StarWind Software

We have 2 Dell R620's on order. They come with:

1.2TB, SAS 6Gbps, 2.5in, 10K RPM Hard Drive x 4
PERC H310 Integrated RAID Controller
Broadcom 5720 QP 1Gb Network Daughter Card

I would like to make a hyper v cluster but we don't have a san so I have come across starwind virtual san.

I have read that the H310 controller has issues so I'm trying to find out if I can change the order and have a H710 instead. Will the H710 with 512mb of cache be ok?

I'm a little confused on how to set this up.

The plan was to have raid 10.

So would I create a system partition install windows then leave the rest for starwind?

Is it possible to buy a 10gb nic for each and connect them directly together without going through a switch? Then say use a 1gb nic via switch link for redundancy?

On dells site there are 2 options for 10gb cards:
Intel Ethernet X520 DP 10Gb DA/SFP+ + i350 DP 1Gb Network Daughter Card [add £311.00 or £11/month-1]
Intel Ethernet X540 10Gb BT DP + i350 1Gb BT DP Network Daughter Card [add £478.00 or £17/month-1]

I'm going to go for the x520 unless there are any known issues with it?

The X520 is a great card and I have not heard or read anyone having issues. I use the Brocade BR1020 DP 10GBe and direct connect the two T620s' directly to each other. It works great.

As to your setup:

4 x 300GB 15k SAS = 1.2TB raw storage? So if you RAID10; you'll end up with 600GB usable; which is more like 575GB actual. 100-120GB for OS partition and you've got just about 450GB of space for your StarWind images.

Or did you mean 1.2TB storage after you RAID10, so you have 4 x 600GB 15k drives?

You should definitely diagram (Visio or even hand-drawn) and post it on here so you can get some feedback from the StarWind fellas.

As to the system; you'll have a Windows volume that you then place your StarWind images onto. You then attach via iSCSI to the local images you've created as well as MPIO to the second node for redundancy. Then you can create your witness volume as well as your CSVs (if you're using CSVs) for Hyper-V cluster mode; and just follow through on the windows portion of your setups.

Just a word of warning; do NOT, I repeat; DO NOT reboot both StarWind nodes at the same time. Occassionally; and with no way to know; it will not allow you to attach to an image (iSCSI) if the system comes out of a hard-down situation and starts full synchronizations. This WILL break your windows cluster until StarWind has finished doing it's synchronizations.

Also, read up on LSFS if you plan on using it; as your calculations for storage space needed will be incorrect if you're assuming deduplication works as you are used to.

Async mode also has an extra bit of space required for journals; each environment has its own requirements based on amount of changed data per time frame against how much bandwidth available at the time. Also; async mode is automatically tuned to not use your full bandwidth with no way to change it right now. Even locally!

Thanks for the info.

We will be using 4 x 1.2TB HDD's so 2.4TB usable

You should definitely have a hot spare drive, and make sure to set your H710 (don't use the H310, you are correct there) for Predictive Failure Replace = True so that you don't ever rebuild your array with an actual bad drive; but only with a predicted bad drive.

1.2TB drive rebuilding can take a bit; and having an unrecoverable read error on the other drive will lead to a bad time. I know this is worse when on RAID5/6; but I believe the rule-of-thumb applies to RAID 10 (or any striped array).

I'm looking at buying these 10gb nics,

Would this cable work to connect the 2?

http://www.uk.insight.com/en-gb/product ... 0001824216

My plan was to use raid 10 on each server but could I use raid 5 since the virtual san will effectively be mirrored anyway?

with raid 10 i will have 8 1.2tb disks which makes it 2.4tb usable seems a bit of a waste!

We had the same issue with the client; RAID 10 just killed too much space. So we've done RAID6; but that is with 21 disks in an array; hot-spare and predictive failure replacement enabled.

As to the SFP+ cables; you'll have to check Intel data sheets about what cables they support. This is the only way to get support if something doesn't work.

hi gents!

You should definitely diagram (Visio or even hand-drawn) and post it on here so you can get some feedback from the StarWind fellas.

+1 to that!

The plan was to have raid 10.
So would I create a system partition install windows then leave the rest for starwind?

Running StarWind images and OS on separate volumes sounds much better imo

Just a word of warning; do NOT, I repeat; DO NOT reboot both StarWind nodes at the same time.

+100500 to that - that is not what HA was designed for.

We had the same issue with the client; RAID 10 just killed too much space.

What can I say? That is just the way the RAID things work. What I can suggest is to switch to 3 node config. I know, it sounds stupid, but lets take a look from another standpoint:
Why do you want to use RAID10? To increase the redundancy of the system. But if you will run 3 nodes there will be no need to have redundancy on the RAID level, thus you can use RAID 0, thus utilizing 6 disks instead of 8.

our 620's are around 4k a server. We can't afford to buy a third it would make more sense to buy a san!

I`m just giving you an options, not forcing to buy something more. You can live with 2 servers - not a problem at all.

actually does the third sever have to be the same hardware or just the same disk size?

if I have a raid10 setup on 2 nodes can I add more disks at a later date?

actually does the third sever have to be the same hardware or just the same disk size?

>>>You can use different hardware. That is why we have ALUA (Asymmetric Logical Unit Access) support

if I have a raid10 setup on 2 nodes can I add more disks at a later date?

Sure thing!

I think one of the only reasons someone says to buy identical hardware; is that you can leverage parts replacements. But as far as identical hardware: no.

If you want the same performance; sure. But you don't need performance off your async storage system; as it's just there for restoration purposes.

We don't require all *storage* hosts having same OS, same hardware or CPU family: not like Hyper-V or vmware hypervisors The only thing we really need is *similar* I/O performance for partners as when doing synchronous replication ACK for the caller would be delivered with the performance of the slower sibling in HA pair (I think it's so obvious it does not really require any further clarification).

thefinkster wrote:I think one of the only reasons someone says to buy identical hardware; is that you can leverage parts replacements. But as far as identical hardware: no.

If you want the same performance; sure. But you don't need performance off your async storage system; as it's just there for restoration purposes.