Looking at best hard drive to choose... ESX 4.1

[abaack]

I'm looking at creating a new file server for our organization and looking to see what the best practice for large drives is now for 4.1.

We're running NFS for all VMs currently and using inside guest ISCSI with NetApp SnapDrive. My two options are:

1. Stay with the inside guest ISCSI (bad thing is it uses the same NIC for data traffic)

2. Create a new VMDK file within the datastore (either the same datastore or create a datastore just for large secondary drives)

This drive has the potential of growing very large... last time I had looked there was a 2TB limit on VMware drive sizes... is this still the case?

Thanks!

[vmroyale]

Hello.

This drive has the potential of growing very large... last time I had looked there was a 2TB limit on VMware drive sizes... is this still the case?

There is a 2 TB - 512 bytes limit for a single LUN presented, but a VMFS volume can be up to 64TB with the use of extents. How big is "large" in this case? Do you have any options on your hosts for presenting separate vSwitches out to the guest to isolate the iSCSI traffic from the regular network traffic?

Good Luck!

[golddiggie]

The 2TB-512B VMFS size limitation remains...

IF you need more than 1TB or 1-1/2TB to share to the entire group, then there are other options to consider. Probably the easiest would be to create a VM and then use the iSCSI initiator to connect it to an iSCSI target of the size you want/need (easily larger than the 2TB-512B ceiling)... You can also use mount points, if that's available to you with your SAN, as an option.

DAS will have the same size limit as SAN...

If you're really a glutton for punishment, you could connect several VMDK's on different LUN's to get more storage. But, you'll be best to present each LUN to the group needing it. That falls flat if one group needs more than 2TB-512B as a single share mount...

Depending on how many users, how much traffic (IOPS needs) and such the share is going to have, you'll want to use fast spindles, and plenty of them (for redundancy as well as performance)... For anything more than ~100 users with very light traffic load, I would go with 10k rpm (or better) drives and at least an array of 12-16 spindles (RAID 50 being preferred)... Also have enough connectors on the SAN for optimim traffic/bandwidth flow...

Just slapping drives together in a box might work in small environments, with light use, but when you start talking moderate to large (either use, TB or number of connections) use, you need to be more selective/carefull...

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.

[abaack]

The proposal was brought up to map multiple drives (partitions) to the guest and then use GPT to span the multiple partitions. It's possible the drive could grow past 1.5TB and maybe even 2TB+

Whichever way we go, what is everyone doing if they're running off NFS datastores these days? In-guest ISCSI or large VMDK files?

[golddiggie]

Small shops: When needing more than 1TB the volume is a mapped iSCSI LUN to the VM (only the VM has access to that LUN)... Done this for a 5.5TB volume at one location. 1TB shares/volumes were VMDK files... The only thing on that LUN was that one server, with it's files (presented a share of about 930GB)... That was almost two years ago now, and I don't think the group has come close to filling it yet (about 50-60% used)...

Large shops/corporations: SAN presented storage as mount points, so it really didn't matter. Wasn't limited by anything to do with what VMware's capabilities were. Then again, that was with EMC storage, so that's in a whole other class...

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.

[abaack]

Not sure what's considered a large shop, but we have appx 8 blades running around 40-50 VMs. Using a NetApp with around 15TB of storage.

Also, I don't understand the 'mount point' reference... are you talking about RDM for the VM?

[golddiggie]

46-50 VM's (even using 8 blades) is what I would call a small shop... Or small environment... I've easily had 40-50 VM's on three Dell R710 hosts (taking up all of 6U of rack space) run 50 VM's and be asking for more (even with HA configured for a single host failure)...

The mount point reference is for EMC storage... Not something you'll probably ever have where you are... Most of the time you won't see that unless there's really high demand, high dollars available for the hardware purchase, or the "powers that be" are willing to spend what it takes to do things like the big boys do (not a bad idea in many cases, but often not critically needed)...

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.

[DSTAVERT]

NFS doesn't have the same 2TB limitations for a datastore. Rather than create single very large drives I would consider using mount points to extend the file system. In Windows you can use either mount points or DFS.