Two LSI 9300-8i for Redundancy?

[Steven Sedory]

We're spinning up a server with the following:

Dell R730xd, 16x 3.5 Drive Bay model - 12x front backplane, 4x mid backplane
14x 4TB NLSAS
256GB RAM
2x 10Core 2620 v3
FreeNAS-9.10.1-U4

Right now, we have a LSI 9300-8i plugged in, both SAS ports plugging into the front backplane. Everything works fine. When I plug in our second HBA both HBA's see all the drives, but once in FreeNAS, there's no sign of multipathing under Volumes as explained in the docu.

I soon realized I may have jumped to the conclusion that this was a safe setup. I called LSI/Avago and they said their HBA's are not meant for multi path configuration and are not supported for it.

That being said, would this setup still be safe for fault tolerance? I can't seem to find a clear answer.

Thanks in advance.

 

[Dice]

I can't recall anyone building a DIY FreeNAS with this required level of uptime. That is to say, this is rather unusual territory.
Have you contacted iX systems for a TrueNAS quote? That might be a good idea.

If really requiring this extreme reliability - you'd probably have a lot less headache having a second machine sitting ready with replicated data.
It might not need to be as fast - if it's purpose is only to keep business afloat for a few hours of downtime while servicing the main machine.

 

[Steven Sedory]

We've built a handful of servers like this that are in production, just with single HBA's. I'm hoping someone can give me a thumbs up/down regarding the dual HBA config.

 

[Dice]

Only mutlipathing I've seen is using multiple cables to the same expander backplane, ie from the same HBA.


Ping @Ericloewe @jgreco

 

[Ericloewe]

SAS theoretically allows this. In practice, the more you deviate from a basic setup, the more resistance you can expect.

Fortunately, this is serious overkill. HBAs don't just die.

 

[Dice]

Fortunately, this is serious overkill. HBAs don't just die.

Unless Linus from LTT happen to fly by in an Boeing 747 at 10.000m height. His karma may or may not influence HBA performance (or at least - kill off half a motherboard worth of PCIe lanes )

 

[Steven Sedory]

SAS theoretically allows this. In practice, the more you deviate from a basic setup, the more resistance you can expect.

Fortunately, this is serious overkill. HBAs don't just die.

This is the warm and fuzzy I was looking for.

Thanks all.

 

[Nicholas M]

We use two LSI9300's connected to a supermicro JBOD. Works just fine, you can even (manually) enable active-active multipath - which speeds up scrubs when you have lots of disks... I think active-active multipath is even on the roadmap for FN10. If you have any queries, PM me :)