Hi,
I have 2 servers which might need to go unattended for
several weeks at a time. They are currently running
vanilla 2.4.10 but my confidence in this (SMP) kernel
has been shaken when it spontaneously froze solid the
other day while I was viewing a web-page in Mozilla.
(And all I was doing was using the scrollbar on an
already-loaded page! No oops messages, no chance to
use Alt-SysRq, nothing.)
All that the servers would be doing would be
connecting to the Internet periodically using PPPoE
and DSL (with NAT), forwarding emails and performing
various CPU-bound tasks. They should both have ample
available memory and should not need to swap much, if
at all.
Does anyone have any kernel recommendations /
counter-recommendations, please? One server is SMP,
the other is UP, and both are Intel architecture.
Cheers,
Chris
__________________________________________________
Do You Yahoo!?
Listen to your Yahoo! Mail messages from any phone.
http://phone.yahoo.com
On Tuesday, October 02, 2001 11:05:02 AM -0700 Chris Rankin
<[email protected]> wrote:
> All that the servers would be doing would be
> connecting to the Internet periodically using PPPoE
> and DSL (with NAT), forwarding emails and performing
> various CPU-bound tasks. They should both have ample
> available memory and should not need to swap much, if
> at all.
>
> Does anyone have any kernel recommendations /
> counter-recommendations, please? One server is SMP,
> the other is UP, and both are Intel architecture.
PPP is not SMP safe in 2.4.x. You'll run into problems on any kernel
there. Even on single processor systems, you need the ppp patch in
2.4.9-ac16 or 2.4.11pre1.
Other than that, 2.4.10 + andrea's vmtweaks patch does well. 2.4.9-ac18 is
a good alternative.
-chris
> > Does anyone have any kernel recommendations /
> > counter-recommendations, please? One server is SMP,
> > the other is UP, and both are Intel architecture.
>
> PPP is not SMP safe in 2.4.x. You'll run into problems on any kernel
> there. Even on single processor systems, you need the ppp patch in
> 2.4.9-ac16 or 2.4.11pre1.
>
> Other than that, 2.4.10 + andrea's vmtweaks patch does well. 2.4.9-ac18 is
> a good alternative.
I'd probably apply them to 2.4.7 based trees as they have more history so
you can meaningfully answer the reliability question in statistical terms.
The others are too new to be 100% sure.
Also for remote systems configure watchdog support. That'll get you out of
so many disasters, software, hardware or other that its a godsend
On approximately Tue, Oct 02, 2001 at 11:05:02AM -0700, Chris Rankin wrote:
> Hi,
>
> Does anyone have any kernel recommendations /
> counter-recommendations, please? One server is SMP,
> the other is UP, and both are Intel architecture.
>
> Cheers,
> Chris
>
The latest ac kernels are stable as hell for me. I would run them on any
production machine.
--
Linux, the choice | What makes you think graduate school is
of a GNU generation -o) | supposed to be satisfying? -- Erica Jong,
Kernel 2.4.10-ac3 /\ | "Fear of Flying"
on a i586 _\_v |
|
Hi,
The UP server does the PPPoE, so that's OK. Has anyone
torture-tested any of the recent kernels? For
instance, I subsequently read a posting from Alan Cox
saying that 2.4.10 didn't survive overnight for him,
implying that he occasionally roasts penguins in some
kind of server-dungeon...
Cheers,
Chris
--- Chris Mason <[email protected]> wrote:
>
>
> On Tuesday, October 02, 2001 11:05:02 AM -0700 Chris
> Rankin
> <[email protected]> wrote:
>
> > All that the servers would be doing would be
> > connecting to the Internet periodically using
> PPPoE
> > and DSL (with NAT), forwarding emails and
> performing
> > various CPU-bound tasks. They should both have
> ample
> > available memory and should not need to swap much,
> if
> > at all.
> >
> > Does anyone have any kernel recommendations /
> > counter-recommendations, please? One server is
> SMP,
> > the other is UP, and both are Intel architecture.
>
> PPP is not SMP safe in 2.4.x. You'll run into
> problems on any kernel
> there. Even on single processor systems, you need
> the ppp patch in
> 2.4.9-ac16 or 2.4.11pre1.
>
> Other than that, 2.4.10 + andrea's vmtweaks patch
> does well. 2.4.9-ac18 is
> a good alternative.
>
> -chris
>
__________________________________________________
Do You Yahoo!?
Listen to your Yahoo! Mail messages from any phone.
http://phone.yahoo.com
> instance, I subsequently read a posting from Alan Cox
> saying that 2.4.10 didn't survive overnight for him,
> implying that he occasionally roasts penguins in some
> kind of server-dungeon...
I run brutal load test sets on the boxes. Not all -ac kernels survive them
either . Thats more useful as "it contains bugs" not "it will break under
normal load".
Alan
On Tue, Oct 02, 2001 at 07:41:24PM +0100, Alan Cox wrote:
> > instance, I subsequently read a posting from Alan Cox
> > saying that 2.4.10 didn't survive overnight for him,
> > implying that he occasionally roasts penguins in some
> > kind of server-dungeon...
>
> I run brutal load test sets on the boxes. Not all -ac kernels survive them
> either . Thats more useful as "it contains bugs" not "it will break under
> normal load".
>
Can you describe, or point to a description of your tests?
> Can you describe, or point to a description of your tests?
Cerberus is the main one I run. It provides suprisingly effective testing
for load triggered bugs
> > Can you describe, or point to a description of your tests?
>
> Cerberus is the main one I run. It provides suprisingly effective
> testing for load triggered bugs
Samium Gromoff said he ran Cerberus over 2.4.10 and it did quite
well
" Robert Szentmihalyi wrote:"
>
> > > Can you describe, or point to a description of your tests?
> >
> > Cerberus is the main one I run. It provides suprisingly effective
> > testing for load triggered bugs
>
> Samium Gromoff said he ran Cerberus over 2.4.10 and it did quite
> well
Yes i did, that was 11+ hours, but i`m not quite sure its long enough...
>
>