Hi LQ'ers.

I've got a server that I've got running a customized Debian Live from a USB stick.

Hardware:
Supermicro X7SPE-HF-D525 (Atom D525 dual core @1.8GHz)
16GB Mushkin Enhanced USB stick in the on-motherboard USB-A socket
4GB Mushkin Enhanced RAM
3Ware 9650SE-8LPML 8-Port RAID card
14, 750GB Seagate Momentus HDDs (in RAID 6 with 1 spare)

Software:
Syslinux (on FAT32 on USB stick)
Debian Live (Wheezy) kernel: 2.6.32-5-amd64 (using USB-HDD option for libe builder)
Boot args: live config noautologin persistent toram

Services:
Captive portal for Wi-Fi
AMP server
NFS server for Wi-Fi clients
MediaTomb server for PlayStation 3

On the first run, I started to get strange behaviour on the 32nd day. Sorry I didn't log them, but I can remember that it started with: BitTornado (btlaunchmanycurses) crashed, then when I tried to restart it segfaulted; MediaTomb stopped generating thumbnails (using ffmpegthumbnailer) and marking objects as 'played'. I tried restarting MediaTomb, but it segfaulted. I restarted the server.

On this second run, it's on the 51st day and the only issue is that MediaTomb has stopped marking objects as 'played'. Ill try restarting it at some point today.

I should mention that I've added a live-old option to syslinux so that when I generate a new system image, I 'rm -rf live-old && mv live live-old && mv /path/to/new/live live && reboot'. I like to be able to fall back to the working system in case something goes wrong. Downtime is BAD.

So my questions are:

What kind of uptime can I expect from this type of system?

What can I do to increase reliability (without re-doing everything)?

Thank you in advance for your constructive responses.

Cheers!
Randall

How much free RAM do you have when the issues occur?
Since you are using a live version, all what is saved in the filesystem is stored in RAM. This may cause trouble when you have many temporary files or very large log-files.

Thanks for responding TobySGD.

RAM usage has never exceeded 490MB (of 4GB). I log RAM, load, net, user and connection stats to a database every minute (read paranoid).

I restarted MediaTomb and it's now marking objects as 'played' again. This behaviour may be something with MediaTomb specifically.

I'm still looking for a) what kind of uptime I can expect, and b) ways to improve reliability if anyone has any knowledge of Debian Live (using aufs) systems.

Cheers!
Randall