New to SANs looking for redundancy and guidance.

[mattgross]

Hello. A little background on my situation. I was doing IT consulting for small business with a consulting firm working with basic level stuff. Anything from workstation to basic servers. I'm now in a roll as the single IT administrator for a small credit union. When I arrived, they had just installed a Dell T410 as the DC and file shares running on the host and Hyper-V with an empty application VM and a Remote Desktop VM. Since I arrived, I have migrated the SharePoint server to VM and installed Exchange on it's own VM. Everything has been working fine up until a week ago when our only remaining physical server (the server that runs our main teller software) had a hicup during an extended power outage after the batteries drained over a weekend. Ended up with a corrupt registry that I was able to repair but found out I had a failing drive that wasn't completely dead, but it was enough for me to convert this box to VM as well.

Now we are up and running on a single server with all VMs and I'm starting to get worried. We have backups and warranty coverage, but there is still that fear in the back of my mind about what will happen if the server tanks? It currently runs everything so my main branch would be down for a day or so and all branches would be out with email as we currently don't have any backup provider. This has me thinking about Clustering Hyper-V with shared storage but dedicated SANs may be out of my price range. I'm also not familiar with this process as I've done more basic jobs in the past.

The current server is a dual processor Xeon E5607 with 48GB of memory with 6 500GB 7200RPM SATA drives in Raid 6. I was curious if I would be able to use this server and build another T420 with similar specs for a Hyper-V cluster using star wind. My question would be, do they have to be the same hardware (processor)? Would the combination of 7200RPM SATA and RAID 6 be a a nail in the coffin for trying to reuse this server?

If this wouldn't work, I spec'd out a T620 with the following:

12 3.5 HD chassis
2x Intel Xeon E5-2650 2.00GHz
64GB of memory
2x 300GB 15k SAS in RAID 1 for the OS
4x 600GB 15K SAS in RAID 10 (or possibly RAID6 if doable) for the StarWind
PERC H710
Broadcom 57810 Dual Port 10Gb NIC
Dual Hot-Plug 495W Power Supplies
Server 2012 standard

I'm also curious about the new version 8 with off site disaster recover. Can someone explain to me what how that works in terms of VHDs? I would only store the VMs on the starwind SAN so when going over to an offsite location, how exactly does that work? If I ended up needed new hardware at the main site, I'd take this T410 to that location but would I need anything else? I haev a 10x10 WAN link to that location, would that be enough?

[anton (staff)]

Few remarks...

1) Booting OS from a 15K rpm in a RAID1 is a waste of money. Boot from a pair of WD RE in RAID1 (even Windows software one would work just fine) and give these two SAS spindles to RAID10 to keep VMs on. They are going to love boost of IOPS. OS is doing boot mostly (tons of sequential reads) and paging (tons of a sequential writes). SATA is fine here and linear read and write is preferred to IOPS.

2) Parity RAIDs are bad friends for typical VM workload. Small writes need whole parity stripe being updated so you have HUGE overhead. Upcoming V8 with a log-structure file
system will fix this but LSFS is not a tool for everybody. So re-build RAID5 (and RAID6 with a few spindles has zero sense) into RAID10.

3) For a two hypervisor setup deploy StarWind running directly side-by-side with hypervisor (no dedicated hardware, pure VSAN scenario).

4) Upcoming V8 will have async replication so you'd be able to deploy a physical or virtual machine remotely and have an async replica of a the whole LUN. With a snapshot history.

[mattgross]

Thanks for the input.

1. Sounds good. With a single controller, I'd still need SAS drives, and a 1tb 7200 sas (smallest on dell's site at the moment) is 243.54 where as the 15k 300gb is 287.10 so I figured why not. This can be adjusted later.

2. Sounds good.

3. Native SAN sounds like the way to go.

4. What type of WAN requirements am I looking at for this? I currently have a 10 up 10 down connection but that is shared with voip traffic and internet traffic from our main site. Would that be enough or would I need up upgrade the speed or have a dedicated circuit just for this traffic?

[anton (staff)]

4) Depends on a) will you deploy LSFS or FLAT containers with StarWind and b) if LSFS how well dedupe will work on increments. In general something between 10 and 100 megabits should be fine. But it's VERY approximate.

[mattgross]

4) Depends on a) will you deploy LSFS or FLAT containers with StarWind and b) if LSFS how well dedupe will work on increments. In general something between 10 and 100 megabits should be fine. But it's VERY approximate.

I'm still new to SAN, what would the difference be with LSFS and Flat containers?

[anton (staff)]

That's our internal terminology so has nothing to do with a SAN concept.

LSFS is a Log-Structured File System. New (in V8) engine for handling VM specific workloads. Does thin-provisioning, snapshots, in-line deduplication and (by name) log-structuring. Eliminates random writes. Very good on stripe RAIDs and flash because of the big block size (4MB+). Good for primary VM workload.

Flat is thick-provisioned, has snapshots on a dedicated node, no dedupe and no LS. The same for everything. I mean works with any type of the workload.

So LSFS *maybe* better. Maybe not. Up to what you do.

4) Depends on a) will you deploy LSFS or FLAT containers with StarWind and b) if LSFS how well dedupe will work on increments. In general something between 10 and 100 megabits should be fine. But it's VERY approximate.

I'm still new to SAN, what would the difference be with LSFS and Flat containers?