FreeBSD unstable on Dell 1750 using SMP?

[Palle Girgensohn]

--On torsdag, augusti 18, 2005 14.30.44 +0200 Rutger Bevaart 
<rutger.bevaart at illian.net> wrote:

> It seems that updating our machine to 5.4-p5 (RELEND_5_4) has solved this,
> or at least made it occur less frequently. Our last reboot was after
> building and installing the new kernel and it hasn't gone down since.

Very interesting. We're still at 5.4-p1. The version bump fixes didn't look 
like they were addressing stability, only security, but why not...

> This
> is with SMP, ACPI and HT enabled on a Dell 1750 with two 3GHz Xeons.

Pretty identical to our system.

>  The
> 2850 has been rock-stable running 5.4-p3.

And you never ran previous versions on that system?

> Whatever is was, it seems to
> have been fixed around that time.
>
> Could be that your issues are amd64 related. We run the i386 branch
> because we need stable systems, not 64bit.

I have indications that the problems have occured equally on i386 and 
amd64, and that amd64 is considered stable, but that might not be quite 
true?


Regards,
Palle


> The issue still persists on 4.11 though. Can somebody explain what the
> ACPI fixes were around that time and if they will be backported to 4.X?
>
> Regards
> Rutger Bevaart
>
> On Thu, August 18, 2005 1:55, Palle Girgensohn said:
>>
>>
>> --On fredag, juli 15, 2005 15.15.24 +0200 Rutger Bevaart
>> <rutger.bevaart at illian.net> wrote:
>>
>>>
>>> hello list,
>>>
>>> For the past year we've been running several Dell PowerEdge 1750 servers
>>> on FreeBSD 4.10, 4.11 and 5.3. All these machines have dual Xeons
>>> running
>>> with HT enabled. This install has proven to be unstable in that the
>>> machine will reboot between 3 days and 170 days without apparant reason.
>>> No log is written. Other machines we have with a single CPU (HT enabled)
>>> do not experience this problem.
>>>
>>> As it is present in both 4.x and 5.x and googling the last year has not
>>> revealed similar experience I'm consulting this list. As all of these
>>> machines are productions machines that have a continuous load (not
>>> heavly
>>> load, but a light average - some peaks) it's not easy to experiment with
>>> HT setting etc. I dislike driving to the datacenter for locked systems
>>> with fubarred kernels ;-)
>>>
>>> The only error i've ever seen just before a reboot is "bge0: discard
>>> frame
>>> w/o packet header" on the 5.3 machine.
>>
>> Late comment while browsing the list for tips...
>>
>> No good clues, I'm afraid, but we have a 2850, and it is far from stable,
>> crashing within hours when running SMP, often but not always under high
>> load. Single CPU works like a charm. This is very annoying, to say the
>> least. See my posts on amd64@ around June 15.
>>
>> FreeBSD 5.4p1 (amd64). Dell 2850 with dual Xeon CPUS, EM64T.
>>
>> /Palle
>>
>>
>
>
> Rutger Bevaart :: illian.networks
>