Calculating Usable Storage?

[Paul Hutchings]

I'm considering vendors for the replacement of our core ESX boxes and SAN.

HP are in the equation. I need to understand the EVA range better, and I trialled the P4000 VSA some time back and was very impressed.

What I'm trying to understand/recall, is how I convert the capacity of a brick/system as sold by HP into the amount of usable storage that I get?

For example the "MDL Starter SAN" (BK715A) is 16tb which is 16 spindles over two nodes.

I found an online capacity tool at http://www.tdonline.com/hp-lefthand/storage-calculator/ which says that will give me 12.5tb usable storage.

I thought it would be closer to 8tb because the fact there are two nodes means it does "Network RAID1" so that if I lose a node I still have a functioning SAN?

I've clearly overlooked something, I just don't know what?

[teledata]

That's actually my capacity calculator.

I've been pondering how to add some functionality to it without making it too confusing to use.

The reason it shows 12.5TB is that IS the total usable space of the cluster. Since your network RAID is configured on a per volume basis, so you may use Network RAID 10 (2-way replication) for critical volumes (thus consuming 2x the capacity), but for a test/dev environment you may use Network RAID 0.. Or for an archive you may use network RAID 5.

So if you plan on doing network RAID 10 for high availability for all volumes on the SAN cluster, take the cluster's total usable space and divide by 2.

The calculator is basically doing this:
1000GB * .913 (to right-size the disks) x 7 (lose capacity of 1 drive for hardware RAID 5 parity) = 6391 x 2 nodes = 12782 / 1024 = 12.48TB

[Paul Hutchings]

Thanks for making the calculator - saves a lot of time :-)

So if I have two nodes (i.e. the MDL Starter SAN) can I actually do Network RAID5 given that I only have two nodes?

I suspect in reality I'd be looking at 4 nodes anyway to give me the capacity I require in which case I definitely have the option.

Also I'm still re-familiarising myself (I did look at the VSA some time back) with the MPIO stack - does LeftHand allow multiple connections to a LUN or if you have a server with gig ethernet iSCSI NICs are you restricted to 1 LUN = 1gbps?

Thanks.

[Mark...]

Hi
For the P4000 range of storage you also have harware RAID. The default is 5 but your other choices are 1 or 6.
Once the nodes are configured into a managemenet group/cluster then you use the network RAID.
Network Raid 0 = 1 node
Network Raid 2 - now 10 = 2 nodes
Network Raid 3 - now 10+1 = 3 nodes
Network Raid 4 - now 10+2 = 4 nodes
Network Raid 5 = 4 nodes
Network Raid 6 = 6 nodes
Yo can mix all the network raid levels as long as you have the MINIMUM required for each level. You can also change the raid level with no penalty.
You can present one Volume/LUN to multiple servers as well.
Mark...

[Paul Hutchings]

OK I understand that each node is hardware RAID so the disks in each node can be configured to a hardware RAID level.

Where I'm still hazy is "best practise" on usable space if I only had 2 nodes?

Sounds like unless I can afford to lose a LUN, with 2 nodes I only have the option of "Network RAID1"?

[Mark...]

Hi,
Once you have decided on the Hardwrae raid for your nodes that will give you the capacity for network raid - well sort of! as you still lose a bit once you get the nodes into a cluster.
The calculator gives you the usable space.
With 2 nodes you could do network raid 0 or 10+1(2). If you did choose to do network raid 10+1 for all your disks/volumes then it would be equal to half of your available storage.
Mark...

[Paul Hutchings]

I think I'm getting it now.

So with two nodes you can only have network raid0 or raid1/10.

Presumably if you choose raid0 you can't force it to all be on a single node can you?

Sounds like 4 nodes is the way to go anyway.

[Mark...]

Hi,
Correct! Network RAID-0 will just stripe the data across ALL the nodes in the cluster - not possible to pick one node.
4 storage nodes gives you more possibilities and flexability like maybe multisite if you wish. You can do multisite with a minimum of two nodes.
Mark...

[teledata]

I did a modification to my storage calculator:
http://www.tdonline.com/hp-lefthand/storage-calculator/

I added a Network RAID configuration. You choose what percentage of the cluster will contain 2-way replicated (Network RAID 10) volumes. I chose NOT to include anything other than Network RAID 0 and Network RAID 10, otherwise I think it gets way too confusing.

It now shows:
Capacity (right sized)
Usable Storage
Total Data (repl and non-repl)
Replicated Data
Non Replicated Data
Also added a Rack Unit field to show rack space required.

I also updated to include the new models (with the 600GB SAS, and the 2TB MDL...)

Although with that new 120TB Capacity SAN It will truly reveal any math errors in my calculator! ;) Can someone from Boulder check my work? I don't have one of the 120TB Capacity SAN in my office to check for accuracy! Although I wouldn't turn it down if they sent me one for customer demos!

[Paul Hutchings]

Thanks Teledata, that looks very useful (I'm in the UK else I'd probably be giving you a call).

I guess this is a fairly open question, but how does the LeftHand performance scale?

For example if I were looking at 4 nodes of MDL SAS presumably because there are more spindles doing the work the performance should be better than 2 nodes even though there would be more "network raid" traffic between the nodes?

I have an idea of our current IOPS, but I don't know how to size LeftHand due to their RAID model(s).