On Sun, Jun 08, 2003 at 01:16:33PM -0400, [EMAIL PROTECTED] wrote:
> Is it true
> that tcp over a 3 hop mesh network link generally works at 15% of expected
> throughput (due to tcp's poor performance in the face of packet loss and
> latency).

  I think it is reasonable to expect less than 2Mb/s in one direction
  over three 802.11b hops, even before TCP overheads, if you use one
  radio per node. This is simply because 802.11b is half-duplex. If you
  add one or two more hops, you will see some overhead that CSMA adds to
  a multihop network, the term for which I think is the "exposed node
  problem." Add in unfairness in the 802.11 MAC, near-far problems,
  and so on, and you see that 802.11b is a crummy radio technology for
  building multihop, ad-hoc networks. That does not keep some of us from
  still trying, though!

  I think you can triple that speed by using two 802.11b radios
  attached to two omni antennas, provided that you tune the radios to
  non-interfering channels and you make the right choice of routes,
  HOWEVER

  1) The right routes are essentially the ones that minimize
  co-interference of TCP packets on the same path. No distance vector
  algorithm will be able to do that. I think that a link-state algorithm
  can do an adequate job at it.

  2) I am not at all certain that two 802.11b radios tuned to different
  channels will not STILL interfere with each other, even if they are
  tuned to opposite ends of the spectrum, when their antennas are on
  the same mast. Can anyone say anything about that?

Dave

-- 
David Young             OJC Technologies
[EMAIL PROTECTED]      Urbana, IL * (217) 278-3933
--
general wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless

Reply via email to