I came across this article last night and was honestly pretty shocked, it talks about the limitations of the new VMware Virtual Storage Appliance that was released along side vSphere 5. I think it is the second VSA to receive full VMware certification after the HP/Lefthand P4000.
The article states
[..]
Plus, this capacity will be limited by a 75% storage overhead requirement for RAID data protection. Thus, a VSA consisting of eight 2 TBs would have a raw capacity of 16 TB, but the 75% redundancy overhead would result in a maximum usable capacity of 4 TB.VMware documentation cites high availability as the reason behind VSA’s capacity limitations: “The VSA cluster requires RAID10 virtual disks created from the physical disks, and the vSphere Storage Appliance uses RAID1 to maintain the VSA datastores’ replicas,†resulting in effective capacity of just 25% of the total physical hard disk capacity.
[..]
That’s pretty pathetic! Some folks bang on NetApp for being inefficient in space, I’ve ragged on a couple of other folks for the same, but this VSA sets a new standard. Well there is this NEC system with 6%, though in NEC’s case that was by choice. The current VSA architecture forces the low utilization on you whether you want it or not.
I don’t doubt that VMware released the VSA “because they could”, I’m sure they designed it primarily for their field reps to show off the shared storage abilities of vSphere from laptops and stuff like that (that was their main use of the Lefthand VSA when it first came out at least), given how crippled the VSA is(it doesn’t stop at low utilization see the article for more), I can’t imagine anyone wanting to use it – at any price.
The HP Lefthand VSA seems like a much better approach – it’s more flexible, has more fault tolerance options, and appears to have an entry level price of about half that of the VMware VSA.
The only thing less efficient that I have come across is utilization in Amazon EC2 – where disk utilization rates in the low single digits are very common due to the broken cookie cutter design of the system.
Who would buy this? Better yet, who has and do they still have a job.
Comment by gchapman — December 2, 2011 @ 12:32 pm
One more thing, Netapp market share appears to be shrinking, though I’m looking at them for a rip/replace of some aging Equallogic rigs. I do like to see the “others” gaining market share, for far too long the choices available have been 3 or 4 at best and its nice to see that IT shops are not just buying because of the name:
http://www.theregister.co.uk/2011/12/02/idc_storage_tracker_q3_11/
Comment by gchapman — December 2, 2011 @ 1:11 pm
Curious are you looking at the NetApp boxes for their iSCSI support or are you moving off the iSCSI equallogic to NAS on NetApp?
Myself just last night I started looking around to see what would be good to provide file services for my company – amount of storage required is small so I really wanted a Virtual storage appliance (which is how I came across the article in the post).
At least at this point I’m quite excited to try out the Nexenta storage appliance, I exchanged a couple emails with one of their system engineers and early signs look good they are fully willing to support a 2-node HA cluster running in VMware guests on top of my existing 3PAR storage. The cost is quite good too – roughly $12,500 for the first year (up to 8TB – the product is licensed based on capacity) and about $2,000/year for support from year 2 onwards (at current pricing). While some of the stuff will be “mission critical” of sorts for the most part I can easily live with this lower tier of storage access vs dedicated appliances for the sort of scale my current company is at. It’ll be another few weeks or even a month before I start testing(hardware is not yet installed) but I do like the features and the price point (and the software back end is fairly robust), and I can leverage the same tier 1 storage hardware as everything else.
HP’s entry level pricing for something that’s remotely comparable is about 5x more (same for NetApp). Which for this use case is overkill.
Comment by Nate — December 2, 2011 @ 3:08 pm