We are configuring our MDS switches for new storage and there are a couple of options. I am not sure which is the best way to go.
We have 6 x 9124's. Our new storage has plenty of ports. Each of the 6 x 9124's will be connected to 2 ports on the storage (12 connections in total). I am going to treat the 6 x 9124s as pairs - so really it will be 3 pairs, and each pair is essentially a fabric A and B.
So the idea would be that when connecting a host to our SAN, we will pick one of the pairs and connect the host to both switches in that pair and zone it to use the storage ports connected to that same switch. This way, traffic will never actually need to "cross" between switches (and in fact it can't as there are no ISLs) - it will always be from the host to the storage all on the same switch. Each of the pairs of switches is acting as a fabric A and B.
In reality we will have 6 fabrics (as each switch will be a fabric). Connecting a new host to our storage will require configuration on two switches (fabrics), depending on which of the 3 pairs we choose to connect that host to. So really it will be 3 seperate "fabric A and B" setups, one for each pair.
This all sounds reasonable to me and should work quite well.
The only thing I am unsure about is whether there is any benefit in connecting ISLs between each of the "fabric A" switches and each of the "fabric B" switches - which would actually mean we really did have 2 fabrics rather than 6.
I am looking to understand the pros and cons of doing this. We have plenty of ports so the ports lost to the ISLs are not an issue. These ISLs would never be used for "storage traffic" to actually flow between switches as we would still zone it so that hosts always talked to the storage port(s) on the same switch. The only reason to create the ISLs would be to consolidate 6 fabrics down to 2 which may (?) make it easier to manage.
Are there real advantages to connecting the switches to consolidate down to 2 fabrics? I'm thinking that potentially it will be "nicer" to connect to DCNM-SAN and only see 2 fabrics rather than 6. I'm also thinking that the "maps" that DCNM produces will be easier to comprehend when there are only 2 - rather than having to look at 6 different maps to get a good picture of what we have. We will only be using the free DCNM-SAN (Essentials I think it is called).
Hopefully my question makes sense. I'm really looking to understand whether it is better to keep 6 seperate fabrics or whether it might make more sense to use ISLs to consolidate down to 2. Whether we keep 6 or go with 2, the concept would remain the same (each host connecting to a single pair of A/B switches and zones to the "local" storage ports) - so both solutions keep all storage traffic within the same switch. The ISLs would effectively only act as "management" connectors such that it could all be seen together within DCNM-SAN. Is there a disadvantage to connecting them together?
How would you do it?
Although I am not a SAN guy I would connect them together from the start. If the need ever arises to use a target on a different switch you would not face a SAN merge. Furthermore the zoning proces will be simpler you simply need to zone on FABRIC A and B and don't need to determine is the fabric A1 A2 or A3 to which the host is connected.