[OpenIndiana-discuss] HBA failover

[Saso Kiselkov]

On 17/06/2013 15:12, Sebastian Gabler wrote:
> Hi,
> 
> it occured to me that obviously some ZFS Storage systems only feature a
> single SAS HBA, including the ZFSSA 7320. At least, as far as I understand.
> From what I saw in the 7320 documentation, each of the two HBA ports is
> connected to each of the two ports of a shelf, which should protect from
> the failure of a single cable, SAS Expander or drive controller (using
> dual ported drives). My understanding is that this topology is drawing
> upon MultiPath. But that still should make the HBA a SPOF, shouldn't it?
> As HBAs are a quite cheap commodity, I wonder why I even don't see a
> secondary HBA as an option for SUN ZFSSAs, or other appliances like the
> Netgear ReadyNas.
> 
> Thanks for any hints on how to exclude the HBA as a SPOF without using a
> cluster.

The reason probably is that if your HBA goes down, your OS is likely to
go with it. Failure of a HBA would entail a creeping (or possibly a
sudden) death of the device from the look of the device driver and I'm
not convinced drivers such as mpt_sas handle this situation in any way.

If you worry about core component failure in a storage head, I'd
recommend going with a clustered head node setup. It's much easier to
design a system where one node shoots the other node in the head if it
behaves erratically, rather than having the self-same node trying to
recover gracefully from core HW failures. Dealing with cable failures or
external device failures is fairly easy in comparison (the interfaces
are designed to tolerate it and handle it).

These are just my guesses though, so take them with a pinch of sail.

Cheers,
-- 
Saso