Re: Short checkpoint durations

Neil Truby wrote: > Sorry about the earlier typos, which obscured the meaning: > > One of customers wants to use SAN replication between remote sites for DR > purposes. As recently discussed, and explained at length and in detail by > Martin Fuerderer, this will have the effect of allowing the DR database to > recover > its state at the last checkpoint. > > Actually the customer wants minimal data loss, but doesn't want to use HDR > for this. To do this we could set very short checkpoints - say every 60 > seconds, or even less. > > I wondered what experiences others have with checpoint intervals as low as > this? > > Do I get this right: any change on disk should be replicated by means of SAN to a remote site? So at any given point in time site B has the same on-disk informations as site A? Basically this should be the same as mirroring disks and later splitting the mirror. Shouldn't unbuffered logging provide you with the most current bits, so that B always is in the same state as A would be if it came to a fast recovery? No immediate need for frequent checkpoints. (I didn't see Martin's comments.)

"Andreas Legner" <andreas.legner@weihenstephan.org> wrote in message news:ec2kem$68l$1@svr7.m-online.net... > Neil Truby wrote: >> Sorry about the earlier typos, which obscured the meaning: >> >> One of customers wants to use SAN replication between remote sites for DR >> purposes. As recently discussed, and explained at length and in detail >> by >> Martin Fuerderer, this will have the effect of allowing the DR database >> to recover >> its state at the last checkpoint. >> >> Actually the customer wants minimal data loss, but doesn't want to use >> HDR >> for this. To do this we could set very short checkpoints - say every 60 >> seconds, or even less. >> >> I wondered what experiences others have with checpoint intervals as low >> as >> this? > > Do I get this right: any change on disk should be replicated by means of > SAN to a remote site? > > So at any given point in time site B has the same on-disk informations as > site A? > > Basically this should be the same as mirroring disks and later splitting > the mirror. > > Shouldn't unbuffered logging provide you with the most current bits, so > that B always is in the same state as A would be if it came to a fast > recovery? No immediate need for frequent checkpoints. (I didn't see > Martin's comments.) Yes, you're right, thanks.