Transcript Peter Clarke (panelist)

GLIF Infrastructure
SC2004 Panel Discussion
Peter Clarke
UK National e-Science Centre
GLIF Panel
Peter Clarke
SC2004
Questions
 Why are Canada, Japan, the Netherlands, the United
Kingdom and the United States involved in GLIF?
 What can be accomplished on lambdas that cannot
be accomplished on "best-effort" networks?
 Where does your infrastructure connect?
 What's missing? When will we implement a totally
functional LambdaGrid?
 What do you see as the future of networking?
GLIF Panel
Peter Clarke
SC2004
Questions
 Why are Canada, Japan, the Netherlands, the United
Kingdom and the United States involved in GLIF?
 Community Lobby
 National Competitiveness in Research
 Acceptance that one size manifestly doesn’t fit all
 Capability for Research
 You cant do it alone (by definition its an end-2-end thing)
 Trust
GLIF Panel
Peter Clarke
SC2004
Capability :
Steering & Migration of Computationally intensive simulations
Cubic micellar phase,
low surfactant density
gradient.
Initial condition:
Self-assembly
Random water/
starts.
surfactant mixture.
Rewind and
restart from
checkpoint.
GLIF Panel
Peter Clarke
Lamellar phase:
surfactant bilayers
between water layers.
SC2004
Source: Reality Grid Project www.realitygrid.org
UK HPC machines connected
to TeraGrid for SC2003
Now a persistent connection
GLIF Panel
Peter Clarke
SC2004
Capability: Breast Screening Programmes
Current system is
“minimal”
technologically
GLIF Panel
(taken from e-DiaMoND Project)
Peter Clarke
SC2004
Future should be
digitised and
processed images
(taken from e-DiaMoND Project)
Remote
Radiographers ?
NETWORK
2nd opinion
Remote Patient
information
• Requires ~ Gbit/s flows for remote access
• Will not be possible without scheduled guaranteed net-services
GLIF Panel
Peter Clarke
SC2004
 What can be accomplished on lambdas that cannot be
accomplished on "best-effort" networks?
 An R&D network is not a production network – you can
do “out of the box” things
 Certainty of having what you need when you need it
(but not scalable today of course) – means research
computing can be planned
 Ability to use non standard transport
 Fixed latency => “backplane” connections between
compute centres => virtual machine rooms
Provide virtual networks a la UCLP, and push
management to the users
GLIF Panel
Peter Clarke
SC2004
CDF – Connection UCL-FNAL
GLIF Panel
Peter Clarke
SC2004
 Where does your infrastructure connect ?
GLIF Panel
Peter Clarke
SC2004
Source: Kees Neggers, SURFnet
10 Gbit/s
New York
MANLAN
Stockholm
NorthernLight
IEEAF
10 Gbit/s
2.5 Gbit/s
10 Gbit/s
2.5 Gbit/s
10 Gbit/s
Tokyo
WIDE
CA*net4
10
Gbit/s
IEEAF
10 Gbit/s
Chicago
StarLight
10 Gbit/s
SURFnet
10 Gbit/s
NSF
10 Gbit/s
10
Gbit/s
Amsterdam
NetherLight
2.5 Gbit/s
SURFnet
10 Gbit/s
Tokyo
APAN
10 Gbit/s
London
UKLight
GLIF Panel
2.5 Gbit/s
Peter Clarke
Geneva
CERN
Prague
CzechLight
SC2004
As a result of the initiative, UKERNA have pushed through a
complementary initiative to provide a nationwide domestic R&D
network in parallel with SupetJANET
NNW
T
UHI Network
Clydenet
FaTMAN
T
C&NL
MAN
C
AbMAN
EaStMAN
T
Glasgow Edinburgh
S
NorMAN
T YHMAN
Warrington
Leeds
T
MidMAN
Reading
London
T EMMAN
T EastNet
Kentish MAN
TVN
Bristol
Portsmouth
South Wales S
S
MAN
SWERN
T
LeNSE
Northern
Ireland
GLIF Panel
T LMN
S
Peter Clarke
London
UKLight
SC2004
 What's missing? When will we implement a totally
functional LambdaGrid?
 When NRENS all adopt hybrid networking as the normal
business model
o Requirement: irrefutable proof of benefit
o Political: national competitiveness, one price for all
issues..
o Financial: proof of cost effectiveness
o Policy:
 When we have a common and “standards” based control
plane infrastructure – not individual solutions : now is the
time to get together on this one (IMHO)
 I have no idea about the last mile issue
GLIF Panel
Peter Clarke
SC2004
 What do you see as the future of networking
Wrong guy to ask – I think from point of view of future of
computationally intensive research where I would say:
We must get to point where network is not seen by
researchers as a big barrier which keeps sites disjoint
 Must ensure it is understood that research capability is
currently inhibited by “traditional” network provision
Must ensure it is understood that one size doesn’t fit all is
a good idea
(there is no way research could be carried out if we all took
commercial IP connections)
GLIF Panel
Peter Clarke
SC2004
… a possible Network of the Future….
Router
Routed IP Network
Router
“layer-2” split out of
wavelength
[e.g. Ethernet]
Provides
• Normal best effort IP network where appropriate
• Extended “virtual LANs” where appropriate
• Switched lightpaths where appropriate
GLIF Panel
Peter Clarke
SC2004