ONTAP Discussions

Use Single VIF for highly available hw_assist

RICHARDM82
3,228 Views

Hi,

Not sure if this is worthwhile, but here goes....

I have a 3240AE in HA. Cluster hearbeat is done over c0a and c0b, and hw_assist is configured to use e0M. All working perfectly fine.

I currently have eache0M plugged in to a Gigabit switch, and recently read a NetApp article that advised against putting the e0M 100MB port on a 1Gigabit switch. The cabinet currently has a Cisco 3548 10/100 switch for server iLOM cards, however it is only the 1, and would prefer to plug each e0M in to a different switch for switch redundancy, the same going for the hw_assist traffic. I dnt have the capacity to add 2x3548s.

I have readd that hw_assist can fail over the e0A if e0M is not configured. I don't currently plan on using e0a and e0b on either controller as I am using 10GB PCI cards/

Can I create a single mode VIF from e0a (dedicated) and e0b (stadnby) on each filer, plugging each each port in to a diffferent 1GB switch? With e0a being the dedicated adapter in the VIF, it means hw_assist traffic runs on the samw switch, and the e0b standby gives switch and port redundancy.

Is this overkill? Will it not work? Is it an unstable setup? I havent tried it, so not sure if it works, but wanted to see if it is a path worth following.

1 ACCEPTED SOLUTION

aborzenkov
3,228 Views

Plugging each e0M in separate switch does not make it more redundant. You still have single interface plugged in single switch. e0M on one head is completely independent from another.

It is possible to use any address (reachable from RLM/SP) for hw_assist, so yes, you can use vif for it. Which still leaves you with single RLM port plugged into single switch.

Do not overengineer. hw_assist is completely optional, it could be handy if it works, but if not - you still have the same behaviour as for years before. Everything will continue to work.

View solution in original post

4 REPLIES 4

aborzenkov
3,229 Views

Plugging each e0M in separate switch does not make it more redundant. You still have single interface plugged in single switch. e0M on one head is completely independent from another.

It is possible to use any address (reachable from RLM/SP) for hw_assist, so yes, you can use vif for it. Which still leaves you with single RLM port plugged into single switch.

Do not overengineer. hw_assist is completely optional, it could be handy if it works, but if not - you still have the same behaviour as for years before. Everything will continue to work.

RICHARDM82
3,228 Views

Thanks for the reply.

I get that I won't get any remote access redundancy from placing the e0M ports on separate switches, it is ensure that if a switch goes down, I can at least still access the other filer to check cf status and run a giveback/failover if necessary.

Was just looking at options - I won't be using the e0a/e0b ports on either filer, so just seeing where I can put them to use; and also whether I could use them to improve performance - notbaly during failover/giveback.

Does the hw_assist actually help with transferring of data during failover? Would it benefit at all by sitting on a 1GB switch compare to a 100MB switch?

aborzenkov
3,228 Views

hw_assist does not transfer any data, it is just for status check. E.g. if you switch off external power for one head, RLM has just enough residual capacity to sent notification about power failure so takeover starts immediately instead of waiting 15 seconds for missing partner. Or in case controller has fatal HW error that brings it down abruptly.

RICHARDM82
3,228 Views

That makes sense - thanks for the description. My current setup will suffice then!

Public