Quanah Gibson-Mount wrote:
--On Wednesday, April 27, 2011 7:41 PM +0200 Michael Ströder
<michael(a)stroeder.com> wrote:
> Quanah Gibson-Mount wrote:
>> From what I see on the list, people who use normal syncrepl
>> do not have reliable replication.
>
> Frankly this sounds scary!
> Especially since you have more knowledge of the internals here... :-/
>
>> With delta-syncrepl, the changes are read directly out of a changes
>> database, in order. The gains are significant.
>
> IIRC with delta-syncrepl all write operations are serialized. Yes?
Correct.
Couldn't this be the cause for delta-syncrepl to seem more reliable than
normal syncrepl (without slapo-accesslog)?
This improves performance significantly over normal syncrepl.
It may seem counter intuitive, and yet it reduces the contention in the
underlying database. I tested this heavily in the past. ;)
I've read the results of your tests before. Many thanks for that.
Ciao, Michael.