kardel@medusa.informatik.uni-erlangen.de (Frank Kardel) (12/12/90)
Possible improvements for SUN clocks when using NTP... I don't know whether anybody has noticed this already, but I cannot remeber any discussion about setting _noprintf to 1 in the Sun kernel. Last night, when our server clock took off again (1.8 secs), I noticed an excessive count of ieX: no carrier (ok: our local problem), but what counts is that these printf's of the kernel are generated at virtually any interrupt level. This means, that the clock can stop for quite a while (several 100ms in case scrolling on the console is necessary). There is a variable in the kernel to turn off the kernel printf to the console. Since we have a graphic console (it might also hold for RS232) this had an almost instant effect of stabilizing the kernel clock to within the clock tick granularity. The kernel messages are still reported to syslog so there are not lost! They are just not printed on the console. The clock improvement for our Sparc server was drastic. It's not tested yet with SS1's but improvements can be expected there too. I do not know whether this will help, if somone uses the console without a window system - we have not tested this either. If enough interest exists (by email please) I'll be willing to put an updated version of Dennis Ferguson's tickadj into our ftp server. Hope this will help Frank Kardel (kardel@informatik.uni-erlangen.de) Disclaimer: setting undocumented kernel variables might void your warranty 8-)