[Voyage-linux] Re: [Madwifi-devel] Real throughput tests at 4.4km p2p link

[Vytautas]

Hello,

  I wonder, what actual throughput do you get ? And for everybody using
WRAP's, what is the maximum throughput that you could achieve between 2
WRAP boxes (both on wired and wireless links)?

The reason I ask is that i can't get more than 50Mbit/s anyhow. I am
testing wired interfaces and wireless interfaces.
The connections are all routed (no bridging). All wired links are
100Mbps FD. Wireless link is 11a forced to turbo mode. No encryption at
all. RSSI is about 55-60, because the boxes ar about 2 meters apart ;)
I use iperf both on the WRAPs itself and on the boxes connected to them
via ethernet. Tuning iperf parameters (window size, bandwith, mss, etc.)
only adds ~1-2Mbits.
Below is ASCII network scheme:

                    +-------+            +--------+
+------+            |       |--wireless--|        |            +------+
|host A|--ethernet--|WRAP AP|            |WRAP STA|--ethernet--|host B|
+------+            |       |--ethernet--|        |            +------+
   |                +-------+            +--------+               |
   |                                                              |
   +---------------------------ethernet---------------------------+


The tests i performed were following:

------------------------------------------------------

1. WRAP to WRAP via wireless
WRAP AP -- wireless -- WRAP STA
WRAP STA -- wireless -- WRAP AP

Results: TCP: 26 Mbit/s
         UDP: 36 Mbit/s (without loss)

------------------------------------------------------

2. WRAP to WRAP via ethernet
WRAP AP -- ethernet -- WRAP STA
WRAP STA -- ethernet -- WRAP AP

Results: TCP: 34 Mbit/s
         UDP: 43 Mbit/s (without loss)

------------------------------------------------------

3. Host to WRAP
HOST A -- ethernet -- WRAP AP
HOST B -- ethernet -- WRAP STA

Results: TCP: 48 Mbit/s
         UDP: 48 Mbit/s (without loss)

------------------------------------------------------

4. WRAP to HOST
WRAP AP -- ethernet -- HOST A
WRAP STA -- ethernet -- HOST B

Results: TCP: 34 Mbit/s
         UDP: 44 Mbit/s (without loss)

------------------------------------------------------

5. End to end w/o wireless
HOST B -- ethernet -- WRAP STA -- ethernet -- WRAP AP -- ethernet -- HOST A
HOST A -- ethernet -- WRAP AP -- ethernet -- WRAP STA -- ethernet -- HOST B

Results: TCP: 34 Mbit/s
         UDP: 49 Mbit/s (without loss)

------------------------------------------------------

5. End to end involving wireless
HOST B -- ethernet -- WRAP STA -- wireless -- WRAP AP -- ethernet -- HOST A
HOST A -- ethernet -- WRAP AP -- wireless -- WRAP STA -- ethernet -- HOST B

Results: TCP: 26 Mbit/s
         UDP: 34 Mbit/s (without loss)

------------------------------------------------------

6. Host to host (non-traffic-free link)
HOST B -- ethernet -- HOST A
HOST A -- ethernet -- HOST B

Results: TCP: 95 Mbit/s
         UDP: 96 Mbit/s (without loss)

------------------------------------------------------

From the results it seems, that the main limitation is WRAP boxes.
Ethernet forwarding speed is <50MBit/s, ethernet traffic generation
speed <35Mbit/s. The wireless link is limited to 26-27Mbit/s.
Also the differrence in bandwith pattern between iperf server being AP
or STA is minimal.
I would be very interested in similar measurements' results, so if
anyone has done some work on that, I would greatly appreciate if you
replied. Wireless link speed is of the greatest interest, because our
main goal is to have as fast Wireless link as possible using WRAP.
If you have any questions about the details, I'll gladly answer any of them.

Regards,
Vytautas

Beat Meier wrote:

>unky wrote:
>
>  
>
>>Hi Beat,
>>
>>Did you use vmstat or others in your test to see if the WRAP boxes are
>>CPU bound (i.e. 0% CPU idle)?  I have experience that runing iperf
>>client and server would consume all your CPU power.  So in my
>>performance test, I don't run iperf on the measured boxed anymore. 
>>Instead, I use wget to remote download a big file from another local
>>server.
>> 
>>
>>    
>>
>I have about 50% cpu load
>I have also done the test with connected pc with iperf and scp but with a
>ftp, scp, wget
>the limiting factor is the disk (even worse a cf)! I know the new disk are
>fast but
>do a test with an old disk ...
>For bandwidth test you should really use iperf or something similar.
>BTW: No difference in tests on wrap or with connected pc
>
>Greetings
>
>Beat
>
>  
>