[DRBD-user] database on drbd

Lars Ellenberg Lars.Ellenberg at linbit.com
Mon Mar 8 20:12:25 CET 2004

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


/ 2004-03-08 12:32:40 -0500
\ Carson Gaspar:
> 
> 
> --On Friday, March 05, 2004 21:49:18 +0100 Lars Ellenberg 
> <Lars.Ellenberg at linbit.com> wrote:
> 
> [ Re: database consistency ]
> 
> >As long as you use protocol "C", this is NOT true.
> >
> >If some user (file system, database, any program) wants to
> >complete a "transaction", it should make sure that whatever was
> >modified is on stable storage.
> >To do so typically (f)sync is used.
> >With protocol "C" in connected mode, this completes only
> >*after* sync completed on *both*, active AND passive node.
> 
> As I understand it, protocol "B" should also be OK, as long as the 
> secondary doesn't reboot at the same time, and may improve performance over 
> protocol "C".

What can go wrong, will go wrong, so
"Should be OK, as long as" == Is not ok.

And, though one might think it has better performance, and it was
*intended* to have better performance, benchmarks show that it
DOES NOT have better performance. We will eventually find out why,
and fix it.
Meanwhile: Use protocoll "C", and enjoy transactional semantics
*and* best performance for short distance links.

"A" is for long distance / high latency links.
Using "B" does currently not gain anything, but degrades
throughput, and loses transactional semantics, so should be avoided.

	Lars Ellenberg



More information about the drbd-user mailing list