[maemo-developers] How to detect that connectivity is lost

[From: Kalle Valo]

ext Tomàs Jiménez Lozano <tomas at oranginalab.com> writes:

> To reproduce the situation you only need to move away from de access
> point. I've tried what Frantisek suggested, and ping fails always
> before the device notifying it has lost connection.

Thanks, I think I managed to reproduce this now. Would you be kind
enough to file a bug to bugzilla for easier tracking:

https://maemo.org/bugzilla/

> I've checked how far from AP is raised de dbus signal notifying
> disconnection and the average distance where ping starts to
> definitiveley fail and ping always fails quite more near from IAP than
> disconnection signal is raised.

That's what I saw. Ping was failing just at the edge of AP radio
coverage but N800 didn't drop from the network. After I moved few
meters, only when the connection was dropped.

> If, as it seems to be, it is a lower level bug I beg you to try to fix
> it as soon as possible.

Naturally. But it would help if you could file a bug to bugzilla. I
receive way too many emails, and sometimes I loose some of them. But I
have never lost a bug in bugzilla :)

> In the meanwhile, we need to use the pinging workaround because for
> our application it is critical to detect (and recover) from any
> connectivity lost. As you feared, Kalle, the battery performance
> reduces drastically (what becomes another issue for our app). We do
> only ping every 15 second in order to let the device enter in PSM
> between pings but the battery performance is still too poor.

Are you implementing a customised application for a specific group of
customers and not for the maemo community? If yes, I thinkg what you
could do is to use the wlan_sleep_timeout. I wrote about it some time
ago:

http://maemo.org/pipermail/maemo-users/2007-February/004092.html

For a temporary workaround, you could set it to zero to enable Full
Power Save. That way the periodic pings would reduce the power
consumption. But not without a cost, naturally. The WLAN troughput
will be smaller and there are buggy APs (especially the older ones
which are not Wi-Fi certified) which might not work properly with Full
Power Save.

> On the other hand, the pinging workaround has fixed another problem we
> were detecting in our app. As I told before, our app need to be always
> WLAN connected, and we have checked that the device after a given amount
> of time of communication inactivity enters in what we suppose is some
> kind of higher lever PSM.

There's only one level of Power Save Mode in 770/N800. The
wlan_sleep_timeout is the only thing controlling when the WLAN chip
goes to sleep, and that's 1000 ms by default. So after a second of
transmission inactivity the device would go to sleep.

> When in this state, the device is still connected but it does not
> answer to ping requests. When it needs to send any data it wakes up
> automatically but for us it may be too late because communication
> with other network devices may has been lost. I've been looking for
> information about this supposed high level PSM and how to avoid it
> but I've found nothing.

I think that's also a bug in our WLAN stack. When you see this, could
you take a look at kernel logs (for example with dmesg) and post them
here. Actually, it's even better to file a bug to bugzilla. Attach the
logs to the bug, if you manage to get them.

> I hope you don't mind if I ask you one more question: Is there any other
> way to force disconnection appart form the shutdown ICd dbus method. It
> works fine but we would like to do it without prompting the user?

I'm so involved with WLAN nowadays that I keep forgetting details about
user-space more and more :) Patrik, can you help me here, please?

-- 
Kalle Valo