WiFi slowdown / stalling. Advice?

[hmc]

Hi guys,

I'm having trouble with one of my Pandoras. The WiFi seems to be broken

I did some tests using wget, fetching a 200MB file from my web server, observing the throughput wget reports.

Following results:

Test with a reference Pandora, that works well:

500..700K/s (abt. 5..6 minutes downloading time). Relatively constant throughput, only normal temporary deviations.

Test with the problematic device:

1. wget at first reports 650K/s / 6m ETA, too. 

But then the throughput constantly decreases, down to about 70K/s resulting in ETA of about 50m.

2. I aborted the transfer and restarted it, but then it began with 70K/s.

3. I rebooted and tried again:

Throughput again begins with 650K/s / ETA 6m. After about a minute it's down to 50..70K/s again. Then it raises a bit to 85K/s. 

During the remaining test throughput mainly alternates in the range of 60..120K/s.

For short periods of time throughput raises to 300..400K/s again.

Prior to those tests, I also experienced complete stalling of transfers, e.g. when installing software using PNDManager. I could not reproduce such stalling now though.

The above results are reproducible at any time, i.e. I have made sure that no other data transfer in the network or external interferences like microwave oven etc. slow down that throughput. It's really the Pandora.

I use the latest firmware, SuperZaxxon 1.54 with all updates installed as of today.

dmesg shows the exact same output for WLAN activation, including firmware version 6.0.4.235, on both units.

Both units are equally far away from the WLAN access point, varying the positions slightly during the test, to make sure that no signal reception "hole" affects the results. (you know, I have written my master thesis about IEEE802.11b WLAN measurements in conjunction with multimedia data streams, so I know what traps there can be in a wireless LAN ;-).

I may later dig out my measurement scripts from back then - it's been 10 years ago now - adapt them and let them produce a few nice graphs of the throughput. Maybe this sheds some more light on the problem by showing some patterns or dependencies.

Any idea what the reason can be for the problems?

Anything I can do to debug this problem further?

Thanks!

 

[mcobit]

http://boards.openpandora.org/index.php/topic/11416-wifi-mysteries-resolved-once-and-for-all-%E2%80%94-power-saving-is-clearly-the-culprit/


Just in case you missed it. Don't let the title fool you. There is no real solution but there are some tests and speculation.

 

[Mr_Loon]

Probably not the answer you're looking for but I bought 1 of these for my CC Pandora and have never looked back.

The battery impact is pretty minimal if you toggle it on and off as and when you require WIFI,it takes only a few seconds to toggle the USB host. Added bonus : it works out of the box, Drivers are part of the firmware.

 

[hmc]

http://boards.openpandora.org/index.php/topic/11416-wifi-mysteries-resolved-once-and-for-all-%E2%80%94-power-saving-is-clearly-the-culprit/


Just in case you missed it. Don't let the title fool you. There is no real solution but there are some tests and speculation.

oh thanks,

I indeed missed it.

Tested switching off power management now, and it did not help.

Then I disassembled the entire Pandora, also separated the WiFi antenna from the screen (attached to it using conductive adhesive tape to join the grounds of the antenna and the screen).

When assembling the parts without a case in front of me on the table, and restarting the wget test, guess what?: A quite steady throughput of over 600K/s! Reproducible!

I know about that common problem of that adhesive tape connecting the actual antenna contacts and thus short-circuiting the antenna. But that was definitely not the case here, because I took very good care of that when I assembled the Pandora some months ago. I have no idea what the cause could be, that without the case now WiFi works well. Maybe it was the antenna's ground connection to the screen, which caused the problems? Probably not: I just reconnected both as before, and still get that good throughput. 

Maybe one of the antenna cable connectors were not properly connected? Maybe...

But I have another interesting result:

Even if I disconnect the antenna from the antenna cable, I still get 600K/s of throughput.

So next (probably tomorrow) I will reassemble the Pandora (adding the LED keyboard light and hinge resting angle mods ;-) ), being very careful about the WiFi antenna and its connection to the motherboard, and then retest and report here.

This result leads me to the assumption, that maybe in many more cases when WiFi is not working properly and disabling power management does not help, there might be some slight hardware problem and disassembling and reassembling the Pandora might indeed help.

 

[notaz]

Do you get that on CC pandora? I get mostly the same on my CC (the chip throttles itself to ~70K/s no matter what after few seconds of operation), and 1GHz doesn't have this problem. MWeston did some tweaks for rebirth/GHz edition boards that seemed to help in my case.

 

[hmc]

No, my problematic Pandora is a 1GHz model. The CC unit works fome for me.

 

[hmc]

okay guys,

I did some testing over the weekend. 

The disassembled Pandora was not always acting well. It degraded in transmission speed over time again.

So I reassembled both Pandoras again, but in different hardware combinations, several times. I tested different combinations of the available mainboards, antennas and antenna cables. Used always the same system (booted from SD) for testing, so no software differences.

Result:
It's reall a difference of the mainboards, that lets WiFi acto so differently.

One mainboard has steady 600...700 Kb/s throuput, the other mainboard always slows down after a few seconds and then stays at 70 Kb/s, sometimes slightly higher or lower. Also switching off power management of the WiFi chip does not help.

Too bad :-( I hoped that it's some other hardware component that's easier / cheaper to exchange, or that it's some difference in software.

Those tests involved my two 1GHz Pandoras.

I also own one CC Pandora. That one also has the 70 Kb/s problem, very similar in behavior as the problematic 1GHz unit.