Subject: Re: tinkering with wi interfaces (Prism 2.5)
To: Wojciech Puchar <wojtek@tensor.3miasto.net>
From: Jeff Rizzo <riz@boogers.sf.ca.us>
List: current-users
Date: 01/21/2004 12:54:21
On Wed, Jan 21, 2004 at 10:13:08AM +0100, Wojciech Puchar wrote:
> > I've read over the wi(4), ifconfig(8), ifmedia(4), and wiconfig(8)
> > man pages, and I'm trying to find how to hardwire certain
> > things in the new world order where most settings are done in
> > ifconfig rather than wiconfig.  (To give you a hint on the last
> > time I played around seriously with this stuff, it was about
> > the time the "wavelan" page on the netbsd website was originally written, and
> > *it* references 1.4G).
> >
> 
> general good ideas with WiFi links are:
> 
> 1) if stations aren't moving frequently, so there are same conditions most
> of time, turn off speed autonegotiation and set maximum speed that works
> reliably.

There's the rub;  I haven't figured out how to stop autonegotiation from
happening at some level;  I thought that setting it with "ifconfig wi0
media DS11" would do it, but it would appear that some (if not all)
frames are sent at lower speeds anyway.

> 
> 2) all antennas have to see each other, or "hidden node" problem produces
> lots of collision and retransmition

Yup, I'm aware of that.  Only two nodes, anyway.


> 
> 3) ad-hoc mode is generally fastest.

OK.

> 
> 
> i'm working on my own PtP and PtMP mode program that bypasses part of the
> 802.11 protocol sending everything as broadcasts (no ACKing this case with
> adhoc mode) and doing all air access control and retranmitions by myself.
> mostly deciding exactly when each node transmits data, instead of "try and
> maybe you are lucky" method in standard WiFi.
> 
> it's mostly for long distance outdoor network, but could be usable indoors
> too. of course unix only.

Which is exactly what I'm doing.  :)  The nodes are about 3 or 4 km apart.

> this eliminates most common problem - high packet loss where there are
> lots of connections and full speed transmisions. this case "try end error"
> algorithm of transmission have low chance of sending data without making
> collisions.

I'd certainly be interested in looking at anything you're doing in this
area...

+j
-- 
Jeff Rizzo                                         http://boogers.sf.ca.us/~riz