
consequence of fEC on a major scale I.e. decimation of legitimate access to computing for
research for a few years.
My own model (1)
Any research led University must provide a baseline free at point of
access service if it is serious about research excellence
Agile local Computing access is like a phone
-
-
would we really want to say you cant make a phone call if you don’t have a DA cost
on a grant to pay the call charge ?
Would we want to inhibit access to library on same basis ?.
It should ideally be paid for via an indirect cost rolled up with other such research
support like libraries. This could be agreed with RCs
If not is must be an agreed DA cost on all grants (possibly weighted according to
discipline)
The University must then provide a way to channel those funds to support this !
It is not very expensive to provide this within a University spread across all its research
My own model (2)
Any research led University may provide a chargeable guaranteed
resource service as part of the hierarchical agility-capability
pyramid.
It must be paid for as a DA cost on grants
It must have auditable service level delivery
It will inevitably appear to be much more expensive than the baseline
service. We all have to understand this.
It must be discounted heavily over a 5 year transition period to fEC to
match the practicalities of transition.
See example =>
ECDF Price Roadmap
%baseline service
2006-7
2007-8
2008-9
25
2009-10
2010-11
Price/Cost factor
Power
Staff
Depreciation CPU
Depreciation SAN
Buildings+other
Unit
%
%
%
%
%
100
0
50
25
100
100
0
50
25
100
100
50
50
50
100
100
100
50
50
100
100
100
50
50
100
Total price for baseline service : to aggregate in University overall Indirect cost
Јk
57
57
110
154
154
Price for guranteed_resource_service : Directly allocated cost on grants
Price per kSI2k per year
Ј
95
95
158
Price per Tbyte per year
Ј
475
475
1113
222
1550
222
1550
My own model (3)
Everyone must get away from the outdated idea that they have to own
the hardware to provide the service, or to be a great University.
We are all happy with the virtualisation/devolvement of networks - why not TBytes or
CPU-cycles?
Its all about virtualising - and raising the dotted line
What is important is to provide the service the research community needs at a sensible
price. If this means ACME-storgae PLC can provide the actual TBytes then so be it.
You could imagine a PowerGen <=> Scottish Power <=> User relation
This applies probably to TBytes and commodity CPU cycles in near term
This probably doesn’t apply to specialist HPC use in very near term
My vision: The dotted line will rise
Scientific Data, Applications & Knowledge Discovery
Virtualised Computing Services
Computers & Storage
The Internet
Researchers will learn not to care where resources are - same as they
don’t care about the network now
Relation to National Services
I (now) see no conflict
For HPC
There is a clear separation of need for Campus scale <=> National Scale resources.
National services are already paid for realistically.
Campus service just need to fit in on on open market.
For commodity computing and storage
There is no plan to fund masses of this centrally via JISC
NGS role is not to provide free computing for all (nor could it)
NGS provides much more important things. !!
So I now don’t think a free NGS undermines campus sustainablity
But this is worthy of debate
The more important question is what should NGS do so that JISC/RCs/Institutes
will be happy to fund it in the way they do SuperJANET
…my list
A baseline hardware service
To set standard
For evolution of service
For development by users
Defining a common standard for federating University resources
Software certification, development, deployment planning
Distributed community support
Aggregation of requirements across Institutes
Access to specialist services which it would be stupid to provide at each site
Helpdesk
….. and more….
We live in a federated world