Re: [KISMET] [PATCH] Support for prism2 hostap drivers (fwd)


From: Pedro Estrela (pedro.estrela_at_inesc.pt)
Date: 2002-07-27 14:22:50 UTC


> PS. As a totally unrelated asside, I recall seeing a huge thread on jumbo
> frame support to speed up gig-ethernet cards. With standard packet sizes,
> interrupt calling was killing the CPU. Would it be possible to use
> something like jumbo frames to increase speeds on 802.11b? It looks like
> the headers are HUGE (2304 total for tcpip frame size of 1500? Is that
> right?).

The main 802.11 problem on high speeds is the level 1 header stuff that includes syncronization (just like livel 1 ethernet of Manchester encoding) and important information to decode the higher levels like the speed of the packet (1Mbit/s, 2, 5.5, 11 Mbit/s). The problem is that to support multirate stations on the same 802.11 cell and to solve the "hidden station problem" also on multirate, this level 1 header info must be sent at the minimum speed... always. Calling by head, i've seen that the level 1 header equals only 24 bytes, BUT transmitted at 1Mbit speed. comparing it to 11Mbit operation, it then equals 264 bytes PP. Altough ethernet also has this syncronization, it is always operates at 10 Mbit/s speed. :-o
Then consider that each and every single packet must be acknolodged, it doubles this figure for 528 worth of 11 Mbit/s bytes for each packet. :-O
Then add other stuff like DCF exponentional backoff, SIFF/PIFF/DIFF time waiting, level 2 MAC header size, lost packets that are retransmitted until they pass or reach N tries, station collisions, CPU latency interrupt time, etc. (the list continues...), to explain the actual measured rates on 802.11b that go nowhere near the 11Mbit/s mark... :-( )

check this excel study that i've made earlier for check the maximum measured rates for different 802.11b MTU, to simulate from voice calls (100 bytes pp), MPEG1/4 Video, and FTP (1500 bytes pp). the actual measures are explained on the .xls file.


back to your tought, of course that jumbo frames would be great, but it would introduce incompatibilities of hardware/firmware level (802.11 defines MTU somewhere around 2300 bytes mark), and could kill voice latency for high priority voice traffic.

Pedro Estrela




This archive was generated by hypermail 2.1.4.