Hi, thank you both for your answers. > -----Original Message----- > From: Flemming Frandsen [mailto:[EMAIL PROTECTED]] > Sent: Mittwoch, 20. November 2002 21:36 > To: [EMAIL PROTECTED] > Subject: Re: Problem with hibernate: 7.4 Beta randomly crashing > (Windows) > > > Stephen Gutknecht (SAPDB) wrote: > > You obviously don't work in the same areas I do :) Our > serves must be > > synced to a known references constantly. >
Ours too, but in differential mode... see below > Naturally. > > On Linux (and I expect other Unixen) the ntpd daemon slows down or > speeds up the system clock for a while depending on the error when it > got the time from the server, this means that time never goes > backwards jumps, unless the clock is many hours out of sync. > On SAP backward jumps are completly forbidden. If time is to far behind a reboot is enforced... > I imagine something similar must exist on windows. > > http://www.eecis.udel.edu/~ntp/ntp_spool/html/ntpd.html > > > Eventhough it is clearly possible to keep accurate time and > avoid jitter > it would be good if sapdb could fail with a little more > verbose message > or maybe log a warning in stead? > > It seems heavyhanded to blow up just because the clock is defective. > We are working on a more defensive strategy, accepting the defective clock and other sources for unexpected time shifts. The idea is to detect the defect already in the clock routine itself and only issue a warning message if a time shift happend. This seems exactly what you want. To detect the time shift, we will always use the last time result as reference and sum up the shifts detected... This way we will build a monotonic accending time reference for the caller of the routine. The advantage is that in case of hibernation, the complete hibernation period is disregarded, which is IMHO almost the natural behaviour. But the SAPDB developers use the vmake in a two step hierachy of source directories splitted over the network. The 'platform local' directory is almost always empty, but the developers source directory is synced on his local machine (unfortunatly almost always a W2K workstation). Our UNIX servers are all synchronized using NTPD in 'differential mode' slowing down or speeding up the system clock as Flemming described it. We really have fun with make dependency if the NTPD fails... vmake is more defensive and allows a maximum jitter of 5 minutes, but the results of a differential makes are unpredictable if a time shift happens. For that reason the factory is always forced to do a scratch make for a production version... CU jrg _______________________________________________ sapdb.general mailing list [EMAIL PROTECTED] http://listserv.sap.com/mailman/listinfo/sapdb.general
