Loading ...
Sorry, an error occurred while loading the content.

Re: [nslu2-linux] My experience with the debian installation

Expand Messages
  • Rod Whitby
    ... Can you run free repeatedly whilst it s creating the ext3 filesystem, to see if it s running out of memory? Also the output of dmesg after mkfs.ext3
    Message 1 of 7 , Jan 3, 2007
    • 0 Attachment
      docgoeken2000 wrote:
      > I got a second NSLU2 and decided to upgrade it to debian. (My first
      > slug has been running unslung for more than a year.) I wanted to
      > document my experience with the upgrade in the belief that it may
      > help the developers make a great product better.
      >
      > If any of the developers want me to try additional tests, I can delay
      > loading additional packages on the debian slug before I put it into
      > full time use.

      Can you run "free" repeatedly whilst it's creating the ext3 filesystem,
      to see if it's running out of memory? Also the output of "dmesg" after
      mkfs.ext3 has failed would be useful.

      Thanks for documenting your experience and investigating the problem.

      -- Rod
    • Gordon Farquharson
      ... I have also seen the partitioner stop at 33% with a 250 GB hard drive, but I was using custom built installer images. However, I only saw the problem once
      Message 2 of 7 , Jan 3, 2007
      • 0 Attachment
        On 1/3/07, docgoeken2000 <dgoeken@...> wrote:

        > I first took the stock slug, modified it for 266 speed, and then
        > formatted a 250 GB usb disk just to make sure everything was
        > working. Then I flashed using the rc1 debian installer from www.slug-
        > firmware.net. Rebooted, did the ssh and followed the installation
        > instructions. Everything went well until the `Formatting Partitions'
        > step. Kept stopping at the 33% mark - similar to what has been

        I have also seen the partitioner stop at 33% with a 250 GB hard drive,
        but I was using custom built installer images. However, I only saw the
        problem once or twice, and I think that I got a red error screen when
        the paritioner quit. For the early December daily build images
        (post-rc1), I don't remember experiencing this problem. Sounds like I
        should test out the daily build with 2.6.18-4-ixp4xx + the open source
        ethernet driver and NPE firmware in the next couple of days to see if
        I can reproduce the problem.

        > Have the problems reported been only with a 1 GB drive and is my
        > problem unrelated since I'm using a 250 GB drive? Could there be a
        > problem with the formatting program (mkfs.ext3)?

        I don't think so - I suspect that it is caused when the system runs
        out of memory, but I have not done extensive testing because I don't
        think that I ever saw the problem with rc1.

        > Which usb port is the best port to use with debian? Another
        > interesting issue – when I was successful, the drive was plugged into
        > USB2. However, the slug's lights show Disk1 attached. If I switch
        > the drive to the USB1 port, the slug still boots and the slug's light
        > still show Disk1 attached.

        The illumination of the Disk 1 LED is hardcoded in /etc/init.d/zleds.
        As far as I know, there isn't an implementation for automatically
        illuminating the disk LEDs for any of the non Linksys based firmwares
        - I could be wrong, but it is definitely not in Debian/NSLU2.

        > Another question: installation gave me a choice of 3 kernels to load,
        > which is the best to use? (I am currently using '2.6.18-3-ixp4xx'.

        This version is the latest kernel in testing. It has a few issues, but
        most of them have been worked out in unstable. Which versions were
        offered, BTW ?

        > If any of the developers want me to try additional tests, I can delay
        > loading additional packages on the debian slug before I put it into
        > full time use.

        Try looking at the memory usage (like Rod suggested) and let us know
        what you find.

        > Sorry for the length and for some of the ramblings, but hope this
        > helps.

        Not a problem. Thanks for sharing your experience.

        Gordon

        --
        Gordon Farquharson
      • Rod Whitby
        ... That is correct. -- Rod
        Message 3 of 7 , Jan 3, 2007
        • 0 Attachment
          Gordon Farquharson wrote:
          > As far as I know, there isn't an implementation for automatically
          > illuminating the disk LEDs for any of the non Linksys based firmwares

          That is correct.

          -- Rod
        • docgoeken2000
          I did a reinstall of rc1 debian. I repeatedly ran, ps , dmesg , and free from a shell. Result: looks like the problem is caused from the system running
          Message 4 of 7 , Jan 5, 2007
          • 0 Attachment
            I did a reinstall of rc1 debian. I repeatedly ran, "ps", "dmesg",
            and "free" from a shell.

            Result: looks like the problem is caused from the system running out
            of memory as suspected.

            free initially outputs a report like:

            total used free shared buffers
            Mem: 30080 19516 10564 0 0
            Swap: 0 0 0
            Total: 30080 19516 10564

            During the initial file downloads, the free value does drop to around
            5000 (from the 10564 above).

            Then swap space is created:
            Adding 546168k swap on /dev/scsi/host0/bus0/target0/lun0/part5.
            Priority:-1 ext

            When mkfs.ext3 /dev/scsi/host0/bus0/target0/lun0/part1 started, free
            reports:

            total used free shared buffers
            Mem: 30080 28996 1084 0 7736
            Swap: 546168 16740 529428
            Total: 576248 45736 530512

            Before the formatting failed, I saw the free value drop below 700
            (from the 1084). However, when I really wanted the output from
            ps/dmesg/free, the shell was frozen from the formatter failing on the
            slug.

            Some of the failures did produce a red error screen. Again this time
            the formatter failed 4 or 5 times and then was successful. This time
            I was successful in doing the formatting from the install gui.

            dmesg did report:
            Out of memory: Killed process 21308 (mkfs.ext3).
            Out of memory: Killed process 23487 (mkfs.ext3).

            I did some cut and paste from the ssh window and put it in a text
            file. It's not pretty, but if you are interested in seeing the raw
            info, the file can be found here:
            http://goeken.homedns.org/Weather/nslu2/InstallOutput.txt
            I tried to put a few notes in it. In addition, it looks like the
            first part of the output from dmesg gets truncated when the output
            gets large. If you need me to comment it more or to format it
            better, let me know.

            Not sure which mirror I used the first time, but this time, I was not
            given a choice of kernels. I started with whatever was the first
            mirror choice, but file downloads from it failed, so I had to switch
            to another mirror for this install.

            Does it matter which usb port to use as the boot device with debian?

            What should I be using for a swap space size? (I tried 559.3 MB this
            install.)

            Hope this helps,
            Dave
          • Kurt Pruenner
            ... Well, the default behaviour of the linux kernel is to kill processes that use more memory than is physically available, no matter how much swap there is.
            Message 5 of 7 , Jan 7, 2007
            • 0 Attachment
              docgoeken2000 wrote:
              > dmesg did report:
              > Out of memory: Killed process 21308 (mkfs.ext3).
              > Out of memory: Killed process 23487 (mkfs.ext3).

              Well, the default behaviour of the linux kernel is to kill processes
              that use more memory than is physically available, no matter how much
              swap there is.

              Try running

              echo 2 > /proc/sys/vm/overcommit_memory
              echo 80 > /proc/sys/vm/overcommit_ratio

              in a shell before creating the file systems but after activating swap,
              that should help.

              Here's some more info on the problem:

              http://www.win.tue.nl/~aeb/linux/lk/lk-9.html#ss9.6

              --
              Kurt Bernhard Pruenner --- Haendelstrasse 17 --- 4020 Linz --- Austria
              .......It might be written "Mindfuck", but it's spelt "L-A-I-N".......
              np: Boris Kauer/Autocyklus - Time Control (Für Dich)
            • Martin Michlmayr
              ... This sounds really interesting. Unfortunately, it doesn t seem to work. I made some settings and have enough swap but the ssh/mkfs.ext3 process is still
              Message 6 of 7 , Jan 8, 2007
              • 0 Attachment
                * Kurt Pruenner <leak_usenet@...> [2007-01-07 15:40]:
                > > Out of memory: Killed process 21308 (mkfs.ext3).
                > > Out of memory: Killed process 23487 (mkfs.ext3).
                >
                > Well, the default behaviour of the linux kernel is to kill processes
                > that use more memory than is physically available, no matter how much
                > swap there is.
                >
                > Try running
                > echo 2 > /proc/sys/vm/overcommit_memory
                > echo 80 > /proc/sys/vm/overcommit_ratio

                This sounds really interesting. Unfortunately, it doesn't seem to
                work. I made some settings and have enough swap but the ssh/mkfs.ext3
                process is still being killed.

                total used free shared buffers
                Mem: 30000 28728 1272 0 5624
                Swap: 96348 4692 91656

                --
                Martin Michlmayr
                http://www.cyrius.com/
              Your message has been successfully submitted and would be delivered to recipients shortly.