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

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

9.2. Preferential Flow H<strong>and</strong>ling Using DiffServ 225<br />

under 90% AF allocation still show serious problems with the cwnd dynamic<br />

which consequently results in reduced goodput performance.<br />

The main difference between ScalableTCP <strong>and</strong> HS-TCP flows are that the<br />

occurrence of the slow start like events for HS-TCP are much lower, making<br />

the effects of the congestion moderation much less noticeable. As such,<br />

the HS-TCP cwnd dynamics are much cleaner <strong>and</strong> nearer to the theoretical<br />

traces than those of Scalable TCP. However, having congestion moderation<br />

turned off has less overall effect on the cwnd dynamics when AF is allocated<br />

900Mbit/sec - regardless of the TCP algorithm used.<br />

The inference is that even without the implications of congestion moderation,<br />

SACK processing is still a dominant factor which causes cwnd dynamic<br />

deviations from theory that result in the observed decreased goodput performance.<br />

When the AF allocation is 700Mbit/sec, it was observed that the<br />

dominant factor is congestion moderation - as the removal of the function<br />

call completely resolves the problem of deviations of the cwnd dynamic from<br />

theory. But when the AF allocation is 900Mbit/sec, the flux of SACKs is so<br />

high that removing the congestion moderation code only partially resolves<br />

the problems of cwnd dynamics entering low values.<br />

9.2.5 Active Queue Management<br />

In this section, the negative effects of SACK processing are mitigated using<br />

an active queue management solution. The objective was to lower the rate<br />

of SACK events after multiple drops <strong>and</strong> therefore allow the sending host to<br />

process the SACK Blocks appropriately.<br />

Neither St<strong>and</strong>ard TCP nor HSTCP experienced as many slow-starts as<br />

ScalableTCP. Therefore, an AQM solution was implemented for ScalableTCP

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

Saved successfully!

Ooh no, something went wrong!