LVM, SMVI, NFS and maybe some other letters

[djaquays0]

So, I'm not sure the best place to put this, but this section is probably the best.  We're in the process of implementing a new EHR and the vendor has signed off on us virtualizing everything.  Our current EHR does things significantly different than the new one, and was never authorized to be virtualized, so sits on a FC SAN by itself and isn't going to be much of a reference for the best way to move on.  This leaves us with several questions and hopefully you guys can provide some answers.

1)  The new system uses an FTP server to hold all scanned data and most of the documentation for patients.  They recommend that we allocate roughly 1TB of space to the data partition of the FTP server.  This server is going to be Linux (SLES10sp3) and virtualized in a vSphere 4 cluster with NFS datastores sitting on our NetApp FAS2050 and backed up via SMVI 2.0 and then vaulted via SV-SMVI.  Now for the questions.

  a)  Is better to create a single 1TB vmdk or to create multiple smaller vmdk's and use LVM to concatenate them?

    i)  If we should use multiple vmdks, can/should they be split across datastores.  We use cross blade LAG groups for both our NetApp filers and our vSphere hosts.  Spreading the data across stores mapped to different IP aliases allows the LAGs to load balance across NICs, so there may be some performance benefit.

    ii)  It looks like SMVI2 now does some sort of smart datastore spanning for backups, has anyone used this?

2)  The new system uses MySQL as its main database application with replicas for reads, all will also be running on SLES10sp3.

  a)  It seems the best choice for performance would be to make these data drives independant persistant and place them on datastores that are not backed up via SMVI, then have one of the replicas use mysqldump (or any other backup dump method) to backup the database and transaction logs to a separate partition that is on a datastore backed up by SMVI.  This should significantly decrease the backup storage needed for the application, as there will only be 1 copy saved instead of 4+.  Has anyone done anything similar to this?  Any thoughts on the process?

Thanks for any direction/assistance that can be provided.  This has been cross posted to both communities.netapp.com and communities.vmware.com

[BrendonHiggins]

Hi

1a. ~  How would you want to backup and recover your vms?  What is the I/O for each VM.  This should help you answer the question.

1a i & ii ~ Not sure enough to advise

2a ~ Not sure and have heard of people who have done it both ways.  We are MS SQL and have luns connected to the VM for the database.

Hope this helps {a little}

Bren

[djaquays0]

BrendonHiggins wrote:
Hi
1a. ~  How would you want to backup and recover your vms?  What is the I/O for each VM.  This should help you answer the question.
1a i & ii ~ Not sure enough to advise
2a ~ Not sure and have heard of people who have done it both ways.  We are MS SQL and have luns connected to the VM for the database.
Hope this helps {a little}
Bren

1a)  We use SMVI for backup.  My understanding is that SMVI can now backup a VM that spans datastores, though I have not tested this to see how it works.  I was hoping to find someone else who had.  If that works as I understood it to, the recovery would be the same as recovering any other SMVI snapshot.  As for I/O, we have no idea how intensive it's going to be.

2a)  We could still go the iscsi route with Linux.  I'd prefer to just use NFS.  My main concern is with this is backup.  I would prefer to only backup 1 copy of the database if at all possible.  In my head, not backing up the database drives at all and having a separate drive to store backups on only 1 replica makes sense.  I wanted to see if it made sense to anyone else as well or if I was looking at this all wrong.