Need Help - Looking to Build our Own SAN

[FusionTech]

Hi all,

I saw a few old threads addressing the same issues but basically, we've looked at a few commercial solutions such as the Powervault series and some other brands but think at the moment we're better off making our own.

We spec'ed out a pair of MD3000i (3GBPS SAS ports + 4 x NIC ) and a pair of MD3220i (6GBPS SAS ports + 2 x NIC)

We're planning on running a cloud deployment for VMS over 20-40 servers (dual-xeon 2.5GHZ 12MB Cache - 16GB RAM) - nodes and need a SAN solution.

Trying to understand

a) What is more important? an HBA, a good RAID card with ample amounts of cache, a set of GIGABIT / 10GB NIC with TOE or just more CPU power

b) For drives, to maximize throughput and reliability, I'm starting to think more 10k rpm drives is better than bigger 15k rpm drives or even 7.2k RPM drives but at what point do you find the equilibrium between # of drives and spin speed

We're really looking for a solution that will be a balance between ample IOPS (which is more dictated by the drives both # and type) as well as ample throughput for read/write given the potential large number of VMs.

For the backup system, we will likely go with an identical system but a lesser system may be sufficient as well? I suspect it can be less intensive as all it is doing is snapshotting the volumes on the main system but not sure how HA works in these instances.

c) Want to understand how Starwind is better than OpenFiler, SANSymphony, Open-E

d) Does Starwind now do de-dupe, snap shotting, replication etc?

We want to create a solution whereby if the storage on the server fails say anywhere from a drive to the entire server, the HA would kick in and it would go to the backup and activate snapshots of the VMs.

e) We would be using Cloudstack to manage the cluster - anyone using Starwind with Cloudstack? I know OnApp does a bunch with Starwind

[Anatoly (staff)]

OK, lets go through this one-by-one.

a) What is more important? an HBA, a good RAID card with ample amounts of cache, a set of GIGABIT / 10GB NIC with TOE or just more CPU power

Everything is important - your SAN should be in the golden middle of ballance, otherwise you will get the bottleneck.

b) For drives, to maximize throughput and reliability, I'm starting to think more 10k rpm drives is better than bigger 15k rpm drives or even 7.2k RPM drives but at what point do you find the equilibrium between # of drives and spin speed

We're really looking for a solution that will be a balance between ample IOPS (which is more dictated by the drives both # and type) as well as ample throughput for read/write given the potential large number of VMs.

For the backup system, we will likely go with an identical system but a lesser system may be sufficient as well? I suspect it can be less intensive as all it is doing is snapshotting the volumes on the main system but not sure how HA works in these instances.

Wait a second, have I understoud you correctly that your statement is that HDDs with lower rpm value is better then high-speed? If yes then why?

c) Want to understand how Starwind is better than OpenFiler, SANSymphony, Open-E

Much better, in my hummble opinin I`ll try to explain: it is really flexible woth wide range of enterprise features that are all included into one product - thin provisioning, snapshots, deduplication, SPTI and of cource HA (true active-active, as you may know). Also it doesnt require WS2008R2 only for its functioning - you can run it on Server Core, and since its Windows based the management of the software id really easy, so you can use intuition instead of bunch of KB articles and Great Book of Manuals to configure something with it.

d) Does Starwind now do de-dupe, snap shotting, replication etc?

We want to create a solution whereby if the storage on the server fails say anywhere from a drive to the entire server, the HA would kick in and it would go to the backup and activate snapshots of the VMs.

As I mentioned in my previous answer - yeap:) Unfortunatelly in current release it is not possible to use everything bundled, but it will be in next major release build that is comming really soon, so stay tuned

e) We would be using Cloudstack to manage the cluster - anyone using Starwind with Cloudstack? I know OnApp does a bunch with Starwind

To be honest on my experiense I haven`t saw any of such scenarios with Cloudstack, but I`m pretty sure that it shold work fine anything that works with iSCSI and have initiator can use our SW.

[mooseracing]

c) Want to understand how Starwind is better than OpenFiler, SANSymphony, Open-E

When comparing to the Opensource offerings I kept coming back to Starwind because I could not get everything configured and get the expected results as fast as I could with Starwind. I even used Starwind's Free version and Hyper V server, for a "free" setup.

I am also alot more comfortable fixing and using Starwind issues. Have you looked at the support forums on Openfiler?

[Anatoly (staff)]

Oh! We are so on the same page on it! Can you remind me how much their manual cost? $60?

[iceknight]

I wonder if the OP has any feedback with his Cloudstack/StarWind deployment

Anatoly, I am looking into a slightly different deployment scenario, perhaps you can share some insight if you have encountered it or thoughts on its feasability:

1. Install SW on Esxi in a hyper-converged mode with SW installed as a vm and using local storage across hosts to provision shared storage for all the vms on the host.
2. Deploy Cloudstack on top of Esxi/SW using SW provisioned storage as the primary storage for Cloudstack thus removing the need to deploy SW on multiple redundant hardware boxes in as a traditional SAN.
3. Result is better performing primary storage (no network latency or bottlenecks) at lower cost (no hardware)

Any thoughts?

[anton (staff)]

Hold on for a second... If you get rid of a network here how are you going to implement redundancy / replication between hosts?

I wonder if the OP has any feedback with his Cloudstack/StarWind deployment

Anatoly, I am looking into a slightly different deployment scenario, perhaps you can share some insight if you have encountered it or thoughts on its feasability:

1. Install SW on Esxi in a hyper-converged mode with SW installed as a vm and using local storage across hosts to provision shared storage for all the vms on the host.
2. Deploy Cloudstack on top of Esxi/SW using SW provisioned storage as the primary storage for Cloudstack thus removing the need to deploy SW on multiple redundant hardware boxes in as a traditional SAN.
3. Result is better performing primary storage (no network latency or bottlenecks) at lower cost (no hardware)

Any thoughts?

[iceknight]

You don't get rid of the network, you just get rid of a physical SAN. The hosts will still be interconnected to a (multiple redundant) switch as well direct NIC to NIC links between hosts for SAN replication. Cloudstack should be able to access the hosts normally through their IP's.

[Oles (staff)]

iceknight, you are correct.

Please let us know if anyone has any questions left, thank you!