NTP maxerror patch for 2.x

• Patch Index 2.0 Unofficial Patches
  
• Previous Patch: Fix for Signal 7 on kernel compile
  
• Next Patch: Patch for the MO-drive Fujitsu 2513a - Sector size = 2048
  

I have an unimportant patch for the Linux 2.0 kernel, thus it can be included in the next versions of 2.0 and 2.1. I haven't checked 2.1, but I guess a appropriate patch is not there.

The patch just increases the largest accepted value for "maxerror", a kernel variable used for NTP time synchronization. As the previous value was according to Dave Mills reference implementation (512ms), the new value is much larger (about 16s). The value is the maximum acceptable value and the initial value after booting.

Experience has shown that for a typical Linux NTP client served by a typical NTP server (i.e. stratum 2 over the Internet) the estimated error is around 20ms. Also with the typical polling interval of 1024s, the maximum error exceeds 512ms from time to time, thus declaring the clock as unsynchronized.

Another sitiuation is initial synchronization where the first estimated error is usually about 1s.

The value itself is a matter of discussion, with only two restrictions: It may never overflow (into negative or back to zero), and it should be reasonably large. I have made an estimation which should really be acceptable (stratum 3, rather poor reference clock, high network delay).

For those people not concearned with NTP the patch does change nothing.