Continuous Replication on a large VM
-
Hi,
We have a large Windows Server 2012R2 VM that comes in at 4.5TB of storage across several disks (largest disk is 2TB) and has 180GB of RAM. It runs SQL Server. We were wondering about whether we could enable continuous replication on this to speed up recovery times in a failure situation.CR supports snapshot with memory - I was wondering what it has been tested to and if there is a limit to size on this and whether 180GB of RAM would end up being problematic if we wanted to do it hourly?
Does 'normal' snapshot in this process do quiescing with VSS? I'm a bit nervous that if we can't do the memory CR then we'll end up with corrupted databases effectively when it boots up. Oh this is running on XenServer 7.0 still although I want to get it over to XCP-ng ideally.
Thanks,
John
-
In theory, a snapshot shouldn't corrupt your database. 4.5TiB is very big, so the first CR will take very long. After that, it should be relatively fast.
Regarding memory, I would probably test without it, since 180GiB of RAM will take a LONG time to save.
So the best way is to experiment, see how long and then validate the CR on the destination VM.
-
olivierlambert Testing is a pain due to the resources required for this (!) and duplicating the setup, particularly as its something we were looking at as a proposal to the client atm.
I meant corruption could occur on boot up of the replicated VM as if it didn't have memory included then it would be booted like it had crashed effectively so there could be some corruption potentially from that effect.
Where does the saving of the memory occur to if its 180GB?! I.e. where do we need that 180GB of resource? Is the VM 'paused' whilst the memory is saved?
cheers
john
-
Yes, it needs 180GiB on a SR to store that, and the VM will be paused all along.
Testing your backup is important, even if it's once or twice a year.
-
Hello,
I have windows 2008R2 with SQL 2008R2 and a database of 4.3Tb, I use CR successfully (for 2+ years now), I do the transfer on a LAN link in http mode.For a transfer on a 1G LAN, you will have a transfer time of 60-72 hours. You must set in XCP-ng the time to kill the task for 72 hours minimum, in /etc/xapi.conf line 306 set: pending_task_timeout = 259200 # 3 days in seconds.
For a transfer on a 10Gb LAN you will have a transfer of 7-14 hours.
I don't use snapshot with memory but the recovery was ok, I had an event. -
Gheppy cool thanks. Good to see some real-world experience with a similar setup!
