Stability problems with low profile Atom machine

W

wildtollwut

Hi,

I own a fit-PC2 which is a small PC with Atom processor, 1 GB RAM, HDMI video output consuming only about 5 watts. I primarily use it as a NAS system but also as mail, samba and LDAP server. At first I installed FreeBSD 8.0 (i386) about a year ago using ZFS RAID1 as the root filesystem and GELI encryption (I know, this is a bit overkill for a 1 GB machine but it ran just fine as headless server without X etc).

Then, after upgrading to FreeBSD 8.1 the system began to get unstable for reasons I cannot pinpoint. After a few days or even a few ours the machine won't respond to a ping, let alone an ssh connection request or anything else. My problem is, that I can't attach a monitor since I only own old VGA-connected screens and don't want to buy an expensive HDMI to VGA adapter. I had hoped resolving the issue by upgrading to 8.2 and upgrading my zfs pool to version 15 but no luck. It still decides to go out of business randomly. Tuning zfs for low memory consumption didn't change anything, unfortunately :(

Skimming through various log files in /var/log I couldn't find anything suspicious at all. The logs just stop at a point :/ I suspect some kind of kernel panic (caused by whatever) which I cannot see because I can't attach a monitor. The temperatures are within normal levels as far as I checked.

So how could I get to the source of my problems with only headless tools? Thanks a lot for any suggestions.
 
Don't you have a monitor with DVI? HDMI is pretty much the same signal and a HDMI->DVI converter is cheap, only a few euro.
 
Unfortunately the Atom has no AMD64 capabilities (it's a Silverthorne Z530 I think). And all my monitors are VGA only, so no DVI ;) I know I should buy one and will do some sometime soon but there has to be a possibility to detect the issue without a monitor?
 
Your system is almost certainly panicking. 32bit address space and only 1GB of RAM is just asking for trouble with ZFS.

1GB simply isn't enough and 32bit platforms also need a lot of additional tuning to make ZFS run stable.
 
jem said:
Your system is almost certainly panicking. 32bit address space and only 1GB of RAM is just asking for trouble with ZFS.

1GB simply isn't enough and 32bit platforms also need a lot of additional tuning to make ZFS run stable.
Click to expand...

That also depends, I use 1GB RAM on my dual core (Intel T8100) home made NAS with FreeBSD amd64 8.2-STABLE, use ZFS/samba/nfs/transmission/shell/... and everything works like a charm without any problems, even the ZFS pool (mirror of 2 * 2TB drives) is 91% full.

So as you can see, your mileage may vary:
Code: 
$ uptime
 8:57PM  up 30 days, 10:58, 11 users, load averages: 0.10, 0.06, 0.02

$ zpool list
NAME      SIZE   USED  AVAIL    CAP  HEALTH  ALTROOT
storage  1.81T  1.66T   154G    91%  ONLINE  -
 
After connecting a monitor and not seeing anything I was more or less out of ideas :/
Then, after a while (12 hours or less) and the system hanging without an error message, I rebootet and tried to continue working normally. I noticed that the system had very long response times to my keyboard inputs... I had to press some buttons 10 to 20 times to get the desired action. Besides from that it seemed to work somehow. top showed me 20% - 30% interrupt load which caught my attention.
I restarted in no-ACPI mode and since then the system runs as stable as a rock (ok a small rock :)) So ZFS doesn't seem to be the culprit :) I have no idea what exactly causes this, but as I'm not dependent on any ACPI feature (it's not a desktop) I can live with this solution.
The strange this is that a change in ACPI must have happened between 8.0 and 8.1 because I don't remember having any stability problems on 8.0.
 
Me again :) Upgraded to 9.0-RELEASE recently and ran into new problems after a year of realiable operation with 8.2-STABLE and ACPI disabled. Now normal boot with ACPI still gives me very slow response times, a constant load of 1.5 and eventually system freezes - same phenomenon as with 8.1 and 8.2. However, since 9.0 booting without ACPI isn't working anymore :| It just stalls after the first few initializations, three or for lines after APIC if I remember correctly (will check this again tomorrow) so I'm between having a slow server for max three days or none at all.

I have not yet tried updating to the latest BIOS, but will give this a shot in the next days (not much hope though). I'm going to try reading a bit on debugging ACPI, but I'm also not too confident this may resolve my issues.

Any other suggestions? I have seen a few other posts in the net regarding ACPI stability problems but most of the time the solution was to disable it completely which for some strange reason isn't working for me.
 
BIOS update didn't do anything. A solution that works (so far) is to disable the APIC via
Code: 
hint.apic.0.disabled = 1
in /boot/loader.conf. I can now boot with ACPI. No interrupt storm or other issues seem to be present.
 
You must log in or register to reply here.
Back
Top