Chapter 11.4

END-TO-END ISSUES

Optical Internet

Optical technology Protocol translates availability of giga

bit bandwidth in user-perceived QoS

WDM network architecture

WDM wide/metropolitan area network As an Internet backbone

Access network Ex: campus network

Domain border gateway High-speed IP router

WDM network architecture

Open issue

Access network Packet loss and delay due to

congestion Optical backbones

High transmission rates Extremely low bit error rates

Bridging the gap between access and backbone network is an issue

End-to-end TCP

It is not practical, the reasons is below TCP slow-start algorithm constrains very

large bandwidth available in the lightpath until the steady state is reached

Socket buffer is not enough congestion and flow control features

needed in access network

Split TCP connection models (1/2) It is an evolutionary approach to the TCP end-to-

end model to adapt to the specific characteristics of each of the network segment

It is not an efficient solution for optical networks due to the wavelength speed

EX: 10-Gbps wavelength bandwidth

10-ms propagation delay bandwidth delay product (BDP)=25MB File sizes in the Internet are smaller than such BDP Because of different round-trip delays, it is difficult

to optimize TCP windows to achieve transmission efficiency

Split TCP connection models(2/2)

Most TCP features( congestion and flow control) unnecessary in optical network Extremely low loss rate in the optical

network makes retransmission unlikely to happen

The optical network can operate in a burst-switched mode in the optical layer, so there are no intermediate queues in which overflow occurs

11.4.1 TCP for High-Speed and Split TCP Connection

First approach to solving the adaptation problem between access and backbone TCP connection can be split in the

optical backbone edges TCP extensions for high speed

a larger transmission window no slow start

Files over lightpaths (FOL) Files are encapsulated in an optical

burst in order to be transmitted across the optical network.

Simulation Topology

An optical channel(1Gbps) which connects several access routers located at the boundaries of optical networks

Network parameters Namely, link capacities, propagation delay,

and loss probability Performance metric

Connection throughput

Simulation

Simulation Result For small file sizes

The connection duration is dominated by setup time and slow start, which does not allow the window size to reach a steady-state value

For large file sizes The TCP reach steady state and the

throughput is equal to window size divided by roundtrip time.

Such behavior is expected in a large BDP network, in which connections are RTT-limited rather than bandwidth-limited

11.4.2 Performance Evaluation of File Transfer (WWW) Services over WDM Networks(1/2)

11.4.1 is in error-free condition Ex: a first-generation WDM network (static lightp

ath between routers) second-generation WDM network suffer blo

cking probability Limited number of wavelengths Burst dropping due to limited queueing space in

photonic packet switchs Split TCP becomes inefficient

Performance Evaluation of File Transfer (WWW) Services over WDM Networks(2/2)

In FOL, files are encapsulated in an optical bursts through the optical backbone using a simple stop-and-wait protocol for error control.

Assuming the setup of an optical burst takes RTT/2

Simulation Result

TCP congestion avoidance limits transfer efficiency

This serve to illustrate that the throughput penalty imposed by the TCP congestion control mechanism is rather significant

The main difference between a simple FOL protocol and TCP is in interpreting congestion TCP considers loss is produced by queuei

ng overflow FOL is aware that loss is due to blocking