Q: Why would you want to make a Windows Server 2012 Scale-Out File Server cluster access a SAN via the file servers using SMB3.0?

A: Windows Server 2012 introduces many advancements to both the SMB 3.0 protocol and the file server. Capabilities such as Active-Active cluster let multiple hosts in a cluster share the same SMB share whose content is on a Cluster Shared Volume. Transparent Failover lets SMB clients move between SMB servers with no loss of handles or downtime.

One architecture used is to attach a cluster of Windows Server 2012 file servers to a SAN, for example by using fibre channel, then making other servers in the data center communicate with the SAN via the file server cluster using SMB instead of by direct communication to the SAN. This architecture is shown below.

fileservclussmb3sml

The question is, why would this architecture be desirable?

The figure above might actually give the answer for many. Consider the hundreds of, in this example, Hyper-V Servers (or they could be other types of servers), that all have virtual machines (VMs) stored on the SAN. If they directly accessed the cluster, then every server would need two connections to the SAN via a very expensive fibre channel HBA (needing two for redundancy from failure), plus multiple large fibre channel switches with hundreds of ports.

Additionally because a Failover Cluster can have only 64 nodes, it would be necessary to break up the LUNs based on the cluster access. This is a very complex and costly solution.

By placing the Windows 2012 Scale-Out File Server cluster in front of the SAN, only the four file servers in the cluster require the fibre channel HBAs, and a much smaller fibre channel switch is required. All the servers connect via SMB 3.0, which could be using the existing network infrastructure depending on requirements and the capability of the current network. With SMB 3.0 on a 10Gbps network, there should be no performance degradation from a server directly accessing the fibre channel array on a standard 4Gb connection.