[Madwifi-users] Throughput tests with madwifi and hostap/wpasupplicant

Jim Thompson jim at netgate.com
Fri Mar 24 21:53:29 EST 2006


On Mar 22, 2006, at 10:45 AM, Beat Meier wrote:

> Hi all
>
> I've done some throughput tests with iperf.
>
> Configuration
> **************
>   AP: wrap2c, SR5
>   Client: wrap2c, CM9
> Software: madwifi-ng-1457, hostapd/wpa_supplicant-0.5.2 using WPA-PSK
> Rate was set fixed to the value indicate on ap and client.
> This is done with 802.11a (5.825 GHz)
> This was done without long distance i.e. only 4m but values
> of cts etc. are set to 6000m!!
> iperf was done at least 15 times per rate and the values were very  
> equal
> (iperf has default tcp window size (client: 16 Kbyte, server: 85.3  
> KByte)
> This has not been in the past!! It seems that either madwifi (which is
> more probable) or hostap/wpa_supplicant has changed to the good  
> site!!!
> In the past I had "fluctuations" about 3-4Mbps now around 0.1Mbps
>
> Iperf server on cliente wrap
> Iperf client from gatway before AP ie.e
> gw -> ap -> client
>
>
> Testresults
> ***********
>
>  6M:  4.3 Mbps (28%)
>  9M:  6.6 Mbps (27%)
> 12M:  8.2 Mbps (32%)
> 18M: 12.6 Mbps (30%)
> 24M: 16.0 Mbps (34%)
> 36M: 21.6 Mbps (40%)
> 48M: 25.7 Mbps (46%)
> 54M: 29.2 Mbps (46%)
>
> The percentage is the overhead (loss of bandwidth) at this rate.
>
> I've seen a lot of mails about asking max. throughput on wrap and
> that's cpu limited to max. 20Mbps but this seems not to be true as  
> seen by
> 48M,54M.

I can get around 30Mbps on freebsd, too.

> Has anybody more better results??
>
> Now my question.
> Does anybody know because until 18M (24M) the protocol overhead or  
> whatever
> is about 30%
> and later there is going to 45%?
> Is it because other modulation? (BPSK,QPSK,16-QAM,64-QAM). Has anybody
> information
> which modulation is used at which rate?
> Retries were in the normal range (I think).

If I had to guess, you're dropping more packets at the higher  
modulation rates.  At 48Mbps and 54Mbps, you're typically limited by  
EVM, not SNR.

Jim




More information about the HostAP mailing list