Reproducable time problem with SMARTD & XNTPD

Reproducable time problem with SMARTD & XNTPD

[Nohez]

We have a Intel 875WP1-E motherboard with 2 onboard SATA ports.
In addition to the two 80GB SATA disks attached to these ports we
have three 160GB IDE HDD and one 300GB IDE HDD attached to 2 Promise
controllers (PDC20269). The server has xntpd configured on it to sync
time from remote timeserver. It has been running without any problems
for more than 2 months.

Last week we installed SMARTD on this box. SMARTD was configured to
monitor the 6 HDDs. The time on the server started to go out of sync
with the remote time server. And the difference increases over time.
As soon as smartd daemon was shut XNTPD started to get the server time in
sync slowly.  We tried this 2-3 times and have noticed the same results.

Has anyone faced a similar problem ?

Nohez


Linux Distro: SuSE Linux 8.2 Professional
Kernel      : k_smp-2.4.21-108 (SMP because of Hyper-threading)
XNTPD       : xntp-4.1.1-177
SMARTD      : smartmontools-5.1.4-20

Re: [smartmontools-support]Reproducable time problem with SMARTD & XNTPD

[Bruce Allen]

> Last week we installed SMARTD on this box. SMARTD was configured to
> monitor the 6 HDDs. The time on the server started to go out of sync
> with the remote time server. And the difference increases over time.

This is the first report we've had of this type.

> SMARTD      : smartmontools-5.1.4-20

Could you please try this with the current 5.20 release of smartmontools?

Bruce

Re: Reproducable time problem with SMARTD & XNTPD

[Nohez]

I removed the 2 SATA disks from /etc/smartd.conf and found that the server
time remains in sync. The moment I add any one SATA disk the time starts
to go out of sync.

So looks like SMARTD + SATA combo is causing the system time to go out
of sync.

I am also trying to relate this problem with a similar problem that
we are facing on four HP Netserver LH6000. System Time behaves
erratically.  We noticed that all time related apps (sendmail,
ping, top, cron etc) stop. We noticed that time goes forward &
backward in seconds only.  These servers do not have any IDE
disks and does not have SMARTD running.  Detailed thread at:
http://marc.theaimsgroup.com/?t=104393747300002&r=1&w=2&n=6

All servers are running a SMP kernel and are facing a system time problem.

Nohez


On Mon, 13 Oct 2003, Bruce Allen wrote:

> Time to write back to linux-kernel.  I have no idea what's wrong.
> Bartlomiej, any ideas?  It seems that running smartd with SATA causes
> the system time to go out of sync...
>
> Bruce


On Sat, 11 Oct 2003, Mark Hahn wrote:

> > Has anyone faced a similar problem ?
>
> I doubt it; sata is still pretty rare, especially in combination
> with xntp.  it's a fascinating problem though - the only explanation
> I can think of is that smart queries eat your timer interrupts.
> that shouldn't happen of course!  I wonder if 'hdparm -u1' would
> help or even make sense for you...


On Tue, 23 Sep 2003, Michael Marxmeier wrote:

> Hello Nohez,
>
> I found your posting in the LK archives. Did you ever resolve
> this problem?
>
> We seem to have a similar problem with a customer using a
> dual CPU TC4100 NetServer.
>
> The effect is that timer interrupts could slow down to about
> once in 30 seconds rather than 100/sec. This seems to happen
> sporadic.
>
>
> Thanks
> Michael