Re: TODO for etch ?
I demand that Andrew Suffield may or may not have written...
> On Mon, Jun 20, 2005 at 07:22:08PM +0100, Darren Salt wrote:
>> I demand that Florian Weimer may or may not have written...
>>> * Olaf van der Spek:
>>>>> You should set the clock using NTP *before* starting any daemons. Most
>>>>> daemons don't use monotonic clocks (I'm not even sure if Linux supports
>>>>> them at the required level), and some of them fail in strange ways if
>>>>> the system clock warps.
>>>> Doesn't Linux or NTP support gradually changing the clock exactly to
>>>> avoid such warps?
>>> Gradually skewing the clock doesn't exactly work that well if the offset
>>> exceeds a few minutes. You don't want to run with a wrong clock for
>>> hours or even days.
>> Maybe ntp, ntpdate etc. should recommend adjtimex?
> adjtimex is pretty near useless and should not be used. It can make things
> worse rather than better, especially with the clocks in modern boxes (which
> are grossly inaccurate).
I find that it improves matters.
> Under *no circumstances* should adjtimex be used at the same time as ntpd.
That's not a problem here - ntpdate is run regularly, and I don't have a
permanent net connection so (AIUI) ntpd isn't really practical here, and I
don't run it.
> The clock will jitter all over the place because they won't agree and will
> keep slewing it in opposition to each other.
(There's a hint of 'adjtimex is a daemon' about that, but let's ignore that.)
If altering the kernel's system clock rate variables while ntpd is running is
the cause of the problem which you mention, then perhaps it's worth
mentioning that in the ntpd and adjtimex man pages. I don't see that altering
them without a running ntpd should cause a problem (so long as the values
used are good), even if ntpd is then started.
| Darren Salt | nr. Ashington, | linux (or ds) at
| sarge, | Northumberland | youmustbejoking
| RISC OS | Toon Army | demon co uk
| Say NO to software patents
File not open, 0:1