Standard Cabling SPOF...

[Sebastian_Goetze]

Hi all,

I'm at a customer site and the standard cabling according to A 210-05224+A0 was used to cable the system, when it was set up.

I see a SPOF there:

It's a 3220A-Pair. 4 SAS shelves in 2 stacks.

Stack A is connected to 0a (both controllers, one in the A-'loop', the partner in the B-'loop'

and to 3b (extra-Quad-SAS card, last shelf, one in B-'loop', partner in A-'loop')

Stack B is connected to 3a and 3d (same principle as before)

==> If there's a problem with the card in slot 3 of controller 2, his B-stack completely looses connectivity, the partner (A) has to take over.

Why not connect the B-Stack with 0b ?

Then there would not be a SPOF...

I do not see any advantages following the cabling in this cabling guide (regarding this detail).

Does anyone see a reason for this?

Can we just re-cable 3a -> 0b ?

TIA

Sebastian#

[aborzenkov]

The main reason is to ensure you are using ports of different ASICs. Of course you may change cabling now, but should you ever connect third stack, you will face the same issue then.

That's why you have partner controller after all.

[Sebastian_Goetze]

Well, that's my point...

Controller A's stack is connected with 0a and 3b,

Controller B's stack is connected with 3a and 3d.

There won't be a third stack connected any time soon, but whenever it will be, it'll be 'just as bad' as the current B-Stack (3a and 3c), so no advantage there.

So why not connect like this:

Controller B's stack is connected with 0b and 3d.

That way it would be more symmetric and less error prone, since you eliminate the SPOF of the Card in slot 3 (controller B, specifically).

Don't you think?

I can't think of a reason to connect both Controller B stack connectors to the same card/slot, if you have an alternative (0b)...