• Home
  • Help
  • Search
  • Login
  • Register
Pages: [1]
Author Topic: Corruption on SD card?  (Read 3861 times)
DamonHD
Full Member
***

Karma: 4
Posts: 169


View Profile WWW
« on: September 28, 2009, 11:55:43 AM »

Hi,

I'm seeing some evidence of corruption on files written to my SD card (application-level checksums failing, once in SVN, once in tar).

Is anyone else seeing this?

The filesystem is ext3 and I have a long commit= time, so it could be corruption in memory instead before stuff is flushed to the card of course.

Rgds

Damon
Logged

aeg
Newbie
*

Karma: 0
Posts: 3


View Profile
« Reply #1 on: October 06, 2009, 11:17:24 PM »

I see you posted this a week ago and got no replies.

Have GlobalScale or Marvell done a root cause analysis and issued an ECO?

If not, the usual solution for this sort of thing is to send it back and buy something else.
Logged

DamonHD
Full Member
***

Karma: 4
Posts: 169


View Profile WWW
« Reply #2 on: October 07, 2009, 12:37:40 AM »

Hi,

The plug seems to be generally working very well, and I am not sending it back (in fact I ordered a second).

But I just noted the corruption under heavy load as a data point in case anyone else sees this.

There's all sorts of places the problem could be of course; the SDIO code, the DRAM, my dodgy power supply, etc, etc.

Rgds

Damon
Logged

Rabeeh Khoury
Administrator
Full Member
*****

Karma: 5
Posts: 218


View Profile
« Reply #3 on: October 11, 2009, 03:06:56 AM »

Can you please send details on how you do the test?
Logged

DamonHD
Full Member
***

Karma: 4
Posts: 169


View Profile WWW
« Reply #4 on: October 11, 2009, 09:12:20 AM »

Hi,

It wasn't so much a formal test as blasting all the stuff from my old server to my new one, either over sftp or via a USB stick, and noting that a couple of things with checksums were reported as broken.

Presumably there's also some corruption that I don't know about... Bv<

Rgds

Damon
Logged

rooster
Administrator
Sr. Member
*****

Karma: 8
Posts: 311


View Profile
« Reply #5 on: October 11, 2009, 09:36:37 AM »

Have you experienced this with several SD cards? USB DOK/HDD?
Logged

birdman
Sr. Member
****

Karma: 4
Posts: 443


View Profile WWW
« Reply #6 on: October 11, 2009, 06:35:59 PM »

I've juts seen something odd that looks like it might be related to the SD card.

I use the Plug to copy files of a DVR recorder (over USB).  It's been doing this a few times over the last two months to an external USB disk.  No problems.
I updated the system to installerv1.0, put in a larger SDHC card (Transcend 16GB) and kernel 2.6.31.3 over the weekend.  With the additional SDHC space I copied the files from the DVR to there instead of leaving in an external USB drive during the copy.   Only one file copy completed OK.  The other four were truncated (one of which was zero bytes).  There were a few syslog messages generated, but these were for a known 2.6.31.3 bug.   There was no sign of filesystem (metadata) corruption at all.  I reverted to kernel 2.6.30.5.  The download problem persisted.  There were no kernel errors.  (Might download script also wasn't reporting any errors - the file transfer just stops).
At this stage a download of the same files directly to the external USB drive ran through perfectly OK.
Next I switched back to my original, smaller, SDHC card (Verbatim 8GB).  It suffered the same problem with truncated files on downloading to it.
So what I'm seeing appears to be that writing to the SDHC card produces some interaction with reading from my USB download port.
Odd - and I'll see whether I can get any more info about this.
Logged

DamonHD
Full Member
***

Karma: 4
Posts: 169


View Profile WWW
« Reply #7 on: October 12, 2009, 12:53:27 AM »

For my part I only saw this with my 32GB SDHC card, but of course I wasn't looking for it at all, and I don't think that there can be a lot of it going on else I'd have noticed, and I did have dodgy power at the time (a loose power lead!).  It might have been nice to have had BTRFS or ZFS to keep an eye on the data integrity for me, but I'm definitely not getting my nose out of joint with this.  I will possibly work my new SheevaPlug harder (with my 8GB SD card) and see if I can reproduce anything.

But stressing the power system would be unkind to the the SD card, etc, if that's what the problem was.

Rgds

Damon
Logged

birdman
Sr. Member
****

Karma: 4
Posts: 443


View Profile WWW
« Reply #8 on: October 12, 2009, 04:43:59 PM »

But stressing the power system would be unkind to the the SD card, etc, if that's what the problem was.
Well - the DVR doesn't draw any power.  However, I have a USB hub with a Wifi dongle in it.  Normally that hub is unpowered.  But with the external drive in place I power it up (although during my weekend changes that power supply did pack up, which confused me for a while until I replaced it - but the oddities were seen after that).
So - perhaps I should see what happens if I repeat the download-to-SDHC setup but with the USB hub powered up.
Logged

aeg
Newbie
*

Karma: 0
Posts: 3


View Profile
« Reply #9 on: October 24, 2009, 09:07:16 AM »

I'm glad to see Rabeeh Khoury has taken an interest in this problem. I hope this means Marvell is taking this seriously and the problem will be fixed soon.
Logged

DamonHD
Full Member
***

Karma: 4
Posts: 169


View Profile WWW
« Reply #10 on: October 24, 2009, 11:40:31 AM »

Since I sorted out my (solar PV) power-supply issues I have not *noticed* any further corruption.

# uptime
 18:40:05 up 16 days,  6:19,  1 user,  load average: 0.41, 0.30, 0.27

...and counting!

Rgds

Damon
Logged

DamonHD
Full Member
***

Karma: 4
Posts: 169


View Profile WWW
« Reply #11 on: October 28, 2009, 02:41:20 AM »

OK, next data point...

The ext3 partition on my USB Flash drive would not mount this morning and reported shedloads of errors and I thought it was completely dead or at least randomly failing with fsck having difficulty even opening the partition on alternate attempts.  I was just warming up the supplier for a returns number...

However, testing on another machine (my dusty Linux laptop) indicates that it is mostly OK, with possibly one file corrupt (note, not a file written on the Sheeva).  I've tested all those files on the USB drive with built-in checksums, eg ZIP files, and diff -r the main file tree against the previous version on the laptop.

So I'm beginning to wonder if the problem isn't in the mind of the kernel on the Sheeva (uptime 19 days at the moment).

I'm assuming that it doesn't have ECC on its RAM (I didn't check the specs), so crap could have accumulated there causing the kernel to misbehave for example?

Rgds

Damon
Logged

DamonHD
Full Member
***

Karma: 4
Posts: 169


View Profile WWW
« Reply #12 on: October 28, 2009, 04:05:07 AM »

Rebooting the plug didn't fix the problem (though it needed to be power-cycled: a shutdown -r now didn't do it).

Swapping out the USB extender cable between my plug and the hub (for a shorter one as it happens) seems to have the trick, though why a cable should 'go bad' I don't know.  More likely a bad connection at one end or the other.

Roll on btrfs or its ilk with checksummed data, I say!

Rgds

Damon
Logged

Pages: [1]
Print
Jump to: