Best solution for hosting CSV on iSCSI with HA

Radian

Member
Mar 1, 2011
59
5
8
I'm running a HyperV cluster 2012R2 with a Windows Storage 2012 NAS provisioning iSCSI as the CSV, but for all the availability that the cluster gives me, the amount of times I have to reboot the NAS for windows updates means the cluster is down just as often as a single HyperV host, nothing is gained.

What is the best solution for making a CSV highly available? The NAS contains 24TB in RAID6 SATA6G config using RE4 drives.

Any thoughts, does it have to be dual controller type SAN to achieve?
 

TuxDude

Well-Known Member
Sep 17, 2011
618
338
63
My first thought is simply get rid of windows on the NAS - updating the NAS OS less often will mean better uptime. A locked-down hardered linux/bsd/solaris shouldn't be getting critical kernel-security updates that often, and any other updates (SSH is the only other service I'd even leave open/enabled) can be applied online without a reboot.

If you want real HA from your storage, you are going to need either some way to have dual controllers accessing the same set of spindles (eg. a dual-controller SAN), or a second NAS with another 24TB of capacity and synchronous replication between them. The second option is obviously more expensive with all that extra disk space, but it could also be located farther away to protect you against eg. a fire. Though like RAID, real-time replication is NOT a backup - accidentally deleting something important will wipe it out in both places.
 

mrkrad

Well-Known Member
Oct 13, 2012
1,244
52
48
storevirtual VSA running on a pair of hyper-v nodes would do the trick. at 50% storage cost for HA
 

Radian

Member
Mar 1, 2011
59
5
8
I prefer the dual controller idea but can't find which OS to run and controllers seem pricey.

Looking at the VSA idea and it seems like what I need, however, I'm unsure what it would look like with my cluster. In my 2 cluster host's I would put a RAID 1 array as DAS and then run the VSA VM, would there be two VMs, one on each Cluster host?

Once the shared storage is configured, how do I present it to the cluster, just as a regular CSV, iSCSI attached?

I've got the demo so I guess I'll find out, but what is the best config for performance?
 

Chuntzu

Active Member
Jun 30, 2013
383
98
28
Server 2012r2 has cluster aware updating. So when updated are done it would occur on one node then the second with the cluster not going down. It worked well on my 3 node sofs cluster.
 

cesmith9999

Well-Known Member
Mar 26, 2013
1,364
461
83
You have a Hyper-V cluster And a single node Storage server running as a NAS?

Who is your NAS vendor?

you can do a few things
1) add HBA's to both HyperV nodes and use spaces to create a shared pool of disks. then your CSV's will be HA.
2) create a SOFS cluster to host the VHDX over SMB or ISCSI
 

Radian

Member
Mar 1, 2011
59
5
8
You have a Hyper-V cluster And a single node Storage server running as a NAS?

Who is your NAS vendor?

you can do a few things
1) add HBA's to both HyperV nodes and use spaces to create a shared pool of disks. then your CSV's will be HA.
2) create a SOFS cluster to host the VHDX over SMB or ISCSI

The NAS is homebrew so no particular vendor (Avoton C2750 with LSI 9260, SAS extender to 24TB), servicing iscsi to the HV hosts, its the NAS updating that is causing me a headache, as all VM's need to be down (CSV goes away during the reboot).

So creating a SOFS can be done on the same cluster as the HV roles?

Just installed the HP VSA and it looks like it should do the trick, but it only supports single NIC in HyperV, how does that work when I have MPIO enabled from my cluster to iscsi targets, what happens to the redundant iscsi link?
 

Chuntzu

Active Member
Jun 30, 2013
383
98
28
@tjk I use SAS drives with SAS cabling as the interconnect. Very simple and passed cluster validation wizard very easily.
 

TuxDude

Well-Known Member
Sep 17, 2011
618
338
63
Just installed the HP VSA and it looks like it should do the trick, but it only supports single NIC in HyperV, how does that work when I have MPIO enabled from my cluster to iscsi targets, what happens to the redundant iscsi link?
You still get MPIO with the HP VSA. On the target (VSA) side there is only the single NIC, but you get redundancy from the hypervisor assuming it is connected to a virtual switch with multiple physical NICs behind it. On the initiator side (for VMware at least) you have either two iSCSI CNAs or one software-iSCSI-adapter with 2 vmknic's bound to it - in both cases you should end up seeing two paths to storage. My only experience with HyperV is with FC storage, so I'm not quite sure how that part would look here.
 

cesmith9999

Well-Known Member
Mar 26, 2013
1,364
461
83
With HyperV nodes, if they have shared JBOD storage, you can use spaces to create a shared disk scenario. and it is faster than NAS. you just have to remain in mirror mode.
 

Radian

Member
Mar 1, 2011
59
5
8
With HyperV nodes, if they have shared JBOD storage, you can use spaces to create a shared disk scenario. and it is faster than NAS. you just have to remain in mirror mode.
It seems like a simple scenario, but if we did it that way, wouldn't I need 48TB of storage to provision 24TB to clients?
 

cesmith9999

Well-Known Member
Mar 26, 2013
1,364
461
83
Unfortunately, yes. Spaces Parity performance is not really good enough for production for anything other than home use (where you are reading data) and archive.

I also recommend that you add SSD's into the mix to create a Tiered Spaces so that you can gain more performance.

The only issue in this scenario is that you cannot use deduplication on that shared local CSV volume. so your boot volumes cannot dedup against each other.

if you had a different cluster (SOFS - use SMB3 for your VHD's) you could turn on dedup in that scenario. however. the dedup seems to be limited to files no larger than 1 TB in size.

Chris
 
Last edited:

cesmith9999

Well-Known Member
Mar 26, 2013
1,364
461
83
Cluster Shared Volume

It is a volume type that is a modification of NTFS. First used in 2008 R2 (for Hyper-V) then in Server 2012 Upgraded for use in Scale Out File Server (continually available Cluster).

Chris
 

Radian

Member
Mar 1, 2011
59
5
8
What a waste of time that was, VSA itself was quite painless to install but it doesn't support multiple NICs (on HyperV) and requires a third machine to act as failover manager. If I reboot the failover manager node, the CSV is lost?!?! I've just rebooted the other node in the cluster and even that makes the provisioned network RAID volume fail....it just seems overly complicated and another thing to go wrong when trying to achieve high availability.....think I'll look at using a Synology box to provision hardware based iSCSI. This VSA bull is a accident waiting to happen.
 

gea

Well-Known Member
Dec 31, 2010
2,875
1,017
113
DE

Radian

Member
Mar 1, 2011
59
5
8
Does it support HA? I guess what I am looking for is a 4 node, 2 cluster system, one cluster to host the CSV (as a redundant role) then a second cluster to host the VM's stored on the CSV....it begs the question, what is the point of CSV in a cluster if the CSV isn't redundant?

I'm gonna try out storage spaces with two RAID1 arrays, one in each cluster node....this seems a similar approach to VSA, hopefully it works better in practice though!
 

Mike

Member
May 29, 2012
482
16
18
EU
What a waste of time that was, VSA itself was quite painless to install but it doesn't support multiple NICs (on HyperV) and requires a third machine to act as failover manager. If I reboot the failover manager node, the CSV is lost?!?! I've just rebooted the other node in the cluster and even that makes the provisioned network RAID volume fail....it just seems overly complicated and another thing to go wrong when trying to achieve high availability.....think I'll look at using a Synology box to provision hardware based iSCSI. This VSA bull is a accident waiting to happen.
To be fair, HA/clustered storage is complicated. Non-realtime replications are not HA.
In a 2 node setup DRBD may actually have fairly forgiving configuration options.
 

gea

Well-Known Member
Dec 31, 2010
2,875
1,017
113
DE
HA is not a software nor a hardware, its an effort to increase availability.
This can be done on a OS, blockstorage or service level, depending on your needs, knowledge or money.

With ZFS and Solaris/ OmniOS this starts with
- a minimalistic OS that offers a great uptime (no be-weekly security updates)
- always consistent filesystems with snaps, checksums and auto-repair capability during reads

- replication of filesystems down to a minute difference, on problems you can mount the backup
- unlimited snapshots of former states

These are trivial methods

Next is
- Multipath SAS storage with two nodes accessing the same storage
- SAN Block mirroring (create two SAN boxes with iSCSI/FC/IB targets) and Raid-1 them on a storage head or at Windows level

Next is
Active/Active HA like its done with Free or commercial supported products like RSF-1
RSF-1 ZFS HA High Availability Open Storage Cluster
 
Last edited: