Networking Discontinuities NSF eSCcence

1
N E T W O R K D I S CON TIN U IT IES b e l a b o
r i n g t h e o b v i o u s ?
Ron Johnson University of Washington, Pacific/Nort
hWest Gigapop, P A C I F I C L I G H T R A I
L ronj_at_cac.washington.edu 206 543 8252
12/5/01
2
Overall Goals ( P A C I F I C L I G H T R A I
L ver.)

• Bring together leading edge RD sites,
  resources, apparatus and groups in a flexible,
  ultra-high performance experimental ip and p2p
  wave and fiber based network fabric that can
  accelerate progress in science and medicine the
  development of Grid infrastructure and bleeding
  edge applications, architectures and content
  and, the next generations network technology and
  networked apparatus etc.
• Accelerate progress in science, RD econ. dev.
  that is leveraged by ultra-high perf. ip wave
  based experimental network fabric linking key
  research groups, sites and apparatus.

draft 12/05/01
3
RE Network Tiers
NETWORK DISCONTINUITIES
LEADERS NETWORK TYPE
CAPABILITIES/USERS
Research
Academic Research, papers, lab. Experiments,
spools of fiber In labs
Computer Science Net. Researchers
Bleeding-edge e2e services facilities
for key experiments, programs,
projects, researchers for next gen.
tech., architectures, Grids,
content, apparatus, etc.
Things that havent
been done before with,
to, on or over networks.
CS, Comp. Science Grid researchers, major
experiments (eg DTF) net. experimenters
leading edge labs and centers, optics sys.
net. Industry RD groups I-WIRE PACIFIC
LIGHTRAIL Internet-2 wave projects
?s, P2p fiber, gt10g ip Experimental
Developmental Networks
I2-Abilene, GigaPoPs Fed. NGI Nets.
Leading-Edge ip Services for demanding
apps . middleware
Operational High Performance Research Support
Networks
Internet-2, GigaPoPs I2-K20
Common Denominator, Advanced Services Research
Education Network
Advanced Services for RE users
Business,.Gov. General Public
ISPs, hosting .coms etc.
Commodity Internet
12/05/01
4
Some Specific Discontinuities/Gaps

• e2e physical connectivity!
• Diversity of Providers
• Persistence of key parts over enough time!
• Physical Latency Jitter Constraints
• Ultra high performance exchange points
• Configurability
• Measurement
• e2e performance
• Deterministic Behavior
• User control panels
• Repeatability/Predictability
• (vs. a shared net with many variables)
• Mismatched Technologies/Architectures
• (e.g. framing, channels, OSs etc.)
• Middleware (app, auth./security, QoS?)

Horizontal
Vertical
12/5/01
5
Some Specific Discontinuities/Gaps cont.

• Network aware/adaptive apps. OS
• Expertise/Operations (NOCs vs. Res. Groups)
• AUPs! Battling AUPs
• Isnt sufficiently responsive/customizable
• for some/many experiments, programs,
• projects, apparatus
• Hard to explore new techs.
• Doesnt well support net. experiments
• Interoperability between experimental
• operational general purpose RE
• Tyranny in success of IP-) But, will our
• current one size fits all path work?
• Goals, Plan, Mechanism for persistent core
• components cost effective provisioning of
• experimental nets, as and how needed

Vertical cont.
12/5/01
6
Physical

• e2e continuity ge, 10ge, ? or mix
• Last Interfaces
• Last Riser, Closet Horizontal
• Building ltgt Campus hubs
• local loop(s) Campus/Site lt-gt PoPs
• Intra Inter-regional segments
• hardware food chains (packets, waves )
• high performance exchange points, ? switching,
• wire provisioning techs.
• Rebuilding the key regional elements,
• again -(
• (in cases with owned fiber /or ? irus)
• Recalling what we knew about how hard
• this is and what the best paths/hubs are
• Hub/Ring Anchor points local build out
• For experimental/developmental nets

Gaps
12/5/01
7
Where the (reachable) Research is
Top 251 Research Universities in Total, Federal,
Peer Reviewed, Contract and Grant
Obligations1all gt 250 million/yr top few gt
500 million/yr
btw - LIGHTRAIL has 3 of top 5, half of top 10
(and maybe 6 of 11)
Plus major, reachable net oriented centers, labs
etc.
Johns Hopkins (w/APL) U. of Washington Stanford U.
of Michigan UCSD U. of Pennsylvania
UCLA MIT Harvard UCSF U. of Colorado (all)
Wisconsin Columbia Yale Cornell Minnesota Washingt
on U. Duke UNC
UCB Pittsburg Penn State USC Cal. Tech Illinois
1. Variations in how to count and financial
period causes minor variations in order and
inclusion at bottom end 2. Obligations lag
Awards the list of Awards is almost identical
but numbers are higher
draft 11/12/01
8
Most are locally interconnected,
ADDRESSING NETWORK DISCONTINUITIES
increasingly with the broadband even
private waves/fiber needed for e2e
experimental/developmental nets
DTF
Critical Mass Sites
Top 10 Res. Univ. Next 15 Res. Univ Key
Centers, Labs Intl. 10gig ?
12/5/01
9
10 gig. wave/ip base, adding ?s, p2p ?s etc.
L I G H T RA I L Start-up March 02
P A C I F I C L I G H T R A I L

• - Fiber based local loops to provide e2e waves
• Includes significant dark fiber test segments
• Plan to be 1st to 40 gig. waves

10 gbs
10 gbs
10 gbs
or
10 gbs
Priority PoPs
In Process
Stop Sought
Bleeding-Edge e2e Services facilities for key
researchers for next gen. tech., arch.,
Grids, content, sensors, apparatus, etc.
draft 12/03/01
10
P A C I F I C L I G H T R A I L
ADDRESSING NETWORK DISCONTINUITIES

• LONG LINES
• 10 yr/IRUs
• Options for more _at_ discount
• Could buy fiber (nationally)
• METRO
• Owned Fiber Rings/Segs. in
• SEA, SF BAY, LA, SD SAC
• PROVISIONABLE DEDICATED p2p WAVES
• ESTABLISHING KEY PIECES
• FOR FLEXIBLE LONG TERM LOW COST PROVISIONING!

P/NWGP-Fairbanks U. Alaska ARSC etc
Abilene?
R
P/NWGP
SEA
Peerings?
R
S
UW
OHSU/OGI?
PDX
R
SAC
CENIC ONI Tier 1 2
R
UCD
Peerings
UCB
SNY
R
EMV
R
Stanford SLAC
UCSF
SLO
or
R
UCSB
? Caltech ?
SB
R
DEN
Peerings
R
LA
CHI
NCAR?!
R
UCLA
ISI USC
I-WIRE !!!
ANH
NY
R
LANL??
NM
UCI
R
DC
SD
R
SDSC
draft 12/4/01
UCSD
11
Get critical mass connected? Cement keys?
ADDRESSING NETWORK DISCONTINUITIES

• Leverage Regional Connections
• Incent fatter/dedicated pipes
• Enable significant e2e
• Connect Scientists/Labs/Devices
• Establish MetaPop Centers

Critical Mass Sites
Top 10 Res. Univ. Next 15 Res. Univ Centers,
Labs Intl. 10gig ? Key Hubs
draft 12/4/01
12
What to do?

Program to establish facilities for responsively
provisioning customized and/or dedicated
experimental end-to-end networks including
extended backplanes and networking capabilities
providing 1) Custom networks dedicated to
serving specific needs of projects,
programs, experiments and apparatus etc.
and, 2) Networks which are themselves
experimental exercises. i.e. be facilities
that can provision ( instrument) any arbitrary
networks as needed
Lessons Successes (e.g., ARPAnet, early
NSFnet), Failures
12/5/01
13
Keep the Experimental Networks realm from
sliding down the network development Spiral
And, Long Term
Privatization
Commercialization
Todays Internet
Research Experimental Nets
NGI Operational RE efforts
Research and Development
Partnerships
Source Ivan Moura Campos
14
ULTIMATE GOAL

• So we can deliver networks which are fully
  responsive to, and meet the needs of important
  applications, projects, programs and users
• and
• Continue to innovate with networks and network
  technology in, and with the same spirit, that
  brought us the internet.