I'm not seeing a ton of benefit to using SMB 3/SOFS today...
Let's use this as an example:
Hypervisor cluster node (physical box)/running Windows Server 2012)
|
SMB 3 networking (this could be over 1Gb/10Gb ethernet, or one of the 3 flavors of RDMA over SMB)
|
Scale out file server cluster node (physical box/running Windows Server 2012)
|
iSCSI networking (this could be over 1Gb/10Gb ethernet)
|
Shared storage (iscsi SAN)
It's obvious the data path is shortest w/ iscsi presented directly to the hypervisor hosts rather than adding an existing level right? Let's review the new SMB 3 benefits:
So we get SMB multichannel: which is similar to iSCSI MPIO. Net gain is ?
So we get SMB Direct: which is faster and cheaper than 1/Gb/10Gb ethernet. Ok I see this but it does require a new spend if you only have 1Gb Ethernet NICs/switches.
So we get SMB Transparent failover: which is similar to a multi-controller storage array. Net gain is ? I already have a SAN with dual storage controllers (as do most people with a SAN)
So we get SMB Scale out. This has some real net benefit it appears.
Of course, so my link from my sofs cluster nodes will have to be equal to or greater than the link coming into my shared storage (aka isci SAN) right? If we do the older Infiniband QDR (32 Gbps) to the sofs nodes using SMB Direct, the shared storage will become a bottleneck even running over 10GbE (iSCSI MPIO) right?
I'm aware that I can use Clusterd Spaces in liue of an iscsi SAN for the shared storage piece of sofs but the hardware is AWOL with the exception of one or two vendors and there isn't a ton of performance/reliability metrics on this setup available yet.
I guess the new features don't really help with those of us who already have existing investments in DAS (Dell MD1200) or SANs (NetApp, EMC).
Am I completely wrong/missing something? Let me hear your comments.
Let's use this as an example:
Hypervisor cluster node (physical box)/running Windows Server 2012)
|
SMB 3 networking (this could be over 1Gb/10Gb ethernet, or one of the 3 flavors of RDMA over SMB)
|
Scale out file server cluster node (physical box/running Windows Server 2012)
|
iSCSI networking (this could be over 1Gb/10Gb ethernet)
|
Shared storage (iscsi SAN)
It's obvious the data path is shortest w/ iscsi presented directly to the hypervisor hosts rather than adding an existing level right? Let's review the new SMB 3 benefits:
So we get SMB multichannel: which is similar to iSCSI MPIO. Net gain is ?
So we get SMB Direct: which is faster and cheaper than 1/Gb/10Gb ethernet. Ok I see this but it does require a new spend if you only have 1Gb Ethernet NICs/switches.
So we get SMB Transparent failover: which is similar to a multi-controller storage array. Net gain is ? I already have a SAN with dual storage controllers (as do most people with a SAN)
So we get SMB Scale out. This has some real net benefit it appears.
Of course, so my link from my sofs cluster nodes will have to be equal to or greater than the link coming into my shared storage (aka isci SAN) right? If we do the older Infiniband QDR (32 Gbps) to the sofs nodes using SMB Direct, the shared storage will become a bottleneck even running over 10GbE (iSCSI MPIO) right?
I'm aware that I can use Clusterd Spaces in liue of an iscsi SAN for the shared storage piece of sofs but the hardware is AWOL with the exception of one or two vendors and there isn't a ton of performance/reliability metrics on this setup available yet.
I guess the new features don't really help with those of us who already have existing investments in DAS (Dell MD1200) or SANs (NetApp, EMC).
Am I completely wrong/missing something? Let me hear your comments.