High Performance
Networking Group
Nick McKeown
Departments of Electrical Engineering
and Computer Science
High Performance
Switching and Routing
Telecom Center Workshop: Sept 4, 1997.
nickm@stanford.edu
http://www.stanford.edu/~nickm
PhD Students
Shang-Tse Chuang, Nandita Dukkipati,
Yashar Ganjali, Sundar Iyer, Isaac Keslassy,
Pablo Molinero Fernandez, Rui Zhang.
Recent Alum: Pankaj Gupta, Adisak Mekkittikul
1
Some Past Projects
Input-queued switches
1.

Theorems
•
•


First proof of 100% throughput for input queued switches.
First proof of emulating output queueing with speedup = 2.
Algorithms: iSLIP, iLQF, LPF, iLPF, …
Prototyping: The Tiny Tera 320Gb/s MPLS switch.
IP Lookup and Packet Classification
2.

Algorithms: 24-8-DIR, RFC, HiCuts.
Packet Buffer Design
3.

Ping-pong buffers.
2
Current Projects
Integrating optics and routing
1.

Optics in Routers.

Circuit switching at the core of the Internet.
Isaac Keslassy, Shang-Tse Chuang.
With Professors Mark Horowitz, David Miller and Olav Solgaard.
Pablo Molinero Fernandez
Load-balancing and parallelism
2.



Parallel packet switches. Sundar Iyer.
Fast packet buffers. Sundar Iyer.
Shared memory switches. Sundar Iyer and Rui Zhang.
Network Algorithms
3.




Congestion control for very short flows. Rui Zhang and Nandita Dukkipatti.
Crossbar scheduling algorithms. Isaac Keslassy.
Small packet buffers. Isaac Keslassy.
Multipath routing protocols. Yashar Ganjali.
3
Incorporating Optics
Into Routers
High Performance
Switching and Routing
Telecom Center Workshop: Sept 4, 1997.
Isaac Keslassy, Nick McKeown
E-mail: keslassy@stanford.edu
Optical Router Project: http://klamath.stanford.edu/or/
4
Why We Need Faster Routers
To prevent routers from becoming the bottleneck
Packet processing Power
Link Speed
1000
10000
2x / 18 months
1000
2x / 7 months
100
100
10
10
1
1
1985
1990
1995
2000
1985
1990
1995
Fiber Capacity (Gbit/s)
Spec95Int CPU results
10000
2000
0,1
0,1
TDM
DWDM
Source: SPEC95Int & David Miller, Stanford
5
100 Tb/s Optical Router

Collaboration


4 Stanford professors (M. Horowitz, N. McKeown, D.
Miller and O. Solgaard), and their groups
Objective


To determine the best way to incorporate optics into
routers
Push technology hard to expose new issues
• Photonics, Electronics, System design

Motivating example: The design of a 100 Tb/s Internet
router
• Challenging but not impossible (~100x current systems)
• It identifies some interesting research problems
6
Research Problems

Linecard


Architecture


Memory bottleneck: Address lookup and packet buffering
Arbitration: Computation complexity
Switch Fabric



Optics: Fabric scalability and speed
Electronics: Switch control and link electronics
Packaging: Three surface problem
7
Two-Stage Switch Architecture
External
Inputs
Internal
Inputs
1
1
12
External
Outputs
1
1
2
N
N
Load-balancing
cyclic shift
N
Switching
cyclic shift
100% throughput for broad range of traffic
types (C.S. Chang et al., 2001)
8
WGR (Waveguide Grating Router)
A Passive Optical Component
l11, l12 …l1N
l11
Linecard 1
l12
Linecard 2
l1N
Linecard N
NxN
WGR
Wavelength i on input port j goes to output
port (i+j-1) mod N
 Can spread and shuffle information from
different inputs

9
WGR Based Solution
Detector
Fixed Laser/Modulator
l1
Linecard 1 l2
lN
l1
Linecard 2 l 2
lN
l1
Linecard N l2
lN
1
1
l 1, l 2
1
…l N
l 2 1 , l2 2
2
…l N
N
N
l 1, l 2
N
…l N
1
N
l 1, l 2
2
…l N
2
l 1, l
3
NxN WGR …l N
N
1
2
l1
l 2 Linecard 1
lN
l1
l2
Linecard 2
lN
N-1
l 1, l 2
1
…l N
l1
l2
Linecard N
lN
10
Conclusion: why this WGRbased solution is appealing
 In
one word: passive
 No
arbitration
 No reconfiguration
 Low power
 Reliability
11
Backup slides
12