On 8 Nov 2007, Xiaochuan Shen wrote:
> Dear Dave, > > Thank you very much for your answering. > > > >Bandwidth limitation doesn't impact buffering directly .. it works by > >delaying the send of packet N+1 by the transfer time computed for packet > >N. > > > This seems to me that nistnet leaves the queue length in an unconfigurable > manner. It buffers up the packets that awaiting processing automatically > while without a pre-set maximum queue length. > That is my impression, but I'm not positive since I've not explicitly searched the docs and code. > Is there a way to set pre-defined maximum queue length or buffer size in > NIST Net? I know that Dummynet has "pipe length" configuration which can > represent the buffer size effectively. Is there a parallel in NIST Net? and > if not, is it feasible to do so with further development / code > modification. I might be interested in doing this.. NistNet already has the ability to drop packets so there is no functional reason why packets can't be dropped with additional code. I would probably implement three controls: a. Max Queue time b. Max packet count queued c. Max Bytes in Queue The idea is that if too much data is sent to flow over a slow link, in the real world, a router would drop packets .. NistNet should simulate that behavior. For sure, once the delay for a TCP packet is longer than the Ack timeout, a packet is a candidate for drop ... perhaps some amount longer than the time. Dave Morris > > Many thanks, > Xiaochuan > > >On Thu, 8 Nov 2007, Xiaochuan Shen wrote: > > > >> Hello group, > >> > >> > >> > >> I am just wondering if anyone got any idea how much buffer NistNet uses > >> (or in what way it determines how much buffer should be used) when > >> bandwidth limitation is applied. > >> > >> > >> > >> > >> > >> This is what I implemented: > >> > >> > >> > >> > >> > >> server ----------------(100mb link) ---------------- NistNet box > >> ------------------(100mb link) --------------------- Client > >> > >> > >> > >> > >> > >> I am transferring bulk TCP traffic from server to client (a huge file > >> using FTP transfer). > >> > >> > >> > >> If I apply: > >> > >> Cnistnet server client -bandwidth 187500 > >> > >> > >> > >> Then I observed the "queue length" is around 50 packets. > >> > >> > >> > >> If all of these packets are full packet of size 1500Byte, a queue > >> length of 50 packets will introduce 50 * 1500 / 187500 = 400 > >> milli-seconds delay in my configuration. Is this value realistic in a > >> real network? If not, why nistnet use such a length of buffer? > >> > >> I can understand nistnet has to buffer a number of packets before it > >> can process the coming traffic but is this queue length configurable in > >> nistnet? > >> > >> > >> > >> Thanks for anyone looking and answering! > >> > >> > >> > >> Regards, > >> > >> Xiaochuan > >> > >> > >_______________________________________________ > >nistnet mailing list > >nistnet@antd.nist.gov > >http://www-x.antd.nist.gov/mailman/listinfo/nistnet > > > _______________________________________________ nistnet mailing list nistnet@antd.nist.gov http://www-x.antd.nist.gov/mailman/listinfo/nistnet
