[DRBD-user] Diconnected drbd as very fast backup solution ?
Christoph.Adomeit at gatworks.de
Thu Oct 15 14:13:21 CEST 2009
I have a customer with large amounts of data (5 TB) which has to be
backed up in a very short time frame and where the backup should have
as low as possible impact on system performance during backup. And yes,
budget is very limted.
Failover and Cluster Features are not needed at this site.
I have tested a lab setup where the customer data is mirrored to
a second storage box with drbd. The drbd on the secondary node is mostly in
Disconnected state to not degrade performance of the primary drbd Server.
I have found out that the amount of customer data tat changes daily
is not that big and that it only needs 15 minutes sync-time after
reconnect until the drbds are fully in sync. Then i could snapshot the data
on the second drbd and disconnect the drbds again.
I can then backup the data from the secondary drbd box and have a lot of
time for this and it does not have any performance impact on the primary
and disconnected drbd storage.
Even better: The second drbd box can be quite slow and crappy, it only
has impact for the short sync time.
I think this is a good and creative usage of drbd for certain use cases.
What I want to know: is there anything wrong with running a drbd
cluster where the second node is disconnected most of the time ? Will
some memory or buffers run out of space or such things ? Is the performance
of a disconnected drbd disk far slower than the performance of a native
lvm volume ? Is there any other reason why I should not use this
setup in this use case ?
Amtsgericht Moenchengladbach, HRB 6303
Christoph Adomeit, Hans Wilhelm Terstappen
Christoph.Adomeit at gatworks.de Internetloesungen vom Feinsten
Fon. +49 2166 9149-32 Fax. +49 2166 9149-10
More information about the drbd-user