Hello, list.
I've searched the archives superficially but I was unable to get a glance
of any promising subjects WRT this matter.
We're setting up a failover storage system in our company, and we thought
that, instead of having the failover host online all the time, using some
sort of realtime synchronization thing (nbd or rsync or whatnot), we would
prefer it to be brought into sync every n hours, with the disks shut off
in the meantime, which would not only preserve disk lifetime, but also
simplify maintenance and improve safety of such backups if we were able
to copy the second latest set of data upon every sync. Having the failover
experience a n hour lag is no issue, it is more important to us to have
a failover storage location which is at least a couple of degrees more
reliable than the one we're using for data and is easily interchangeable
with the current one.
Now, after having had brainstormed for a while, we came to the following
idea: would it be possible to have xfs_logprint dump the transaction logs
between event X and event Y from a log device set at filesystem creation
time, then these logs be transferred onto another machine and replayed
there somehow ? Machines would be identical, both hardware and software-
-wise, there would be mechanisms for ensuring the standby machine does
never get modified from userland (-o ro).
We feel strongly that this would be the cheapest way of synchronizing two
storage devices that both have capacities of a couple of TB of data,
especially since the volume of transactions on these devices would not
have been too high, but it is important to avoid any kind of scan-like
difference checks in the manner of rsync and such, as this would be a
performance hit too great to handle.
Thank you very much in advance,
--
Grega Bremec
Senior Administrator
Noviforum Ltd., Software & Media
http://www.noviforum.si/
pgpIQjUgc4hMl.pgp
Description: PGP signature
|