Well, this is TCP dynamics! High speed connections will suffer more; even
more connections with high bandwidth and long delay paths will deliver
degraded performance. There are several studies on TCP fairness that are
dealing with similar situations. Also, fairness issues arise when
different enchanced TCP flavors with different underlying transport
technologies are competing for the link.
________________________________________
Pambos Charalambous, MSEE
Network Engineer
Sprint, Broadband Switch Engineering
phone#(816)854-5445
fax#(816)854-5068
[email protected]
-----Original Message-----
From: Tang Oo [SMTP:[email protected]]
Sent: Friday, October 23, 1998 1:53 AM
To: [email protected]
Subject: fairness of TCP
Assume that all the IP routers don't use fair queueing scheme. In case of
buffer overflow, the arriving packets are discarded randomly. The window
size will be reduced by half, if a packet of TCP connection is lost.
Comparing a connection with 64k bytes window and another one with 4k bytes
window, the window of the former will be reduced by 32k, but the latter
only
2k, if there is loss of both connections. The arrival rate of high speed
connection is higher than a low speed one, so the chance of multiple
losses
is much higher for the high speed one. That is, it is more probable for
the
window of high speed connection to be dropped multiple times. Is it fair
for
it?
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:51 EST