Optical Burst Switching CSIT 560 Internet Infrastructure: Switches and Router Prepared By Eric Lo Date: 6 Dec 2005

Agenda Introduction OBS Concept Other research Basic Motivation Burst Assembly Just Enough Time (JET) Scheduling Content Resolution QoS Potential problem Other research R&D activities in Japan Dual header OBS Synchronized OBS Layered Architecture

Introduction Circuit Switching Packet Switching Bandwidth inefficient of bursty data (internet traffic). Frequent long circuit setup Waste bandwidth during off/low-traffic periods Packet Switching A packet contains a header (e.g., addresses) and the payload (variable or fixed length). Each node needs to buffer, process the header, and send it to the next hop. Statistic sharing of link BW among packets with different source/destination (Statistical multiplexing)

Introduction (Bursty traffic) Left: Poisson traffic (voice) smooth at large time scales and mux degrees Right: data (IP) traffic, bursty at all time scales and large mux degrees circuit-switching not efficient (max >> avg)

Motivation Problems: Solution: Explosive traffic growth and bursty Traffic pattern Lack of optical buffer (RAM) for packet switching. Fiber delay lines (FDLs) are bulky and provide only limited & deterministic delays Need fast processing power of header and payload Solution: Find a method between circuit switching and packet switching. To increase the Bandwidth efficiency. That’s Optical Burst Switching comes into the picture. Data traffic growth still doubling every year

Agenda Introduction OBS Concept Other research Basic Motivation Burst Assembly Just Enough Time (JET) Scheduling Content Resolution QoS Potential problem Other research R&D activities in Japan Dual header OBS Synchronized OBS Layered Architecture

OBS Concept Burst Assembly (and Disassembly) at Edge client data (e.g., IP packets) assembled into bursts Burst Switching/Reservation Protocol Control packet (CP) sent an offset time t ahead of burst Dedicated control channel (out-of-band signaling) for CP No fiber delay lines (FDLs) nor O/E/O conversions for burst at any intermediate (core) nodes Photonic Burst Switching Fabric inside Core Leverages the best of optics (for burst switching) and electronics (for CP processing and fabric control)

Packet (a) and OBS(b) switching

Optical Burst Switching Node Multiple data channels share one control channel. Data bursts remain in optical domain while CPs go through O/E/O conversions

Burst Assembly – Step 1 Control channel Data channel Time or length threshold is reached Control channel Assembly queues for different egress nodes Data channel Burst Assembly Node ATM Cell SONET Frame IP Packet

Burst Assembly – Step 2 Control channel Data channel A CP is generated and sent out Control channel Assembly queues for different egress nodes Data channel Burst Assembly Node ATM Cell SONET Frame IP Packet

Burst Assembly – Step 3 Control channel Data channel Assembly queues for different egress nodes Data channel Burst Assembly Node ATM Cell SONET Frame IP Packet

Just-Enough-Time (JET) An offset time between Control packet(CP) and burst No fiber delay line (FDL) required to delay the burst when CP is processed and switch fabric is configured. CP carries the burst length information Facilitates delayed reservation (DR) for intelligent, efficient allocation of BW and FDL (if any), including look-ahead scheduling.

JET – Step 1 OOO OOO OEO OEO CP arrives the OEO node at time t1 Offset = T OOO OEO OOO OEO

JET – Step 2 OOO OOO OEO OEO CP goes through O/E conversion and configure switch fabric OOO OEO OEO OOO

JET – Step 3 OOO OOO OEO OEO CP goes through E/O conversion and leaves O/E/O node at time t1+ OOO OEO OOO OEO

JET – Step 4 OOO OOO OEO OEO When burst arrives at the intermediate node, the switch fabric is already configured

JET – Step 5 OOO OOO OEO OEO Offset = T-  Without any delay, the burst goes through the optical switch fabric

JET – Offset time Control packet can leave right after d = D – s (s is the switch setting time)

Delayed Reservation (DR) DR leads to efficient allocation of BW and any available FDLs (though not shown). Without DR, 2nd burst will be dropped in both cases (and FDLs will be wasted in Case 2).

Burst Scheduling In general approach, it is to schedule incoming bursts in the order of the CP arrivals. It leaves several free time intervals between the scheduled reservations called void. Without void filling (only use open interval such as LAUC). With void filling (Can minimize starting and ending void such as LAUC-VF)

Burst Scheduling

Burst Scheduling – New approach Ordered Scheduling is to schedule bursts in the order of the burst arrivals instead of header arrivals. It places incoming reservations in a buffer and defers the scheduling until just before the actual burst arrives. Able to remove the negative effect of header arrival dynamics.

Ordered Scheduling

Content Resolution (1) When multiple bursts compete for the same output channel, how to avoid/reduce burst loss? Three major strategies Deflection in space, time and wavelength Preemption of an existing reservation Segmentation of a burst into smaller pieces

Content Resolution (2) Deflection Segmentation Space domain: applying deflection routing Wavelength domain: use a different wavelength via wavelength conversion Time domain: wait using a fiber delay line Segmentation Drops, deflects or preempts one or more segments instead of an entire burst

QoS Different offset time is assigned to different service classes over JET OBS networks so as to provide differentiated services in terms of burst loss probability for classes of different priorities. Difficult to maintain the same offset time in all routers. May starve lower priority bursts because higher priority bursts always have more opportunities to make wavelength reservation It may be unfair to long bursts of low priority because it is hard to find a long gap to serve a long burst of low priority in an almost full schedule table.

QoS – New approach (PWRP) Preemptive Wavelength Reservation Protocol (PWRP) Each class is associated with a predefined usage limit. Each switch maintains a usage profile for a class per output link and monitors the current usage of each class.

QoS - PWRP

Potential Problem Tends to have a high blocking probability May require an uneconomically large increase in network transmission capacity.

Agenda Introduction OBS Concept Other research Basic Motivation Burst Assembly Just Enough Time (JET) Scheduling Content Resolution QoS Potential problem Other research R&D activities in Japan Dual header OBS Synchronized OBS Layered Architecture

Other Research R&D activities in Japan

Other Research Dual header optical burst switching (DOBS)

Other Research Synchronous Optical Burst Switching (SOBS)

Other Research A Layered Architecture for Supporting Optical Burst Switching

References OBS Forum tutorial slides by Chunming Qiao R. Parthiban, C. Leckie, A.Zalesky, AV.Tran*, Does Optical Burst Switching have a role in Core Network ? Farid, Vinod, Joel, A Layered Architecture for Supporting Optical Burst Switching Sami Sheeshia, Chun Qiao, Synchronous Optical Burst Switching Neil Barakat and Edward H Sargent, Dual Header Optical Burst Switching: A New Architecture for WDM Burst-Switched Networks Ken, etal., Photonic Network R&D Activities in Japan Wanjiun Liao, Chi-Hong Loi, Providing Service Differentiation for Optical-Burst-Switched Networks

Thank you!