vSphere 4.1 RDP vs. VMDK for fileserver solution

[dchmax]

We are working on migrating our fileserver to a VM.  W2K3 ENT R2 x64 SP2.  The total storage neede will be 4-5 tb and we plan to split between 2 VM's.  Questions.....

RDP vs VMDK?

From what I've read I should avoid RDP if possible and there is very little performance increase.

If I go VMDK I'll be adding the disk to W2K3 and creating a dynamic disk so I can create a larger volume.  My concern is around snapshots and vmdks that have been strung together to create 1 dynamic volume in windows.  How much space should I leave free on the VMFS for snapshots.  Can I disable snapshots on the data disk but snap only the OS volume?

Any pointers, things to avoid is very helpful.  I figure someone has been there done that.

David

[DSTAVERT]

Unless you absolutely need a contiguous disk I would avoid it. Use smaller VMDKs and reconstruct your file tree using DFS or mount points. Isolate your historical data (perhaps read only) and your most critical data.

[dchmax]

When you mention smaller VMDK's, what size would you be refering to.  I'm looking at getting 500 gb luns from the storage team and carving up 250-475GB VMDK's, connecting them together to create a large volume.  I'm in a pinch in the sense that time is ticking and little time to re-engineer the setup.  We looked at Distributed File System as a solution but was afraid of file changes not replicating in time since there are many shared files, atleast under W2K3.  What would a mount point buy me over a logical drive?  Thanks for your feedback.

[DSTAVERT]

A mount point would allow you to create an identical file tree using multiple disks. The disk is mounted to a folder rather than as drive letter. For DFS you can create your new server and your DFS structure including the old server. Use replication to migrate the data. Left out a lot of steps here but.

[dchmax]

I ended up with something I'm a bit more confortable with.  I think DFS is a great option but in the time frame I was working with I would not have had time to test.  I did end up sticking with VMDK's and chaining 2 to 3 VMDK's together to create a dynamic disk large enough for our data.  So far we have been getting great performance.  Backups are being down via a traditional agent.  Thanks for your assistance.

[DSTAVERT]

You will need to be very careful. In a heartbeat everything can go wrong with dynamic disks especially spread over multiple datastores.

[dchmax]

Thank you, it has been in the back of my mind.   The only reason I was ok with it was that we have a very reliable storage array (XP24000) and have had very little to no issues.  I've set the vm disk used in the construction of the dynamic disk to Independent to keep them from getting snapshots.  We are using double-take to replicate to a failover set.  Keeping a close eye on the health of the system.

[DSTAVERT]

Ok then I will keep my fingers crossed for you.

[BradDHancock]

We present several 1tb LUN to our hosts, and use dynamic disk to allow more than 2tb volumes to the VMs. We use snapshot with VDDK and don't have any problems with backing up the data.

What are some of the issues that can happen with the dynamic disk? I always here this is a bad practice, but ive never seen any example of why it's bad.

Brad

[Rumple]

Dynamic disks are software RAID - and Microsoft RAID to boot.

an equivalent configuration is basically trusting your enterprise to a promise raid controller (which is also just software raid).

Anything software based thats there to work around an issue is not designed for reliability or stability..its a quick fix to some other issues

The equivalent in vmware world is extents...sure you can use them...but..shudder...

All it takes for dynamic disks to go sideways is a hard crash or something else and the volumes go offline...like for instance something happening where your VM loses contact with the SAN for a minute (which WILL happen, not if...)  That can cause the volume to corrupt pretty easily.

DFS is works ok if you've tested it, but alot of programs will actually puke on DFS volumes so you need to test it rather extensively.

[BradDHancock]

I understand what you are saying, but like you said it is such a quick fix.

If DFS isn't always a solution how are people providing clueless developers more than 2tb of space in a virtual environment? 

We have two boxes where we are giving them 4tb as 1 drive, so it's not too late to switch the technology; I just want to make sure to do it right if I redo it.

thanks for the response.

[depping]

Many customers I have use NFS or iSCSI but than in-guest to provide their customers with extra storage. Kind of an extra service they offer. This iSCSI / NFS as a service is usually a different storage tiers as it has different requirements etc.

Duncan (VCDX)

Available now on Amazon: vSphere 4.1 HA and DRS technical deepdive