OBS: Academic Hype or Future Internet Transport

1
OBS Academic Hype or Future Internet Transport?

• CUBIN Seminar
• Panel Discussion Series

2
Key Question

• What is the motivation for OBS?
• How will it provide improved performance?
• What will be the performance gain?

3
Some food for thought

• Network design is inherently a multi-layer
  discipline, design any one layer at your peril

4
Agenda

• Key components of an OBS network
• Jolyon White
• Modeling Performance
• Hai Le Vu
• Blocking rates in OBS using Deflection Routing
• Andrew Zalesky
• Discussion Strengths and weakness
• Where should OBS go
• Ubiquitous? Access? Metro? Core?
• What is the motivation for it?
• How will it improve our lives?

5
A Brief History of OBS

• Jolyon White

6
The Spectrum of Optical Networking
Low Multiplexing Efficiency
High Multiplexing Efficiency
Optical Packet Switching
Optical Burst Switching
Optical Circuit Switching
(wavelength routing)
7
Some Motivation

• IP traffic is bursty
• Optical Circuit Switching
• Resources under-utilized
• Cant cope with demand fluctuations
• Optical Packet Switching
• Current technology inadequate
• Better suited to access/metro networks
• Optical Burst Switching
• Achieve high utilization with existing photonic
  technologies

8
Network Architecture
Edge Routers (IP)
Core Switches (OBS)
OBS Network
9
Burst Assembly
Data Bursts
Packets assembled into bursts
OBS Network
Edge Router
IP Packets
Resource reservation signaling

• Advantages
• Switch burst as single unit ? can switch with
  existing technology
• Still achieve good multiplexing efficiency

10
Reservation Protocols Circuit Switching
Source
Destination
11
Reservation Protocols Just Enough Time
Source
Destination
Offset Time
Burst Data Transmission Time
12
Channel Allocation Algorithms
Channel (wavelength)
t
Control Packet Arrives
Burst to be allocated
13
Channel Allocation Algorithms (2)
Channel (wavelength)
t
Control Packet Arrives
Burst to be allocated
14
Some Motivation
Some Academic Hype?
Some

• IP traffic is bursty
• Optical Circuit Switching
• Resources under-utilized
• Cant cope with demand fluctuations
• Optical Packet Switching
• Current technology inadequate
• Better suited to access/metro networks

15
Some Demotivation

• OBS trades dedicated resource reservation for low
  latency
• Consequence bursts can be dropped!
• But Utilization is higher
• To overcome loss client must retransmit
• Customer experiences delay
• Network operator has higher carried traffic
• Do we really need high utilization?
• Current Internet backbone utilization at 40

16
Modelling and Performance Analysis of OBS
Networks

• Hai Vu

17
Switching Schemes in Optical Networks
JET OBS, OBS2w/Signaling
OCS
S
D
1 H-1
time
18
A Set of Fixed Point Equations

• Burst blocking probability at each switch depends
  on its offered workload
• The offered workload to each switch depends on
  the blocking probability at other switches

Translates into the set of equations
19
A Set of Fixed Point Equations

• Burst blocking in link j

• Normalized offered load to link j

where is service rate in link j
20
Reduced Load
21
Performance Evaluation Tool
Wavelength Conversion N/F/L (options)
Routing Policy Fixed, Deflection . . .
Performance Delay, Blocking Probability,
Modul. Analytical Perf. Eval. Tool
Edge/Core Router/Switch diff. models
Topology
OBS with Priorities, QoS
Traffic Model
OBS Scheme JIT,JET,BS,WR-OBS
Single Node Performance Model
Source M. Z., CUBIN/OIRC Workshop, 03
22
Questions/Comments
23
RFC 3366 Advice to link designers on link
Automatic Repeat reQuest (ARQ)

• Section 4 Conclusion
• A lower rate of packet loss generally benefits
  transport protocols and end host applications.
• This reduces impact on applications, allows
  more rapid growth of TCP's congestion window
  during slow start, and ensures prompt reaction to
  end-to-end protocol exchanges (e.g.,
  retransmission, congestion indications).
  Applications using other transport protocols,
  e.g., UDP or SCTP, also benefit from low loss and
  timely delivery.

24
TCP performance on lossy links

• Padhye, J., Firoiu, V., Towsley, D. and J.Kurose,
  "TCP Throughput A simple model and its empirical
  validation", SIGCOMM Symposium on Communications
  Architectures and Protocols, August 1998.