Forum - Aiccu problem caused by NTP clock drift (OT, really an NTP question) :: SixXS

Aiccu problem caused by NTP clock drift (OT, really an NTP question)

Shadow Hawkins on Monday, 22 September 2008 10:53:56

I occasionally lose my aiccu tunnel due to clock drift on my computer. I've set the machine up to sync using ntp against two NTP servers, but I get the impression that this is done either very seldom or only at startup. Since uptime on the machine is usually as long as local electricity company allows this means my machine now and then drift by more than 2 minutes. I guess I could write a cron entry to force my computer to resync using ntpdate every few days, or supervise syslog to catch entries like aiccu: [AYIYA-tun->tundev] [82.96.56.14]:5072 : Time is 121 seconds off for :: So... 1. Can this problem be solved by reconfiguring NTP or do I need to manually take action as above? At the moment I'm trying to digest the NTP docs, so any help or pointers are appreciated. 2. If this happens, and I sync up the system clock using ntpdate, will aiccu sense this or is my tunnel "dead" after falling outside the clock "window"? 3. Any other suggestions or pointers to good clock-keeping?

Aiccu problem caused by NTP clock drift (OT, really an NTP question)

Jeroen Massar SixXS Staff

on Monday, 22 September 2008 11:09:59

Fix your computer's clock. Most likely there is a kernel bug here (some platforms like NSLU2 need kernel patches for them to behave properly), or you have other clock issues, or devices attached which cause clock skew. NTP won't sync more than X time per period, depending on the NTPd you can disable this though; but according to the NTP folks this time-jumping is undesired.

Aiccu problem caused by NTP clock drift (OT, really an NTP question)

Shadow Hawkins on Wednesday, 29 October 2008 02:15:38

NTP should never cause time-jumping. NTP will make the clock "tick" faster if the clock is lagging behind or "tick" slower if the clock is ahead real time. Time will never jump backwards.

Aiccu problem caused by NTP clock drift (OT, really an NTP question)

Jeroen Massar SixXS Staff

on Wednesday, 29 October 2008 10:22:53

Which is why there is a special flag for doing it anyway (-s for most NTP servers) ;)

Aiccu problem caused by NTP clock drift (OT, really an NTP question)

Shadow Hawkins on Wednesday, 29 October 2008 11:12:24

Right! Unfortunately, if NTP servers are unreachable, it seems NTPD (at least in OpenBSD) will hang up forever. At least, this is what happened to me one of my OpenBSD machines where I was supplying "-s" to NTPD: the network cable was accidentally unplugged and the system would stall during boot trying to set up the clock :-/

Aiccu problem caused by NTP clock drift (OT, really an NTP question)

Shadow Hawkins on Tuesday, 23 September 2008 10:22:51

NTP can adjust the frequency of your clock, but not the 'tick' value. If the clock in your computer is stable but always running ahead of time 30 seconds/day, then decreasing TICK with 30/8.64 ~= 4 (thus TICK=9996) should allow NTP to fine-adjust with FREQUENCY the remainder and have a stable and accurate clock in the end. More info about this can be found in adjtimex manpage. (http://www.linuxcommand.org/man_pages/adjtimex8.html) After coarse tuning of your computer clock, the fine tuning of ntp will succeed. Ofcourse I'm assuming here that your OS is linux, with another OS this may be completely different:)

Aiccu problem caused by NTP clock drift (OT, really an NTP question)

Shadow Hawkins on Tuesday, 23 September 2008 21:04:56

Thanks for the reply, I've been looking through the ntp.org site, but the state of the documentation is a bit ... convoluted. I've found out about correcting clock drift but haven't had time to work my way through the docs yet. OS is linux, running on a very old pentium III 733 MHz machine. Measurements so far indicate clock drift is pronounced but stable, so I'll go ahead using your info. Again, thanks.

Please note Posting is only allowed when you are logged in.