Transcript Presentation Slides (powerpoint)

A novel switching paradigm
for buffer-less WDM networks
Myungsik Yoo and Chunming Qiao
EE and CSE Departments
University at Buffalo (SUNY)
Why WDM Optical Internet ?
•
•
•
•
•
•
Explosion of the Internet traffic
Advances in DWDM networking technology
Emergence of Terabit routers with OC-48 line speed
Reduced capital equipment and operating costs
Current incarnation: IP routers over WDM links
Envisioned Next Generation Optical Internet:
– IP over an all-optical WDM layer
Existing switching paradigm
• Optical circuit switching
– set-up delay = round-trip time
– a limited number of wavelengths and thus limited connectivity
– not suitable for bursty traffic and short-lived sessions
• Optical packet/cell switching
– a payload and its header are sent together (tightly coupled)
– need to buffer the payload while processing the header
– high control overhead due to small packet/cell size
• Optical Burst Switching (OBS)
– combines the best of the two while avoiding their shortcomings
Comparison of three paradigms
Optical
switching
paradigms
Bandwidth
Utilization
Latency
(setup)
Circuit
Low
High
Optical
Buffer
Not
required
Packet/Cell
High
Low
Low
OBS
High
Proc./Sync.
Adaptivity
Overhead
(traffic & fault)
(per unit data)
Low
Low
Required
High
High
Not
required
Low
High
OBS based on Just-Enough-Time
Unique Features
• Loosely coupled bursts and their control packets
– uses an offset time + out-of-band control
• Offset time :
T 
H
   , where H is number of hops to go
k 1
– A burst is buffered only at the source node
• Delayed Reservation (DR) of BW at node i
– reserved from tb to tb+l (offset time at node i :
i
T(i)  T   
• Efficient utilization of BW (as well as FDLs if any)
k 1
)
BW Utilization vs Latency
• Comparing JET and NoDR
OBS for IP over WDM
• Run IP on top of WDM switches
• Use a dedicated control wavelength between the IP entities
– for routing control packets
• To send data, a control packet first sets up a connection
– one-way reservation based on offset time and DR
• Then, the data burst cuts through intermediate IP entities
– reduces latency & processing load at IP routers
• Similar to multi-protocol label switching (MPLS)
Support QoS at the WDM Layer
•
•
•
•
•
•
•
•
IP provides only best-effort services
A WDM layer supporting basic QoS is useful
E.g. to let Class 1 have a higher priority than Class 0
Existing approaches (e.g. Fair Queueing) require buffer
Challenge: support priority in a buffer-less WDM layer
Solution: assign Class 1 bursts an extra offset time
Class 1 bursts can reserve BW much in advance
Class 0 bursts can only “buy tickets at door”
Extra Offset Time for Priority
Differentiated Services
• Assume: the length of Class 0 bursts is exponentially
distributed with an average of L
– L may be a few Kbits or a few microseconds at OC-48
• At least 95% of Class 0 bursts are shorter than 3L
– and at least 99% of them are shorter than 5L
• So, if the extra offset time = 3L, at least 95% of Class 0
bursts will not block a Class 1 burst
• Blocking probability of Class 1 bursts will be reduced
– average blocking probability (over all bursts) unchanged
QoS Performance Improvement
• Blocking probability after 6 hops (offered load = 0.8)
Conclusion
• Optical Burst Switching (OBS) is a novel switching
paradigm for the Next Generation Optical Internet
• A buffer-less WDM layer can support basic QoS
• Future work on supporting multiple classes
• Also to interwork with other approaches to achieve
absolute vs. relative QoS guarantee