Sugar on a Stick/Stick Layout

Failure Modes

1. Gets stuck at Fedora Login
   1. With liveuser prefilled
      I was able to repeatably create this failure when I was testing a version of schoolserver.py that had a typo in it.
   2. With nothing prefilled
      This error has successfully been replicated by Restarting multiple times.
2. Bug in initramfs /init detected. Dropping to a shell. Good luck! bash: no job control in this shell.
   Stick abuse, like putting it through the washing machine, seems to precipitate this type of failure.
3. Freezes during use of an activity. Rebooting on the same computer doesn't help it still freezes on the same activity. One time we rebooted on a different computer and it worked.

We don't know the root causes but we have plenty of theories.

Sticks will always fail. There will always be multiple causes for failure. This problem has to be addressed like a process engineering issue. Is stick failure a major cost in time or money? Find the largest cause of failure. Fix that to improve yield. Is stick failure a major cost in time or money? Find the the remaining largest cause of failure.....

This summer we had probably 10 sticks fail for 35 students. That is a major cost in time and money. The most common unrecoverable failure mode was stuck at Login.

We need a more robust files system

It has been shown that if you take a stick that is stuck at login and you copy over a fedora-overlay file from a working stick, the stick will boot again. Thus we are confident that the problem is corruption of the overlay file.

Next Steps - Try some alternate file structures.

1. Create a Fedora Full Install USB Stick, restart it a bunch of times and see if it does better. How much space will the Fedora Full Install take?
   So far its > 4GB http://wiki.sugarlabs.org/go/VMware#Full_Install_with_F11_Net_install_CD_to_USB_Stick
2. Create an Open Suse USB Stick, restart it multiple times and see if it fails. How much space does it use?[1] Note the second partition for persistence
3. Create Alternate ex3 File Structured, non-live, blueberry USB [2] (The resulting partition size on the USB can be resized with gparted if more storage is needed)[3]

Background Links Thread from the Fedora Forum: [[4]]

Here is an idea for an alternative USB Format: [[5]]

Also see Ticket #907

A 2 GB USB is about $.60 more then a 1GB stick. If we need to compress less to get robustness its ok for us to require a 2GB USB. A 4GB stick is about $2.50 more then a $1GB Stick.

Theories that have evidence against them

• ext4 and data loss:http://lwn.net/Articles/322823/ (It looks like ext3 formatting may be more robust in a USB)

Dave Bauer was able to replicate the failure easily by restarting a Sugar stick. This means its probably not:

• The Bulk Copier, Dave doesn't have one.
• The older slower computers and USB 1 ports we are using at the GPA.

Sticks are damaged during formatting or burning

http://wiki.laptop.org/go/How_to_Damage_a_FLASH_Storage_Device [[6]]

There was a detailed discussion thread back in February at: http://lists.laptop.org/pipermail/devel/2009-February/022987.html

If we understood how to win and we had our own stick creation activity perhaps we would have less failures.

• Look at How To Sugarize liveusb-creator:[7] for a working method to do this.

Sticks are of poor quality

Here is an explanation of two processes used to create USB sticks: http://www.solutiongrove.com/blogger/2008/09/08/there-are-two-type-of-usb-flash-memory-slc-and-mlc [[8]]

Sticks are being improperly removed

Yes they are, but its not at all clear to me that that correlates with the sticks that are failing.

Current experements on delayed writes from cache in live usb's [9]

There is a bug around restarting

It is very common to fail after a restart, the system forgets its supposed to go to Sugar. See ticket {{bug|1069]]