[OpenAFS] Problems in stopping the afs-client

[Andrei Maslennikov]

On Fri, 18 Oct 2002, Dan Pritts wrote:

> On Fri, Oct 18, 2002 at 12:06:18PM -0400, Jim Rees wrote:
> > I think the default for afsd should be changed to "-nosettime", and a new
> > option "-settime" should be introduced for those who need it.  Afs has no
> > business messing with the clock.
> > 
> > Opinions?
> 
> That seems like a reasonable default. 
> 
> Complaining about the clock is one thing, changing it is another.
> 

  Hmm. Clearly one of the base things with AFS is that all clients 
  must be working in sync between each other, and with the servers.
  Apart from the obvious authentication problems (famous error "clocks
  are badly skewed"), mismatch in wallclock may lead to unpredictable
  results in the distributed environment. A classical example is that 
  of "make" that has to be run on client A that has a clock behind that
  of client B where the sources are actually being modified. So I dissent
  from the idea to eliminate or disable by default the built-in syncro
  mechanisms (which, by the way, work).  
  
  It is true that other means may be used for syncronization, 
  but then two services (afs and, say, ntpd) should necessarily
  be maintained together instead of one (afs) on every host involved.
  If one starts to use ntpd on client and runs afsd with "-nosettime"
  (option normally used only on servers), then it is essential
  to make sure that servers run ntpd as well - which is not always
  the case. I.e.: switch this option in afsd, and you are suddenly
  before a lot more issues to manage...
  
  Andrei.