Hello,
Having a ESX 3 farm connected to IBM SAN based on DSxxxx disk array, in order design a disaster recovery solution, we are going to sync every source IBM disk arrays to similar disk array in the DR plant, using a mix of PPRC and FlashCopy.
Now the question: before "closing" the target copy of the source LUNs, do I need to make a VMWare SnapShot of the source Virtual Machine?
If I do not create a SnapShot of the source Virtual Machine, what will happen if we close the PPRC while the source virtual machine is writing the disk? (Of course, every virtual machine hosting a database will be freezed in advance, like backup mode of oracle or quiesced of SQL)
Thanks.
Hi,
You don't need necessary to make a Snapshot, If you make it the copy will be more consistent, If you don't make it and close the PPRC the disks in the target are in the same state as if you power off the VM (not shutdown).
I'm using IBM's flashcopy for backups and I'm not making snapshots and I haven't had any problem to restore a VM.
I perfectly understand what you say and therefore my doubts!
If i close the PPRC without shutting down the virtual machine on the source plant, it is like a power off. If I restart the virtual machine on the copied LUN on the target plant, the virtual machine should do at least a checkdisk or try to boot in safe mode. Is it correct?
I I shutdown the source virtual machine and then close the PPRC I am sure that the target is PERFECT.
But I have another intermediate option: do not shutdown the source machine, but make a snapshot.
In the VMWare whitepaper "ESX_backup_wp.pdf" I read: "If your virtual disk files are stored on a SAN, you can use features supplied by your SAN vendor to create a copy of your production LUN, containing all virtual disks. These copies can then be sent to your backup media. With this method, you do not have to use virtual machine snapshotting functionality during the backup process because the SAN snapshot guarantees consistency."
The problem is that I am sure that VMWare snapshot flush cache and does other little things, while the PPRC or the FLashCopy does not care at all about the operating system running on the copied LUNs.
I am confused.
Steff,
I bet you are confused. This space is indeed pretty confusing. Also the terms "snapshot" refers to so many things/technologies that is easy to get lost.
I can't clame to be a backup/storage expert but I'd like to summarize based on my view.
VMware does provide a "snapshot" solution which has <nothing> to do with the storage snapshots. Basically what VMware does is file manipulation so when you issue a snapshot command what you get is a "somewhat" consistent picture of your running vm in the main vmdk files and all modifications being written to the vm since then on a redo external file. In the same directory of the same vmfs voulumes. This is a sw thing and you can do it even on locally attached SCSI/SAS storage if you want.
What most likely you are referring to is storage based replication (PPRC) or Flashcopy. As of today the usage of these technologies allows you to get "crash consistent" state vm's (with the only exception if you use RDM's in "physical mode" where you can use storage integrated utilities to create a somewhat consistent/quiesced image). Otherwise what you will get is a crash consistent vm which, unless you are running a banking accounts database and you are very concerned (for good reasons) about having a CONSISTENT snapshot of the data ....... should be ok (with all the advancements in file systems recovery technologies in the last few years a crash consistent vm should be a no-problem).
We do have customers that are doing this and they told me they have always had a 100% success rate in their tests. Some customers use a mix of PPRC / Flashcopy technologies to accomplish this. They always keep the PPRC link on and they take snapshots of the passive volume on the DR site to present to the actual ESX systems. This is good because you can do weekly/monthly tests of your DR setup without breaking the production PPRC link. Of course the disadvantage is that you need more storage on the DR site to accomplish this.
Just re-reading your last post .... so what you would like to do is take a (VMware) snapshot of the vm's and then break the PPRC (or issue a storage Flashcopy)? Well it's true this would allow you to get a "somewhat" consistent vm backup but yet I think you need to valuate the management burden to do something like this multi-level backup methodology or just "trust" a crash consistent picture of your vm's.
Hope this helps.
Massimo.