Transcript 20040720-Panel

Regional Network Future
Architectures
Joint Techs July 20,2004
TFN/OARnet
[email protected]
Legacy Networks
• Provisioned Circuits
• Many built in the mid to late 90s ATM based
• Looking for a convergence of services
– Ip, voice video
• ATM originally was faster than alternatives
• Attempted to implement telco like signaling
UNI etc.
• Used pseudo circuit switching.
• Could run virtual over lay networks as
separate entities.
Motivations for Dark Fiber
Network
– Reduce long term costs
• It appears that long term costs can be contained by
making large upfront capital investment
– Allow dedicated light paths through regional fabric
• Research community believes that Lambda based
service provides resource that other mechanisms don’t
e.g. enhance research environment.
– Allow interaction with other Gigapops and entities
in a dedicated fashion
Motivations for Dark Fiber
Network (continued)
• Provide better service by having total
control.
• A belief that we now control our destiny
– Sounds good but is it true?
– What about whims of state budgets?
Conflicting Requirements
• Build new network to save money
– Often cited as an incentive
• Build network to support on demand High
Capacity Circuit (Lambda) Switching
– Some progress in “tunable” devices but it still
requires deploying amps, DCU to support under
utilized infrastructure
• Allow for new collaborations e.g. be all things
to all people
Conflicting Requirements
(Continued)
• States have invested Millions
• Higher Ed has driven it, but there is
pressure to add “equity partners”
• Desire to converge different types of
data and networks
– Ip, Voice, Video, ATM, RPR etc.
– Sound familiar?
What will drive Architecture is
Policy as well as Technology
• Political, Social, and Economics will drive as
much as technical requirements
• Belief that Lambda switching will prevail is not
very real because of economics in the short
term
• Major investment to benefit the few. In Higher
Ed research grants might be used to finance
early work.
• Since the shared resource must be
scheduled it will require many users with
non-conflicting temporal needs at a minimum.
Analogy
• Current notions of schedulablity are
much like operators manually plugging
patch chords to complete circuit
Reality Check
What are we doing?
• Alternatives such as MPLS, HOPI are much
more realistic in the short term
• DWDM will drive convergence of various
state entities to share same fiber plant
• The trick will be
– How involve other communities and get them to
pay fair share while….
– Higher Ed stays in the drivers seat
Future Architectures
• DWDM shared infrastructure in the short term
• Longer Term, Lambda (circuit switching) will
be real when
– 1. Cost comes down to deploy fully provisioned
infrastructure
– 2. Switching and demand patterns can be
modeled using Erlanger like statistical analysis,
e.g. an optical switch that is almost a “Phone
switch” in time scale for switching (Microseconds)
and utilization is predictable enough to allow
scalable over provisioning