Linux

Hi,

Am trying to use TCP as transport for a system that can support large number of connections where all of them potentially can send data at the same time. I came across ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=643715&tag=1 that says

TCP's ability to share a bottleneck fairly and efficiently decreases as the number of competing flows increases. TCP's minimum congestion window of one packet is the source of these problems: it causes a few flows to send too fast while the rest wait in re-transmission time-out.

This is an old link (~1997). Just wondering if the latest TCP stack in linux (2.6.32) has similar problem or has this been addressed of late ?

Thanks,
Nirmal

The problem is not the implementation, but the standard definition. TCP/IP is - intentionally so - defined without any quality-of-service in mind. This means for a regular, standard-adhering TCP/IP stack it is not possible to do this - all the possible partner stations wouldn't understand it and not honor it.

This was one of the reasons other network stacks (like ATM for instance) were being invented for backbones.

I hope this helps.

bakunin