More tests on M5 throughput (well, our first :)

[kijoma]

Hi All,

We finally got our sweaty hands on some M5 bullets so thought it would be good to do some initial tests..

These are between two rooms (through some floors) using low gain aerials and the test setup to make things easier is we have the two Bullet M5's set up as follows :-

Latest rc firmware.. ,one as AP WDS, other as STA WDS. on the STA WDS unit we plugged it into an RB433 Routerboard (we will dig out an AH version in later checks). lobbed 3.28 code on the MT board and set its btest server up for no auth and 20 max connections. set it for dhcp on eth1 so we could talk to it , also turned conntrack off for max speed. the 433 can easily saturate 100mbps anyway.

BOTH bullets have the power set to full, hey if your going to test throughput, you need those Power amps running at their hardest / worse case distortion..

the AP WDS unit attaches to a switch here which in turn connects to an RB433AH board, also with 3.28 fw. (this is shared with my net connection so we will dig up a dedicated board later).

Bullets are on 5680 MHz, signal -69dBm between them, everything else set as factory (ssid = kijoma), bridge mode... airmax off..

first test is to see one way UDP as seen on the MT router...

http://arunvalley.net/bullet/UDPTX.PNGhttp://arunvalley.net/bullet/UDPTX.png

Yawn... it does that non stop, fully saturated ethernet basically, ccQ 100% on the bullets. observant people will note this test was done after the one below.

Next lets return fire, UDP RX to the 433ah. this time i stuck the MT result overlayed on the bullet screen..

http://arunvalley.net/bullet/UDPRX.png

equally boring... and nice ! (cant use smilies, uses up pic allowance..)

neither of the MT boards have any issue feeding ~100 mbps UDP..

the nice forum limits us to 4 pictures i now find... so will continue in the next message... pfft. yeah i know there are only 2, it says i have 4!

cheers

http://arunvalley.net/UDPTX.png

[kijoma]

part two...

now... TCP.. the real stuff! , TCP send using 20 connections (this shows the 433AH's higher CPU speed, even then it shows 100% cpu )

http://arunvalley.net/bullet/TCPTX.png

the speed maybe limited by the routerboards each end here.. especially as the speed is asymetric (300 mhz vs 680 mhz cpu's).

the throughput like the UDP is boringly flat and good !

Now lets see what happens if you ram TCP both directions (802.11a usually goes horrible when you do this, lets see what .11n does?).

http://arunvalley.net/bullet/TCPTXRX.png


Interestingly symetric ~35Mbps TCP when ramming data in both directions simultaneously.. i like it ! . this may be affected by the RB speeds so may not represent the capability of the M5's

Last test, i turned airmax on.. (note in UDP it still flatlines at full speed so didnt bother to show it..).

Single direction TCP as previous test.

http://arunvalley.net/bullet/TCPTXMAX.png

a bit faster than with airmax off, interesting... the short red burst shown on the MT graph was the other direction, the 300 mhz cpu of the other RB shows as a reduced speed there.. the purple bit was the earlier both ways test..

Now i know these two work i am going to leave them running doing flat out TCP duplex for some hours to see if they survive and don't crash etc..

cheers

[Ron]

thanks for sharing

[kijoma]

hi,

now the forums back, i left the two m5's running a 20 concurrent tcp connection duplex test all night and part way though this morning..

I set two passthrough firewall rules so i could count the data it passed, the radio links and ethernet sides of the M5 bullets remained solid throughput.

here is the data it pushed back and forth..
http://arunvalley.net/bullet/soaktest1.png

After this i did a load of other tests which show that playing with the settings is probably not a great idea at this stage of the FW development.

Issues with the UI and 5 MHz chan with Airmax on soon showed up.., the latter does not work.. The UI issues are i suspect down to similar quantity of fixable bugs the UI had in the early days of AirOS..

We tested the units AP WDS to AP WDS as a test, that worked as well as AP WDS to STA WDS which the tests we did above were set at..

We also tested ping with traffic, and as with all mediums the ping is actually fine as long as you do not saturate the link we set a one way TCP stream with 20 connections and limited it to 40 Mbps + Airmax, the ping wobbled around 1 to 8 ms when checked simultaneously.

if we saturated it then obviously the pings went through the ceiling...

Of course ALL real WISPs limit the throughput with QoS somewhere to prevent ugly link saturation, don't they? :icon_twisted: ... well we do at least :)

Basically in summary the bullet M5 works fine as long as you don't play with its settings too much yet :)

Now the distance limitation for 40 MHz mode i find more concerning, perhaps that can be explained as it is also present on the norma 5 products and experience with MT kit on long distance 40MHz 802.11a shows there is no practical reason why the ack cannot go higher than that restricted by AirOS ..

answers on a post card.. please..

now when are the rockets and the new 2x2 aerials going to arrive from my order!, i want to dabble..