[plug] High speed WiFi

[Brad Campbell]

Stephen Boak wrote:

> I have recently installed a netgear WG311 802.11a/b/g PCI wireless
> card in my server (Debian Woody, 2.4.22 kernel), and it seems to be
> working faultlessly. The other end is currently a matching netgear
> WG511 PCMCIA card in a indows laptop. Both cards are supposed to be
> 54mMbps (802.11g) at top speed, but I haven't tested the throughput
> or done any optimisation yet. Windows/Samba file sharing is at least
> equal to my wired lan (10Mbit half duplex?) over about 20 metres. 
> 
> The WG311 (PCI) card is based on an Atheros 5212 controller chip,
> supported (mostly) by the madwifi driver set,
> http://sourceforge.net/projects/madwifi/
> madwifi needs at least kernel version 2.4.20.
> 
> Ad-hoc networks are not supported in the madwifi drivers, but running
> the WG311 in 'master' mode seems to achieve the same results.
> 
> I have not yet set up the WG511 PCMCIA card under Linux, I believe
> (hope) it is using the better supported PrismII chipset.
> 
> I can supply more notes on my configuration and a few gotchas along
> the way if you decide to use these cards.

That is the kind of info I'm after, thanks. Benchmarks would be good 
when you get a chance. Even if it's just sucking down an .iso with wget 
I'll get an idea of raw trasnfer rate :p)

I have an idea I'm working on at the moment to use a kernel mode 
encapsulated packet driver, as on the air 802.11b requires an ack for 
every packet sent.
For example I send a tcp packet over the network. It leaves the server 
and hits the client. The wifi card on the client sends an ack back to 
the wifi card on the server to ack the packet. The client's network 
stack then generates the ack to the tcp packet and sends it back to the 
server, the servers wifi card then sends an ack back to the clients card.
So for every packet sent, there are two packets on the air. Now in 
theory when a packet leaves destined to the broadcast address the wifi 
architecture does not require acks on the air. So my theory is to 
encapsulate all outgoing packets and send them using a broadcast MAC 
addr, and do the packet filtering in the kernel module. This will 
increase the cpu usage on the nodes, but should reduce the bandwidth 
overhead on the air, thus allowing more data to be sent..

I'll be working on this over the next couple of weeks, will be 
interesting to see how it goes anyway.

The more I get into the low level of 802.11b the more I realise it's 
just a bastard child of ethernet. Urgh!

Brad


Brad


_______________________________________________
plug mailing list
plug at plug.linux.org.au
http://mail.plug.linux.org.au/cgi-bin/mailman/listinfo/plug