How does SMB 3 Multipath actualy work?

Question

First of all, sorry if there is a better form to post this in, I didn't see a general networking or SMB forum.
I've been trying to find anything about how SMB 3 Multipath actually works at a networking level (OSI Layer 2, Layer 3). I can't find anything that actually talks about how it achieves multipathing once it's out of its exit interface. Even how it actually chooses
potential exit interfaces is still a mystery to me.
How does the MAC address forwarding work? Does it somehow dynamically learn the available MACs on the destination and split the load between them? If that's the case does it work over layer 3 subnets if there are two available routes? How does it learn about
multiple available paths and actually get the network equipment to use them? In other words if Server A and B have 4 NICs they have 4 MAC addresses or more, one or more IP addresses, teamed or not teamed. Layer 2 adjacency vs Layer 3. There are a lot of variables
that it has to work with. Is DNS involved at all? If my connection attempt is the server.domain.local DNS might have more than one IP address that server is reachable from but if I connect to 192.168.1.1 then I'm specifying an IP and DNS doesn't come into
play, does that affect SMB 3 Multipath?
I know SMB 3 works when in a team and not in a team, I've tried it. But if it is in an LACP team then once it leaves the server NIC it's up to the switch to forward that to the destination based on the hash, for a single flow in normal situations the switch
would pick the same interface but I get increased bandwidth so I know the switch is sending to both interfaces, so I assume there are multiple flows. That means something about the flows has to be different, I assume the destination MACs, but I haven't put
a packet capture to look...yet...
Does anyone know of any papers or details that talk about these issues they can point me to? Every blog post or video I've seen is super generic about how it actually works. It's not magic.

Answer 1

Hi,
With Multichannel, SMB will create multiple TCP/IP connections for that single session (at least one per interface or more if they are RSS-capable). This allows SMB to use
the combined NIC bandwidth available and makes it possible for the SMB client to continue to work uninterrupted if a NIC fails.
Although a team always provides fault tolerance, SMB without Multichannel will create only one TCP/IP connection per team, leading to limitations in both the number of CPU
cores engaged and the use of the full team bandwidth.
The related article:
The basics of SMB Multichannel, a feature of Windows Server 2012 and SMB 3.0
http://blogs.technet.com/b/josebda/archive/2012/06/28/the-basics-of-smb-multichannel-a-feature-of-windows-server-2012-and-smb-3-0.aspx
More information:
How much traffic needs to pass between the SMB Client and Server before Multichannel actually starts?
http://blogs.technet.com/b/josebda/archive/2013/01/18/how-much-traffic-needs-to-pass-between-the-smb-client-and-server-before-multichannel-actually-starts.aspx
Hope this helps.
We
are trying to better understand customer views on social support experience, so your participation in this
interview project would be greatly appreciated if you have time.
Thanks for helping make community forums a great place.