[ath9k-devel] Multiple interfaces with one card

[ath9k-devel] Multiple interfaces with one card

[Hakan Coskun]

Hi,

we are doing research in order to study the behaviour of virtual wifi interfaces with channel switching. After a simulative study, we started to perform some real world experiments with ath9k. Running multiple wiphys on top of a card is working quite good. Of course there are several restrictions and performance problems, we are working on some scheduling algorithms to improve the operation.
But now to take it a step further, we plan to go for a mac80211 based virtualization, meaning to run multiple vifs in parallel in station mode. The first step is to run them on the same channel, but as expected this is not working. We can create several interfaces with iw, there is no problem. Even scanning is working, but as soon as we want to connect to an open AP we get into trouble. The authentication is not working.
 
Sure, the mac80211 needs some additional functionality like ath9k, to cope with multiple associations and channel switching, etc.. How can we track the sequence of operations during the runtime in order to get a better understanding of the mac80211. At the moment, we enabled tracing and debug messages for mac80211, this helps but is not sufficient to follow the functions step by step. We can put a lot of debugging messages into the files, but  are there any "best practices" we should follow. 


Thanks,

Hakan

[ath9k-devel] Multiple interfaces with one card

[Luis R. Rodriguez]

On Fri, Aug 06, 2010 at 03:32:20AM -0700, Hakan Coskun wrote:
> Hi,
> 
> we are doing research in order to study the behaviour of virtual wifi interfaces with channel switching. After a simulative study, we started to perform some real world experiments with ath9k. Running multiple wiphys on top of a card is working quite good. Of course there are several restrictions and performance problems, we are working on some scheduling algorithms to improve the operation.
> But now to take it a step further, we plan to go for a mac80211 based virtualization, meaning to run multiple vifs in parallel in station mode. The first step is to run them on the same channel, but as expected this is not working. We can create several interfaces with iw, there is no problem. Even scanning is working, but as soon as we want to connect to an open AP we get into trouble. The authentication is not working.
>  
> Sure, the mac80211 needs some additional functionality like ath9k, to cope with multiple associations and channel switching, etc.. How can we track the sequence of operations during the runtime in order to get a better understanding of the mac80211. At the moment, we enabled tracing and debug messages for mac80211, this helps but is not sufficient to follow the functions step by step. We can put a lot of debugging messages into the files, but  are there any "best practices" we should follow. 
> 

Hakan, I recommend you use the linux-wireless mailing list for this
and also always use wireless-testing.git for your development.

Can you post patches of what you have done to mac80211 / cfg80211?

  Luis

Re: [ath9k-devel] Multiple interfaces with one card

[Luis R. Rodriguez]

On Fri, Aug 06, 2010 at 03:32:20AM -0700, Hakan Coskun wrote:
> Hi,
> 
> we are doing research in order to study the behaviour of virtual wifi interfaces with channel switching. After a simulative study, we started to perform some real world experiments with ath9k. Running multiple wiphys on top of a card is working quite good. Of course there are several restrictions and performance problems, we are working on some scheduling algorithms to improve the operation.
> But now to take it a step further, we plan to go for a mac80211 based virtualization, meaning to run multiple vifs in parallel in station mode. The first step is to run them on the same channel, but as expected this is not working. We can create several interfaces with iw, there is no problem. Even scanning is working, but as soon as we want to connect to an open AP we get into trouble. The authentication is not working.
>  
> Sure, the mac80211 needs some additional functionality like ath9k, to cope with multiple associations and channel switching, etc.. How can we track the sequence of operations during the runtime in order to get a better understanding of the mac80211. At the moment, we enabled tracing and debug messages for mac80211, this helps but is not sufficient to follow the functions step by step. We can put a lot of debugging messages into the files, but  are there any "best practices" we should follow. 
> 

Hakan, I recommend you use the linux-wireless mailing list for this
and also always use wireless-testing.git for your development.

Can you post patches of what you have done to mac80211 / cfg80211?

  Luis

[ath9k-devel] Multiple interfaces with one card

[Hakan Coskun]

Am 06.08.2010 um 20:48 schrieb Luis R. Rodriguez:

> On Fri, Aug 06, 2010 at 03:32:20AM -0700, Hakan Coskun wrote:
>> Hi,
>> 
>> we are doing research in order to study the behaviour of virtual wifi interfaces with channel switching. After a simulative study, we started to perform some real world experiments with ath9k. Running multiple wiphys on top of a card is working quite good. Of course there are several restrictions and performance problems, we are working on some scheduling algorithms to improve the operation.
>> But now to take it a step further, we plan to go for a mac80211 based virtualization, meaning to run multiple vifs in parallel in station mode. The first step is to run them on the same channel, but as expected this is not working. We can create several interfaces with iw, there is no problem. Even scanning is working, but as soon as we want to connect to an open AP we get into trouble. The authentication is not working.
>> 
>> Sure, the mac80211 needs some additional functionality like ath9k, to cope with multiple associations and channel switching, etc.. How can we track the sequence of operations during the runtime in order to get a better understanding of the mac80211. At the moment, we enabled tracing and debug messages for mac80211, this helps but is not sufficient to follow the functions step by step. We can put a lot of debugging messages into the files, but  are there any "best practices" we should follow. 
>> 
> 
> Hakan, I recommend you use the linux-wireless mailing list for this
> and also always use wireless-testing.git for your development.
> 
> Can you post patches of what you have done to mac80211 / cfg80211?
> 
>  Luis
> 
Hi Luis,

thank you for your really quick reply. We are already using wireless-testing for our development. The ath9k extensions will be commited soon. At the moment, we are using the eclipse to index the code and walk through the files. This is working quite good. But how can we see during runtime, what is happening. Should we use something like kgdb to trace the operation or what is your approach you take for development. Can you give us any hints ?

Hakan

Re: [ath9k-devel] Multiple interfaces with one card

[Hakan Coskun]

Am 06.08.2010 um 20:48 schrieb Luis R. Rodriguez:

> On Fri, Aug 06, 2010 at 03:32:20AM -0700, Hakan Coskun wrote:
>> Hi,
>> 
>> we are doing research in order to study the behaviour of virtual wifi interfaces with channel switching. After a simulative study, we started to perform some real world experiments with ath9k. Running multiple wiphys on top of a card is working quite good. Of course there are several restrictions and performance problems, we are working on some scheduling algorithms to improve the operation.
>> But now to take it a step further, we plan to go for a mac80211 based virtualization, meaning to run multiple vifs in parallel in station mode. The first step is to run them on the same channel, but as expected this is not working. We can create several interfaces with iw, there is no problem. Even scanning is working, but as soon as we want to connect to an open AP we get into trouble. The authentication is not working.
>> 
>> Sure, the mac80211 needs some additional functionality like ath9k, to cope with multiple associations and channel switching, etc.. How can we track the sequence of operations during the runtime in order to get a better understanding of the mac80211. At the moment, we enabled tracing and debug messages for mac80211, this helps but is not sufficient to follow the functions step by step. We can put a lot of debugging messages into the files, but  are there any "best practices" we should follow. 
>> 
> 
> Hakan, I recommend you use the linux-wireless mailing list for this
> and also always use wireless-testing.git for your development.
> 
> Can you post patches of what you have done to mac80211 / cfg80211?
> 
>  Luis
> 
Hi Luis,

thank you for your really quick reply. We are already using wireless-testing for our development. The ath9k extensions will be commited soon. At the moment, we are using the eclipse to index the code and walk through the files. This is working quite good. But how can we see during runtime, what is happening. Should we use something like kgdb to trace the operation or what is your approach you take for development. Can you give us any hints ?

Hakan

[ath9k-devel] Multiple interfaces with one card

[Bob Copeland]

On Fri, Aug 6, 2010 at 2:59 PM, Hakan Coskun <coskun@cs.tu-berlin.de> wrote:

> thank you for your really quick reply. We are already using wireless-testing for our development. The ath9k extensions will be commited soon. At the moment, we are using the eclipse to index the code and walk through the files. This is working quite good. But how can we see during runtime, what is happening. Should we use something like kgdb to trace the operation or what is your approach you take for development. Can you give us any hints ?
>

IMHO the best way to do this is a combination of mac80211 tracepoints and
ftrace, the function tracer, via trace-cmd.  Something like:

# trace-cmd record -p function -l ieee\* -e mac80211 sleep 300

...should get you five minutes worth of useful traces.  Of course command can
be anything you like.


-- 
Bob Copeland %% www.bobcopeland.com

Re: [ath9k-devel] Multiple interfaces with one card

[Bob Copeland]

On Fri, Aug 6, 2010 at 2:59 PM, Hakan Coskun <coskun@cs.tu-berlin.de> wrote:

> thank you for your really quick reply. We are already using wireless-testing for our development. The ath9k extensions will be commited soon. At the moment, we are using the eclipse to index the code and walk through the files. This is working quite good. But how can we see during runtime, what is happening. Should we use something like kgdb to trace the operation or what is your approach you take for development. Can you give us any hints ?
>

IMHO the best way to do this is a combination of mac80211 tracepoints and
ftrace, the function tracer, via trace-cmd.  Something like:

# trace-cmd record -p function -l ieee\* -e mac80211 sleep 300

...should get you five minutes worth of useful traces.  Of course command can
be anything you like.


-- 
Bob Copeland %% www.bobcopeland.com