ARP, how fast can you go?

[Jim Carter]

On Mon, 23 Sep 2002, Septiaji Eko Nugroho wrote:
> My problem is I want to transfer small amount of data (only about
> 2000 bytes) from one station to another station in Ad-Hoc mode. But I
> have to finished everything, from host discovery to ending the
> communication, in less than a second.
>
> Actually I don't have any hesitation that Wireless LAN can handle
> that (ftp-ing at best condition only takes less than 6 ms). But my
> problem is the IP connection building, the ARP, normally I found ARP
> is periodic for each 1 s. Can we change this value, and how small
> interval can we make? Several ms? Because even the ARP is periodic,
> say 0.5 s, the TCP/IP stack will be useless for this application.

Try reducing the beacon interval and/or disabling power saving mode, at
both ends.  If I understand this right, I think the cell leader (like
access point in Ad-Hoc mode) has to have its radio on all the time, but the
clients are allowed to turn off their radios until the beacon is expected
from the cell leader.  The cell leader is expected to announce, with the
beacon, "you've got mail", and send the queued packet.  This will be a
duplicate if the partner had its radio on; I imagine there's some heuristic
to suppress that case, e.g. if the cell leader notices an 802.11 ACK of the
packet it's holding.  I may be mixing up features of Managed and Ad-Hoc
mode (this is from memory), but Ad-Hoc mode is not just "Aethernet"; the
cell leader actually does some work for the partnership.

Once a NIC sees traffic for itself, it has to keep its radio on for the
whole beacon interval, so there's a reasonable chance that the whole
conversation can be finished without queueing at the AP or cell leader.

Someone mentioned static ARP, which is a good idea. Someone else mentioned
pinging the partner occasionally, which will keep the ARP table primed, and
also will preload the nameserver cache, if that's relevant. But both of
these assume that at least one of the two machines knows the identity of
the other in advance -- not an unreasonable requirement.

James F. Carter          Voice 310 825 2897    FAX 310 206 6673
UCLA-Mathnet;  6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA  90095-1555
Email: jimc at math.ucla.edu    http://www.math.ucla.edu/~jimc (q.v. for PGP key)