Starwind + Windows 2012 SOFS

[gfischer99]

Hi!

You recently had a press release indicating that Starwind could be use to build out a high performance, highly available Windows 2012 Scale-Out File Server (SOFS) on just two total servers.

Can you advise what your reference architecture for this solution would be?

Any guidance and detail would be appreciated.

Greg.

[Anatoly (staff)]

As the first step I`d recommend you to take a close look at our document that you can find by using the link below:
http://www.starwindsoftware.com/sw-conf ... le-servers

If you will have any additional question - don`t hesitate to ask us.

[gfischer99]

Anatoly ...

Thanks. This document certainly helped to understand the steps in configuration.

However, I am wondering what your suggested configuration of network adapters would be for the utmost in performance of a highly-available SOFS? We use only RDMA NICs with iSCSI Offload. We are thinking:

- 2 10GbE NIC ports per server to support SMB 3.0 and general network traffic
- 2 10GbE NIC ports per server for the internal iSCSI and Starwind sync traffic

Your thoughts/concerns?

Thanks in advance.

Greg.

[Anatoly (staff)]

Are you planning to have HA storage? Will it be 2 or 3 nodes if yes?

[gfischer99]

- The storage will be HA storage - local DAS (FusionIO Flash Storage) with HA provided by Starwind.
- We plan on two nodes at this point.

[Anatoly (staff)]

Well, the main concern is to have identical throughput on every datalink (except HeartBeat)- SyncChanel, iSCSI traffic, general network traffic, and everything else that will use the StarWind in the end.
So I`d recommended to have 1 10 Gig NIC for Sync, 1 10 Gig for iSCSI traffic and HeartBeat, and 1 to support SMB 3.0 and general network traffic

[gfischer99]

Anatoly:

Thanks for the replies ... A couple more general Starwind questions ...

In a 2 node HA cluster:
- does Starwind present the 2 nodes through a single virtual IP or does each node have its own iSCSI target IP?
- how are hosts connected to the HA cluster load balanced either through multiple NICs on each host and/or is a LUN/Volume load balanced for reads or writes accross nodes?
- is MPIO supported and how in a HA environment? do you have your own proprietary DSM or do you plan to have one?

Thanks in advance... our iSCSI experience to date is using Dell EqualLogic so some of the questions are driven by that background.

[anton (staff)]

1) Own. It's not shared IP solution, I/O is split using MPIO technology.

2) MPIO from client and own MPIO-like policy within StarWind cluster.

3) Yes, absolutely. MPIO is fully supported. Own DSM of course.

Anatoly:

Thanks for the replies ... A couple more general Starwind questions ...

In a 2 node HA cluster:
- does Starwind present the 2 nodes through a single virtual IP or does each node have its own iSCSI target IP?
- how are hosts connected to the HA cluster load balanced either through multiple NICs on each host and/or is a LUN/Volume load balanced for reads or writes accross nodes?
- is MPIO supported and how in a HA environment? do you have your own proprietary DSM or do you plan to have one?

Thanks in advance... our iSCSI experience to date is using Dell EqualLogic so some of the questions are driven by that background.

[gfischer99]

Anatoly:

If I am understanding you correctly, each node of an HA cluster presents itself with a separate IP address? And a HA LUN/Volume has I/O that is served up by both nodes of the cluster?

I've done a quick looke and don't see anything directly related. Do you have any documentation that explains how I/O is served from a HA cluter, how MPIO is configured on the client(s) using your provided Starwind iSCSI DSM?

Perhaps a very specific use case:
- we plan on deploying an HA Starwind iSCSI cluster with two nodes. The primary porpose will be to serve up high performance, low latency storage to a number of SQL servers.
- each node will have a couple FusionIO flash PCIe cards. Two 10GbE NIC ports per node for general iSCSI traffic + heartbeat, two 10Gbe ports for sync. This would be a total of 4 iSCSI ports in the cluster to support storage traffic to servers.
- all volumes/LUNs will be HA
- All client servers (ie. SQL servers) will have two 10GbE ports
- for redundancy, each server is connected to 2 active, stacked switches

How would you configure this scenario at the host and client end to maximize performance and redundnacy?

Thanks.

[Anatoly (staff)]

I`d recommend to configure the system as follows.

Drawing1.jpg (17.14 KiB) Viewed 15236 times

As you can see this config requires to have 1 additional NIC on each SANserver but it is required in order to maximize performance and redundancy. I`d suggest you to have one more NIC that`ll serve for non-iSCSI traffic on the client machine as well.

If I am understanding you correctly, each node of an HA cluster presents itself with a separate IP address?

Yes, exactly. And there can be more than 1 IP address.

Here is our document about the MPIO http://www.starwindsoftware.com/iscsi-s ... ltipathing
I hope it will answer your question.

And a HA LUN/Volume has I/O that is served up by both nodes of the cluster?

Yes, each node can process the I/O from client and it will be so if the Round Robin NLB policy is used. It is defaults for MS, and need to be configured for VMware (form the VMware side).