<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META http-equiv=Content-Type content="text/html; charset=iso-8859-1">
<META content="MSHTML 6.00.6000.16640" name=GENERATOR>
<STYLE></STYLE>
</HEAD>
<BODY bgColor=#ffffff>
<DIV><FONT face=Arial size=2>
<DIV><FONT face=Arial size=2><FONT face="Times New Roman"><FONT size=3>
<DIV><FONT face=Arial size=2><FONT face="Times New Roman"><FONT size=3>dear
welzl,<BR><BR>referring to the section <FONT
style="FONT-FAMILY: times new roman,serif"><FONT
style="FONT-FAMILY: times new roman,serif">'3.2 Challenge 2: Corruption Loss'
</FONT></FONT></FONT></FONT><FONT face="Times New Roman" size=3>of the draft, i
think that an important point is missing here, which is the basic idea
introduced at Mobiocm 2001 in the TCP Westwood paper. </FONT></FONT></DIV><FONT
face=Arial size=2><FONT face="Times New Roman" size=3></FONT>
<DIV><FONT face="Times New Roman" size=3><BR>The key concept of westwood tcp is
the following:<BR></FONT></DIV>
<DIV><FONT style="FONT-FAMILY: times new roman,serif" size=3>- when a packet
loss is detected through 3 dupack (or timeout) set the congestion window equal
to <BR>the measured bandwidth at time of congestion. this measurement is
obtained by counting and filtering the ack rate. this setting provides
<BR>a significant goodput improvement in lossy channels, i.e. when packet are
lost not due to congestion.</FONT></DIV><FONT
style="FONT-FAMILY: times new roman,serif"
size=3></FONT></FONT></FONT></FONT></FONT></DIV>
<DIV><FONT face=Arial size=2><FONT face="Times New Roman"><FONT size=3><FONT
face=Arial size=2><FONT style="FONT-FAMILY: times new roman,serif"
size=3></FONT></FONT></FONT></FONT></FONT> </DIV>
<DIV><FONT face=Arial size=2><FONT face="Times New Roman"><FONT size=3><FONT
face=Arial size=2><FONT style="FONT-FAMILY: times new roman,serif" size=3>The
reason why this setting works very well in lossy channel is that, since losses
are not due to congestion, <BR>the "blind" by half window reduction <BR>of
New Reno is avoided, i.e. the window is not over-shrinked.</DIV>
<DIV><FONT face=Arial size=2></FONT><BR><BR>We have run thousands of tests
over real networks using New Reno, Cubic and Westwood and we have found
that all <BR>basically provide similar performance except in the case of lossy
links where tcp westwood provides a neat improvment due to control window
shrinking <BR>based on measured available bandwidth. <BR><BR>On the other hand
it is well-known that using croos-layer signalling is not a working solution due
to inter operability, indeed it violates the fundamental design principle of
layering.<BR></FONT><FONT style="FONT-FAMILY: times new roman,serif"
size=4><BR><FONT size=3>best,<BR>saverio</FONT>
</FONT></DIV></FONT></FONT></FONT></FONT></FONT></DIV></BODY></HTML>