Re: [ns] TCP traffic generation rate

[Brian Lee Bowers <bowersb@lanl.gov>]

On Tuesday 31 July 2001 18:05, Sunil Gowda wrote:
> Thanks Brian,
>     Sorry for not having provided enough information. Here are the
> details
> I am trying to study TCP performance over an optical link. So the
> bandwidth typically is about 2.5 -10Gbps/channel and we have 32-128
> channels.
> Taking the least case i will need 2.5Gbps*32 = 40Gbps.
> The link delay is 100ms.
> So at what rate will i have to generate traffic to saturate the
> link and what should be the window size?

If you need to have 40Gbps with a 100ms RTT, then you need to have 
4Gb in flight during any given RTT to keep the pipe full.  (40Gbps * 
100ms/RTT == 4Gb / RTT).  If you use uniform, 1500 byte packets, then 
you need a window of 333,333 packets to keep the pipe full
	(4Gb/RTT) / (1500B/pkt * 8b/B)
==>	4Gb/RTT / 12Kb/pkt
==>	1/3 Mpkt/RTT


All of this assumes that you aren't crossing a bottleneck link that 
will constrain your throughput.

-- 
Brian Lee Bowers	|	RADIANT Team (Summer Intern)
bowersb@lanl.gov	|	Los Alamos National Laboratory