Transcript Networking-MarcusPattloch_03292007

How networks can fulfil today‘s and
tomorrow‘s Grid demands
- Organisational and Technical Challenges Marcus Pattloch, Karin Schauerhammer, Klaus Ullmann
(DFN-Verein, Germany, [email protected])
29. March 2007, ISGC 2007, Taipei
Page 1
Contents
• User Collaboration Structure
• Economical and Technical Challenges
– Some Technical Definitions
– Examples: X-WiN, Geant2 and the LHCOPN
– OPN Building Blocks
– Forecasts
• VPNs/OPNs, Grids’ Role for Networking, Transmission
Technology, User Demands
– Conclusions
• Bandwidth Provision, Future Developments and Needs
• Organisational Challenges
Page 2
How do networked users collaborate?
• Research collaboration has in almost all
cases an international dimension
• Example: For LHC collaborations the
processes for the four experiment‘s data
evaluation has an international dimension of
networking which is vital for success
• NRENs (National Research Networks) and
Geant2 (Pan European Network) have to
adapt to that situation - not only for the LHC
experiment evaluation process
Page 3
Economical and
Technical Challenges
Page 4
Some Technical Definitions
• VPN: Virtual Private Network
– a „user-owned network“ which is built from a
basic technical platform (IP / SDH / DWDM...)
• OPN: Optical (Virtual) Private Network
– a VPN based on optical technology
• Hybrid Network
– Router based network using a VPN / OPN as a
platform
Page 5
Forecast (F1)
• Communication Market
– F1: Liberalisation of communication market in all
European countries will continue, will (amongst
others) give better opportunities for research
networks and will decrease the „digital divide“ (or
the economic conditions for that divide)
Page 6
Example 1: X-WiN
(German NREN)
Page 7
X-WiN (1): Topology inc. cross-border fibre
KIE
Fibre A
AWI
Fibre B
DKRZ
DES
ROS
HAM
Fibre C
Fibre D
BRE
Surfnet
POT
HAN
MUE
BIE
BRA
MAG
TUB
ZIB
FFO
PSNC
ZEU
HUB
ADH
DUI
FZJ
AAC
BIR
JEN
FRA
BAY
GSI
ESF
KAI
HEI
ERL
FZK
Renater
CHE
ILM
Geant2
SAA
DRE
LEI
KEH
REG
STU
GAR
Richtung Basel
Switch/GARR
13.10.2006
Page 8
X-WiN (2): (Hybrid network) Features
• Platform available for national VPNs/OPNs
and for the national part of international
VPNs/OPNs
• Possible due to lively fibre market in Germany
• Optical technology delivers ample bandwidth,
i.e. 160*10 Gbit/s per link
• Costs per 10 Gbit/s link are relatively low (as in
Geant2) - in the order of 90 K€/a for 10 Gbit/s
Page 9
X-WiN (3): Targets (network design)
• More performance
– performance increase by factor 4 since 01/06 (same costs)
– more performance available as of 01/07 (for the same price)
• More flexibility
– no volume charging (and no usage limitation)
– Ethernet as additional access technology
– hybrid PoPs enable VPNs
• Higher availability
– during design for backbone implicitly taken into account
Page 10
X-WiN (4): Optical platform
• Toolbox for the provision of
– DFNInternet (10 Gbps to 10 Mbps)
– VPNs/OPNs based on optical links
– services like DFNVideoConference, DFN-PKI,
DFN-CERT, DFNRoaming, DFNNews unchanged
• New cost structures for optical networks
enable very economic network solutions for
specialised services like Grids (for example
OPNs)
Page 11
X-WiN (5): Router Platform for IP
EWE
BRE
KIE
ROS
GRE
DES
HAM
TUB
HUB
ZIB
ADH
AWI
HAN
BIE
BRA
MAG
GOE
CRSHAN
MUE
FFO
CRSPOT
DRE
POT
KAS
DUI
MAR
FZJ
CRSFRA
CRSERL
BAY
AAC
ERL
GIE
CHE
FRA
BIR
LEI
REG
WUE
JEN
AUG
XR
43 CISCO7609
KR
SAA
KAI
GSI
HEI
FZK
STU
GAR
ILM
1GE
10GE
2x10GE
Page 12
X-WiN (6) - Status
• Technical concept backbone
– backbone consists of (dark) fibre and leased
circuits
– operational responsibility: DFN
– more than before bought in partial services like
24/7 hotline
– much more than before DFN PoPs
• Economic concept
– backbone put together from different service
offerings from the market and integrated under
DFN responsibility
– most effective usage of competition on the market
Page 13
X-WiN (7): Scalability of architecture
• Options for transfer
– 10 Gbit/s ~ 1014 Byte/d or 100 TByte/d
– eleven 10 Gbit/s links -> more than 1 PByte/d or
roughly 0.5 ExaByte/a
• If one 10 Gbit/s link per T1 is not sufficient
– installation of just another 10 Gbit/s link T0 - T1 if
possible on physically separated fibre path
– architecture covers this completely
– upper limit of the technology is 160 links per
optical path
Page 14
Conclusion (C1)
• Bandwidth Provision
– C1: Bandwidth provision, which has been a major
economic problem for any research network over
the past decade, will not be a big problem in the
future (for 10 Gbit/s and below)
Page 15
Forecasts (F2 & F3)
• VPNs / OPNs
– F2: VPNs/OPNs will in a few years carry the bulk
of scientific data in European networks
(NRENs/Geant) and perhaps also world-wide
– F3: Migration to that scenario is an evolutionary
rather than a revolutionary process defined by user
group’s needs and available new network
technology
Page 16
Forecasts (F4 & F5)
• Role of Grids
– F4: Grids and Grid like systems will for the time
being be a major driver for the VPN migration
scenario process (examples LHCOPN, DEISA
network etc.)
– F5: Networking technology developments like
network management (example: monitoring of
VPNs/OPNs in multi domain environments) or
network security are still needed for serving „Grid
infrastructures“
Page 17
OPN Building Blocks
Page 18
OPN Building Blocks (1): E2E Links
• E2E Links are dedicated optical multi-gigabit connections
• Essentially P2P links, usually using SDH/SONET or Ethernet
GEANT2
E1
NREN1
NREN2
E2E Link 3
E2
NREN3
E3
E2E Link 2
• E2E Links are planned as a regular service of Géant2:
• Cooperation of several NRENs needed to operate E2E Links
• Users need Single Point of Contact (SPOC)
•E2E Link Coordination Unit (E2ECU) brings together Users and NRENs
during operations
Page 19
OPN Building Blocks (2): Workflow E2ECU
• Workflows define the interaction between Actors
• Actors: Authorized Users (no End Users), e.g.
– LHC GGUS (Global Grid User Support)
– E2ECU (End-to-End Link Coordination Unit )
– NREN TNOCs (Thin Network Operation Centres of NRENs)
• Mostly human/organisational communication
• Full life cycle of E2E links is covered
• For now, only Workflows for technical aspects are
defined
• Refinements still under discussion in GN2-JRA4/WI3
Page 20
OPN Building Blocks (3): E2E Link Monitoring
• Status information corresponds to network
layers 1/2
• Multiple technologies / vendors used to provide
an E2E service
– Status information is logical abstraction from vendor
solution
– No information about physical devices necessary
• Status of partial links (within domains and
connecting domains) is provided by NRENs
• E2E link status is an aggregation of partial links
Page 21
Demo Monitoring
• http://cnmdev.lrz-muenchen.de/e2e
Page 22
Example 2: Geant2
Page 23
Geant2 topology as of 09/06
Page 24
Example 3: LHCOPN
Page 25
LHC TIER0 – TIER1 OPN, scenario based
on work by R. Sabatino (DANTE)
RAL
NorduGrid
BNL
FNAL
CERN
T0
TRIUMF
S-Janet
ASCC
NorduNet
SWITCH
Surfnet
GEANT2
Renater
SARA
DFN
GARR
GRIDKa
Rediris
IN2P3
PIC
CNAF
Page 26
LHCOPN in Europe
• T1-T0 primary connection
– for „Geant2 fibre cloud NRENs“ through Geant2
• T1-T1 secondary connection
– on separate fibre paths through other fibre.
Secondary connections provide resilience
• OPN approach
– high data volume expected, Grid middleware
driving this approach; „low“ prices for optical links
due to liberalised situation per country enables it
Page 27
T2 communication to T1 in DE (1)
• Open Issues
– Which access pattern is requested by the T2s to
T1? Only GridKa or other T1s as well? „Other
T1s“ would be T1s in other NRENs.
– Which access pattern is requested by the T2s to
other T2s or T3s?
– What are quantitative access patterns of T2s?
Page 28
T2 communication to T1 in DE (2)
• No specification available yet
• In Germany T2- and T3-sites are known,
networking them is now on the agenda
• Principles for this part of networking could be:
– T2 sites need 1 Gbit/s access to T1 (which one?)
– Build resilient ring of core-T2 sites in Germany
– T3 sites access data through extended
DFNInternet service
• Problem has to be solved in 2007
Page 29
Forecasts (F6 - F8)
• Transmission Technology
– F6: On a (per user-) stream basis 10 Gbit/s will be
the main bandwidth to be used for the next 2-3
years, perhaps even longer - the difficulty is to get
data from sources to sinks at higher speeds than
10 Gbit/s
– F7: 40 Gbit/s or 100 Gbit/s per (user-) stream will
follow
– F8: 10 Gbit/s equipment will be very „cheap“
Page 30
Forecasts (F9 & F10)
• Qualitative User Demands
– F9: The user (group) demand in the research area
is in almost all cases „multi-domain / multi vendor“
in networking terms (--> see LCG example)
– F10: Users will require „intelligent networks“, i.e.
network technology which adapts (at best
dynamically) to their requirements
Page 31
Conclusions (C2 & C3)
• Developments
– C2: One of the main future challenges for the
developments of research networking is to further
work out solutions for multi-domain environments
for operational purposes
– C3: Work started in Geant2 („E2E“) but solutions
have to be driven further according to developing
demands for example from Grid communities like
the particle physics community
Page 32
Conclusion (C4)
• Future Needs
– C4: Intelligent networks (i.e. „intelligent“ VPNs /
OPNs adaptable and more flexible to user
needs) have to be further developed in the
future, i.e. VPNs „on demand“ or dynamic VPNs
Page 33
Organisational Background
Page 34
Governing Structure
• Presently the NREN Policy Committee
(NRENPC) is successfully governing the
networking policy definition and always
devised a flexible substructure (for example
Exec) to adapt to management needs
Page 35
The NRENPC as of 01/07
Country
NREN
Country
Austria
Belgium
Bulgaria
Croatia
Czech Republic
Cyprus
Germany
Estonia
France
Greece
Hungary
Ireland
Israel
Italy
Latvia
Lithuania
Luxembourg
Malta
Netherlands
(ACOnet)
(BELNET)
(BREB/ISTF)
(CARNet)
(CESNET)
(CYNET)
(DFN)
(EENet)
(RENATER)
(GRNET)
(HUNGARNET)
(HEANet)
(IUCC)
(GARR)
(LATNET)
(LITNET)
(RESTENA)
(UoM)
(SURFNET)
Nordic Countries (NorduNet)
Denmark,
Finland,
Iceland,
Norway,
Sweden
Poland
(PSNC)
Portugal
(FCCN)
Romania
(RoEduNet)
Russia
(JSCC)
Slovakia
(SANET)
Slovenia
(ARNES)
Spain
(RedIRIS)
Switzerland
(SWITCH)
Turkey
(ULAKBIM)
United Kingdom (UKERNA)
NREN
Plus Non-Voting Members:
DANTE, TERENA
Perm. Observers: CERN, AMREJ, MARNET
Page 36
Summary
• Economic situation for data networks improved
drastically within the last 10 years
• 10 Gbit/s VPNs economically achievable today
• High performance network technology is/has
been introduced in a couple of NRENs and
Geant2; they will be upgraded according to
available new network technology
• Main development topic: multi-domain issues
• Close coordination between demanding user
groups and networkers is absolutely necessary
Page 37