2013 09-22, transport sdn and ecoc, ping pan
DESCRIPTION
Transport SDN in ECOC'2013TRANSCRIPT
SDN in Transport/IP Networks?!
Ping Pan, Ph.D. September 2013
The Paradigm Shift: the separation of control-plane and data-plane
Applications and Services
Physical Storage Physical Processors Physical Networks
Physical Storage Physical Processors Physical Networks
Physical Storage Physical Processors Physical Networks
Physical Storage Physical Processors Physical Networks Physical Networks
Virtualized Networks
OR… IS THIS FOR REAL?
Can make network BIGGER…
Can make service deployment FASTER…
Can make network operation SIMPLER…
Can virtualize network devices and make them CHEAPER
Incumbent: my SDN is BETTER than yours!
Confusion… over Transport SDN
Bigger? Optical transport networks can get bigger without SDN
Fact: Wavelength demand has increased 500% since 2004, while revenue growth is far less Claim #1: SDN is about scaling up networks (packet, optical, enterprise, data center and backbone) as a whole. Considering optical transport network SDN in isolation makes no sense!
Faster? Existing network architecture is the bottleneck for slow service deployment
Claim #2: Sorting out inter-networking interfaces as “NNI’s” makes no sense. We need to virtualize the network as a whole, and run IP services in an Over-The-Top (OTT) manner
Simpler? Don’t look for standard bodies to help you
IETF: • The routing infrastructure is
getting more complex each everyday
ITU: • OTU/ODU is not for dummies
neither
ONF: • Multi-tables in OpenFlow…
really?
Claim #3: We are not making networking simpler through standard bodies
Virtualization in storage and computing: • Definition: Take commodity x86 servers, represent them as uniform resources,
and utilize them for efficiency
• Goal: efficiency in utilization
• Means: commodity hardware
Reality in networking: • High-performance networking ASIC’s are all different in behavior and properties
• Transport gears don’t need to inter-op (including date encoding) to get deployed in Tier1 networks
• The driver behind transport efficiency is in packing advanced functions and features, and has little to do with commodity hardware
Claim #4: Virtualizing of network components makes no sense. The name of the game is in virtualizing the networks in terms of throughput, delay and protection capability, and provide programming interfaces.
What does VIRTUALIZATION mean?
Incumbent’s Notion of SDN: …make sure it’s called OpenXYZ
Router Networks
SDN
OpenFlow? Why stop there? Add more
Claim #5: Router vendors think the router networks are not complex, slow and expensive!
Why cannot we do this?
Vendor X
Vendor X
Vendor X
L2 Switch
Vendor Y
Vendor Y
Vendor Y
Metro DC Metro DC Long-Haul
Controller
Application
10GigE
10GigE
Physical Network
10GigE
10GigE
Virtualized Network Via simple interfaces
A
B
C
D
Virtualized Network by Application -- backend load balancer
P1
P2
P3
P4
P5
P6
P7
P8
10GigE
10GigE
Controller
A
B
C
D (ABCD, P1, P7, BW)
P1 P7
Application
P1
P2
P3
P4
P5
P6
P7
P8
10GigE
10GigE
Controller
A
B
C
D (ABCD, P1, P7, BW)
P1 P7
Load Balancer
Data Center App’s
VM distribution
P1
P2
P3
P4
P5
P6
P7
P8
What if… • We scale up the transport network
What if… • We virtualize the network as whole
• We run IP services on top
What if… • We make the IP services 1/10 of the existing price, and sell
to 10x of the customers
Claim #6: It’s not What-If. It’s WHEN
The Last Claim for the Day