05.08.2014 Views

An Investigation into Transport Protocols and Data Transport ...

An Investigation into Transport Protocols and Data Transport ...

An Investigation into Transport Protocols and Data Transport ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

6.1. Survey of New-TCP Algorithms 115<br />

3500<br />

50<br />

1<br />

cwnd <strong>and</strong> ssthresh (packets)<br />

3000<br />

2500<br />

2000<br />

1500<br />

1000<br />

500<br />

cwnd<br />

ssthresh<br />

0<br />

0 50 100 150 200 250 300<br />

Time (seconds)<br />

α<br />

40<br />

30<br />

20<br />

10<br />

α<br />

β<br />

0<br />

0<br />

0 50 100 150 200 250 300<br />

Time (seconds)<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

β<br />

(a) Congestion Window<br />

(b) AIMD Parameters<br />

Figure 6.2: cwnd dynamic of ScalableTCP (Single flow on Dummynet, link capacity<br />

200Mbit/sec, RTT 150ms, queue size 500 packets.)<br />

recommended values of α = 1<br />

100 <strong>and</strong> β = 7 8<br />

are used.<br />

In order to provide compatibility with St<strong>and</strong>ard TCP, a variable W Switch is<br />

defined to determine the mode switch point from St<strong>and</strong>ard TCP behaviour<br />

to Scalable TCP behaviour.<br />

It is defined in terms of the cwnd, so that<br />

when cwnd ≤ W Switch St<strong>and</strong>ard TCP is invoked. The recommended value of<br />

W Switch = 16 segments is used.<br />

Figure 6.2 shows the cwnd evolution of ScalableTCP as a function of<br />

time. The small decrease of cwnd upon packet loss guarantees high throughput<br />

for individual flows. Meanwhile, the multiplicative increase means that<br />

approximately a constant number of RTTs would be required to reach the<br />

value of cwnd just before congestion.<br />

The scaling property of ScalableTCP means that it implements an exponential<br />

increase in the sending rate when no loss is detected. This kind<br />

of increase, with the associated multiplicative decrease, is known as Multiplicative<br />

Increase Multiplicative Decrease (MIMD). [CJ89] states that MIMD<br />

based algorithms do not converge to fairness in drop-tail environments.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!