[B.A.T.M.A.N.] about batman performance

[B.A.T.M.A.N.] about batman performance

[zb1981gm]

I have used batman on my project for several months.It performs very well,stable especially! 
But,actually,I am not satisfied with the performance of Batman.I found that the bandwidth would drop sharply with the wireless hop increase. 
Now I provide my test result here: 

routerA and routerB is composed of compex wp543ahv mainboard (AR7161 platform) and two AR9220 wireless card.Firmware is Openwrt backfire. 
one card is working on 2.4G channel as AP mode,the other is working on 5G channel as mesh backbone  
batman version:2.0 
notebook1 and notebook2 is equiped with 802.11n usb wireless card which access to the AP mode card of routerA and routerB 

            channel 1                 channel 36              channel 11 
notebook1 ---------------  routerA --------------  routerB -------------- notebook2 

I use iperf to test bandwidth 

notebook1 ----------- routerA    45Mbps 
routerA   ----------- routerB    46Mbps 
routerB   ----------- notebook2  45Mbps 

But the bandwidth between notebook1 and notebook2 remains 8Mbps 
bandwidth between notebook1 and routerB remain 18Mbps 

So the conclution is that the bandwidth will be halfed with the 1 hop increase! right? 

I also found that the interface alternating mode and bonding mode doesn't take effect 

              channel 1 
routerA  ------------------  routerB 
              channel 36 

The test result is no difference whatever I configured alternating or bonding or just single radio 

According to the batman official site, alternating or bonding mode would improve the performance compared with the single radio mode 

looking forward  somebody's help 
highly appreciated!

Re: [B.A.T.M.A.N.] about batman performance

[Simon Wunderlich]

[-- Attachment #1: Type: text/plain, Size: 2935 bytes --]

Hello zb1981gm (btw, do you have a real name?),

On Tue, Dec 06, 2011 at 05:01:13PM +0800, zb1981gm wrote:
> I have used batman on my project for several months.It performs very well,stable especially! 
> But,actually,I am not satisfied with the performance of Batman.I found that the bandwidth would drop sharply with the wireless hop increase. 
> Now I provide my test result here: 
> 
> routerA and routerB is composed of compex wp543ahv mainboard (AR7161 platform) and two AR9220 wireless card.Firmware is Openwrt backfire. 
> one card is working on 2.4G channel as AP mode,the other is working on 5G channel as mesh backbone  
> batman version:2.0 

What is batman version 2.0? We have batmand (layer 3), but I guess you are using batman advanced
as you talk about bonding/interface alternating later. These have 0.X (for old releases) or 201X.Y.Z
version schemes (since 2 years). Please clarify.

> notebook1 and notebook2 is equiped with 802.11n usb wireless card which access to the AP mode card of routerA and routerB 
> 
>             channel 1                 channel 36              channel 11 
> notebook1 ---------------  routerA --------------  routerB -------------- notebook2 
> 
> I use iperf to test bandwidth 
> 
> notebook1 ----------- routerA    45Mbps 
> routerA   ----------- routerB    46Mbps 
> routerB   ----------- notebook2  45Mbps 
> 
> But the bandwidth between notebook1 and notebook2 remains 8Mbps 
> bandwidth between notebook1 and routerB remain 18Mbps 

Have you compared this to static routing?

> 
> So the conclution is that the bandwidth will be halfed with the 1 hop increase! right? 

It should not, as different radio modules are used (and correctly tuned on different channels).
However, you may have a limitation in your CPU power or something else. 
You may want to try static routing to find this out.

> 
> I also found that the interface alternating mode and bonding mode doesn't take effect 
> 
>               channel 1 
> routerA  ------------------  routerB 
>               channel 36 
> 
> The test result is no difference whatever I configured alternating or bonding or just single radio 

To make bonding work, there are a few conditions:
 * routers must be equipped with multiple radios which can connect to each other
 * activate bonding (batctl bonding 1)
 * make sure the transmit qualities are similar (check with batctl o), should not be more than a difference of 50

If everything is fine, you can get a boost of approx. 50%.


> 
> According to the batman official site, alternating or bonding mode would improve the performance compared with the single radio mode 
> 

Yup, it should help. Alternating is not helping in your scenario as far as I understand (your
mesh is only 1 hop long, as notebooks are connected via infrastructure mode). Bonding may
help as described above.

best regards,
	Simon



[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: [B.A.T.M.A.N.] about batman performance

[Wuchi]

Hi zb981gm,

we had a (somewhat) similar performance issue with our setup some time ago. We found out that a setup of two n-Cards in a single room decrease performance heavily, if both cards are under load at the same time. Forcing the cards into b/g mode did not help, so changing the cards to b/g hardware solved the problem for us.

May had to do with those fallback window override issues some n-Cards have, but not sure. In general, don't expect too much of an performance increase with n-Hardware on a USB 2.0 interface.

May be worth a try to look at first before fiddling around with the protocol, IMHO.

- Wuchi

Am 06.12.2011, 10:01 Uhr, schrieb zb1981gm <zb80211@gmail.com>:

> I have used batman on my project for several months.It performs very well,stable especially!
> But,actually,I am not satisfied with the performance of Batman.I found that the bandwidth would drop sharply with the wireless hop increase.
> Now I provide my test result here:
>
> routerA and routerB is composed of compex wp543ahv mainboard (AR7161 platform) and two AR9220 wireless card.Firmware is Openwrt backfire.
> one card is working on 2.4G channel as AP mode,the other is working on 5G channel as mesh backbone
> batman version:2.0
> notebook1 and notebook2 is equiped with 802.11n usb wireless card which access to the AP mode card of routerA and routerB
>
>             channel 1                 channel 36              channel 11
> notebook1 ---------------  routerA --------------  routerB -------------- notebook2
>
> I use iperf to test bandwidth
>
> notebook1 ----------- routerA    45Mbps
> routerA   ----------- routerB    46Mbps
> routerB   ----------- notebook2  45Mbps
>
> But the bandwidth between notebook1 and notebook2 remains 8Mbps
> bandwidth between notebook1 and routerB remain 18Mbps
>
> So the conclution is that the bandwidth will be halfed with the 1 hop increase! right?
>
> I also found that the interface alternating mode and bonding mode doesn't take effect
>
>               channel 1
> routerA  ------------------  routerB
>               channel 36
>
> The test result is no difference whatever I configured alternating or bonding or just single radio
>
> According to the batman official site, alternating or bonding mode would improve the performance compared with the single radio mode
>
> looking forward  somebody's help
> highly appreciated!

Re: [B.A.T.M.A.N.] about batman performance

[zb1981gm]

> Hello zb1981gm (btw, do you have a real name?),
I am sorry,you can call me Ice_zheng

> What is batman version 2.0? We have batmand (layer 3), but I guess you are using batman advanced 
> as you talk about bonding/interface alternating later. These have 0.X (for old releases) or 201X.Y.Z 
> version schemes (since 2 years). Please clarify. 

It's Batman_adv 2011.2.0

> Have you compared this to static routing?
So, you think the test result is abnormal,right?
As regard to the IP address configuration of routerA and routerB,
I bridge the eth0,wlan0 ,wlan1 all together,so routerA and routerB  has only 1 IP address respectively

                                                     batman
                  channel1                           channel 36                            channel 11
notebook1(wlan0) -------- (wlan0)routerA(wlan1) -------------------- (wlan1)routerB(wlan0)---------- (wlan0)notebook2
10.10.10.1                10.10.10.2                                  10.10.10.3                     10.10.10.4

you suggest static routing,means without batman routing,just use ad-hoc mode,than add static route on each node,right?

notebook1(wlan0)          routerA(wlan0)            routerA(wlan1)            routerB(wlan1)           routerB(wlan0)           notebook2(wlan0) 
10.10.10.1                10.10.10.2                20.20.20.1                20.20.20.2               30.30.30.1               30.30.30.2

then add next hop gateway at each node(route add .....)
that is what you mean?
If no batman routing protocol,It certainly should has a better performance 
actually,my test result shows that(single radio mesh,if relay node are equiped with dual radio worked on interface alternating,it would be better)
if via 1 hop,bandwidth has x
then via 2 hop,bandwidth remain x/2
then via 3 hop,bandwidth remain x/4
.......


> It should not, as different radio modules are used (and correctly tuned on different channels). 
> However, you may have a limitation in your CPU power or something else. 
> You may want to try static routing to find this out.

I check the cpu(ar7161) loading (using top utility),less than 30% ,memory still remain 16M, so it has nothing to do with cpu loading and memory consumation
As regard to the channel assignment of different wireless card,I still have some puzzle.
I originally think that no interference should be available if radios are configured on non_overlap channel such as 1,6,11.
but,actually not.interference is still available though radios are configured on channel 1 and channel 11
                    channel 1
routerA(wlan0)  ------------------  routerB(wlan0)
                    channel 11
routerA(wlan1)  ------------------  routerB(wlan1)

if start iperf only on wlan0 or wlan1 singlely , bandwidth has 50Mbps almost.
if start 2 iperf simultaneously,both drop sharply

you must think that interference should be absolutely  not available at 2.4g channel and 5g channel.
but still has a little, because I change channel 11 to channel 36,if start 2 iperf simultaneously,bandwidth still drop
when I turn down the txpower of wlan0 or wlan1 ,drop degree would be better.
so channel interference is still a problem,OMHO.But I can't fix that except turn down the txpower.
do you have some advice?
the two radio's antenna distance is about 10cm

by the way, have you ever used the atheros 9220 wireless minipci card,the best bandwidth you have ever test is how much?

the ar9220 wireless card is 802.11n 300Mbps ,so I think 100Mbps should be available at least 

and It looks as if HT40 mode doesn't take effect, performance has almost no difference with HT20 mode

> To make bonding work, there are a few conditions: 
>  * routers must be equipped with multiple radios which can connect to each other 
>  * activate bonding (batctl bonding 1) 
>  * make sure the transmit qualities are similar (check with batctl o), should not be more than a difference of 50 

> If everything is fine, you can get a boost of approx. 50%. 

I promise that all the conditions you mentioned are met.but no boost with bonding mode

> Yup, it should help. Alternating is not helping in your scenario as far as I understand (your 
> mesh is only 1 hop long, as notebooks are connected via infrastructure mode). Bonding may 
> help as described above.

I try the 3 point scene test,alternating did take effect

                 channel1                           channel 36                            channel 11                       channel 36
notebook1(wlan0) -------- (wlan0)routerA(wlan1) -------------------- (wlan1)routerB(wlan0)---------- (wlan0)routerC(wlan1)------------(wlan1)notebook2

routeA-wlan0:AP 
routeA-wlan1:mesh
routerC-wlan1:AP
routerC-wlan0:mesh

notebook1 iperf routerA:40Mbps
routerA iperf routerB:36Mbps
routerB iperf routerC:40Mbps
routerC iperf notebook2:40Mbps

when routerB is configured as single radio for batman
notebook1 iperf routerC : 18Mbps
notebook1 iperf notebook2:9Mbps
when routerB is configured as dual radio work on interface alternaing(2.4g and 5g alternating)
notebook1 iperf routerC : 25Mbps
notebook1 iperf notebook2:9.5Mbps

so my conclution is still that,if batman_adv work on single radio router,bandwidth would be halfed when 1 hop increase,if the relay node equiped with dual radio work on interface alternating,it would be better a little

by the way,I still have another 2 problems,could you give me some advice
1.is there some approach to add static route for batman,configure nexthop forcely ,just ignore the  TQ value,I want to measure the bandwidth performace via wireless hop exactly
2.could you give me some advice about testplan for about 20 nodes mesh network (besides ping,iperf,ftp...).


best regards,
ice_zheng
2011/12/7

Re: [B.A.T.M.A.N.] about batman performance

[Sven Eckelmann]

[-- Attachment #1: Type: text/plain, Size: 503 bytes --]

Hi,

I tried to read your mail to find out what you did and why you did it, but I
failed miserably. We wanted that you test a setup without batman-adv that uses
static routes between nodes in adhoc mode. This would be help us to find out
whether batman-adv is the reason for the bandwidth drop or your hardware. And
the static routes should help to recreate the multi-hop tests.

Maybe you can restructure the explanations of your setup (and which results you
saw with each setup).

Kind regards,
	Sven

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]