Deepti set up an experiment with 2 badges and 3 anchors in one evening. By the next morning, both badges had stopped working. We set out to understand why they were failing:
Mote 219
- Started fully charged, 5pm June 22
- Stopped writing to flash after 44596 mote seconds,
which is about 12 hours.
- One of the anchors has data coming from mote 219 until 44760 (9 extra
broadcasts) and then no more data from 219.
- By that time the mote had acculumated 37120 bytes of data in flash
(at least that much had been committed).
- Read flash ran with no problem, suggesting that the data was
correctly written.
- Checked battery level 10 am June 23. Battery level 3.67, so should
high enough to work.
- Then reset the mote, the mote went through the light cycyle, resynched,
but continued to not broadcast.
- Then cleared flash and reloaded program and it operated normally.
I suspect a bad pointer problem in writing to flash, possibly a
bad segment in memory, or a wrap-around problem when getting to the
end of address space.
Mote 227
- Started fully charged, 5pm June 22
- Last message recorded at 38373, about 10 hours into experiment
- Memory was not readable for 3 attempts.
- Used our program to write a few more bytes to memory, then repeated
TFlash and successfully read 32,512 bytes (including about 30 we
added to the end)
- The last message heard by the anchor occured at 38439
- We noticed that the broadcasts from 227 to the anchor were widely
spaced, sometimes a minute or more apart.
- On inspecting the device, we noticed that the battery conector was
loose. We had measured the voltage on the board to be 3.25V, but the
battry itself had 3.9V.
We conclude that the mote died because of a poor power connection.
Deepti/Geb: if the flash software is suspect, then we might consider bypassing the TinyOS file system and writing to flash directly. This would have benefits and possibly also some disadvantages to consider.
Comment by tedherman — June 25, 2010 @ 12:38 pm