[ath9k-devel] ath9k (ad-hoc?) problems in compat-wireless trunk as of sept 6/2

[ath9k-devel] ath9k (ad-hoc?) problems in compat-wireless trunk as of sept 6/2

[Nicolás Echániz]

We are testing current wireless in QuintanaLibre and we are finding some strange behaviors.

The first symptom we have been able to reproduce is this:

Given nodes A and B which are neighbors with a good link quality (-68 signal, 15Mbps throughput) and a third node C which has a comparable link to A and a low signal link to B. When we turn of C's wireless the latency between node A and B will be extremely high (in the thousands of ms) for almost one minute.

In this example, nodes A and B are marisa and nogal. And node C is gerylu (which we turn off 200 sec after the ping starts).

http://pastebin.com/MzWSCDRb


the moment when the ping times return to normal is coincidential with the iw event (as seen from marisa -Node A- ):

wlan0 (phy #0): connection quality monitor event: peer 92:f6:52:c6:00:ed didn't ACK 50 packets
wlan0: del station 92:f6:52:c6:00:ed

92:f6:52:c6:00:ed   <-- this is gerylu (node C)


We are observing an overall instability in the network which consists of 40 nodes (100% ad-hoc) some of which hear up to 16 neighbors in the "iw station dump", so this scenario may be occurring frequently.

Lastly, we also observe that some nodes are not associating on one of their interfaces (they are dual band), although they did associate correctly before the upgrade.



We will keep some nodes with this version for a couple of days in case someone (Felix?) want's to log in and do some live testing.

Guido will add some more info from the tcpdumps we collected if it's relevant.


Cheers,
Nico

[ath9k-devel] [OpenWrt-Devel] ath9k (ad-hoc?) problems in compat-wireless trunk as of sept 6/2

[Nicolás Echániz]

I have kept testing this and was able to reproduce the same problem
every time.
The performance perception when the network is at "rush hour" is very
unstable so I tried to reproduce traffic conditions during late night.
I sent multiple netperfs through different routes and while the netperfs
were running turned of one node. Latency climbed to the thousands and it
took a while for it to stabilize again.

Then I turned off another node, latency went sky high again for some
seconds until I lost conectivity and could monitor no more.

I waited more than 15 minutes for the net to be reachable again from my
local node. (Nodes are automatically reset when they cannot reach the
network gateway for more than 15 minutes)

All ping tests are done on fe80 local IPv6 addresses so no routing
protocol is involved in case someone's wondering. Also, from what I see
(iw station dump), routers are associated but can send no data.

After 15 minutes of waiting, this is what station dump was showing from
my local node:
root at nogal:~# iw wlan0 station dump
Station 64:70:02:3d:85:0a (on wlan0)
	inactive time:	40 ms
	rx bytes:	175693
	rx packets:	2642
	tx bytes:	0
	tx packets:	0
	tx retries:	0
	tx failed:	0
	signal:  	-69 dBm
	signal avg:	-65 dBm
	tx bitrate:	1.0 MBit/s
	rx bitrate:	1.0 MBit/s
	authorized:	yes
	authenticated:	yes
	preamble:	long
	WMM/WME:	yes
	MFP:		no


station dump shows no tx data at all...

and of course no ping

root at nogal:~# ping6 fe80::6670:2ff:fe3d:850a%wlan0
PING fe80::6670:2ff:fe3d:850a%wlan0(fe80::6670:2ff:fe3d:850a) 56 data bytes
From fe80::92f6:52ff:febb:ec58 icmp_seq=1 Destination unreachable:
Address unreachable
From fe80::92f6:52ff:febb:ec58 icmp_seq=2 Destination unreachable:
Address unreachable



I'll try to keep this nodes on this version to help debug, but the
network is in production, so I can't do this for long without getting
complaints :)

Cheers.

Nico.


El 07/09/13 22:40, Nicol?s Ech?niz escribi?:
> 
> We are testing current wireless in QuintanaLibre and we are finding some strange behaviors.
> 
> The first symptom we have been able to reproduce is this:
> 
> Given nodes A and B which are neighbors with a good link quality (-68 signal, 15Mbps throughput) and a third node C which has a comparable link to A and a low signal link to B. When we turn of C's wireless the latency between node A and B will be extremely high (in the thousands of ms) for almost one minute.
> 
> In this example, nodes A and B are marisa and nogal. And node C is gerylu (which we turn off 200 sec after the ping starts).
> 
> http://pastebin.com/MzWSCDRb
> 
> 
> the moment when the ping times return to normal is coincidential with the iw event (as seen from marisa -Node A- ):
> 
> wlan0 (phy #0): connection quality monitor event: peer 92:f6:52:c6:00:ed didn't ACK 50 packets
> wlan0: del station 92:f6:52:c6:00:ed
> 
> 92:f6:52:c6:00:ed   <-- this is gerylu (node C)
> 
> 
> We are observing an overall instability in the network which consists of 40 nodes (100% ad-hoc) some of which hear up to 16 neighbors in the "iw station dump", so this scenario may be occurring frequently.
> 
> Lastly, we also observe that some nodes are not associating on one of their interfaces (they are dual band), although they did associate correctly before the upgrade.
> 
> 
> 
> We will keep some nodes with this version for a couple of days in case someone (Felix?) want's to log in and do some live testing.
> 
> Guido will add some more info from the tcpdumps we collected if it's relevant.
> 
> 
> Cheers,
> Nico

[ath9k-devel] [OpenWrt-Devel] ath9k (ad-hoc?) problems in compat-wireless trunk as of sept 6/2

[Felix Fietkau]

On 2013-09-09 1:29 PM, Nicol?s Ech?niz wrote:
> I have kept testing this and was able to reproduce the same problem
> every time.
> The performance perception when the network is at "rush hour" is very
> unstable so I tried to reproduce traffic conditions during late night.
> I sent multiple netperfs through different routes and while the netperfs
> were running turned of one node. Latency climbed to the thousands and it
> took a while for it to stabilize again.
> 
> Then I turned off another node, latency went sky high again for some
> seconds until I lost conectivity and could monitor no more.
> 
> I waited more than 15 minutes for the net to be reachable again from my
> local node. (Nodes are automatically reset when they cannot reach the
> network gateway for more than 15 minutes)
> 
> All ping tests are done on fe80 local IPv6 addresses so no routing
> protocol is involved in case someone's wondering. Also, from what I see
> (iw station dump), routers are associated but can send no data.
I've committed some fixes that might be related to this issue. Please
test current OpenWrt trunk.

Thanks,

- Felix