Can I change the Chassis/FEX discovery Policy ONLINE from "none" to "port Channel"

[Stefan Gruber]

Hello UCS frinds

My Question is

-Can I change the Chassis/FEX discovery Policy ONLINE from "none" to "port Channel"

After I set it to Port Channel i have to reacknowledge the Chassis

Can I Do this online?

Without intetrrupt?

[Wes Austin]

If you are running on later firmware releases, you can ack a single IOM at a time, which will only cause disruption on A/B side respectively.

In earlier releases, the entire chassis will need re-ack which requires downtime.

Regardless, we recommend making this change in a maintenance window for safety.

-Wes

[Stefan Gruber]

That means normaly I can

Switch from the Option "None" to "Port Channel" without interrupt?

After that i Have to reack IO Modul 1 and then if all is up again  IO Modul 2?

Thats it?

[Wes Austin · Accepted solution]

Yes, you can switch those without interruption. The port-channels will not be build until you re-ack the IOM. 

Yes, you need to re-ack each IOM individually. Please ensure that all faults have cleared and all VIF have re-pinned before acking the second IOM to minimize traffic disruption.

HTH,

Wes

[dmarkham_peak]

Just to be clear.  I am doing this for a UCS domain which has 4 chassis's this weekend.  Am I safe to acknowledge all IOMs on the B Fabric at the same time and once everything comes back up properly on the B Fabric I can acknowledge all IOMs on the A Fabric at the same time?

[helloworld2024]

I am migrating from 6248 to 6454 and have the same issue where I have 2 links from each IOM with the policy set to none. 

Cisco recommends switching to portchannel before migration. 

My questions are:

1) Since my current environment has 2 chassis with fully populated blades in production - would I experience an outage where the servers may reboot? 

- also do I acknowledge the IOMs or the chassis?

2) since I only have 2 links to the FI for each IOM - what's the recommended number of links to set if I go with the port-channel setting?

[Steven Tardy]

Acknowledging one IOM will not cause a complete outage (provided your installation has proper redundancy configured.)
But without reviewing your setup it is hard to say for sure. TAC has seen some poorly configured UCS through the years.
After acknowledging the one IOM, then the other IOM can be done (after waiting enough time).

Acknowledge the Chassis WILL cause a complete outage. So don't do that.

When you say "2 links to the FI for each IOM" do you actually mean that or saying the entire chassis has two (total) links (one to each IOM)?
It is perfectly valid to have a single link in a port-channel. More links provides more bandwidth, but not required.

[helloworld2024]

Thanks, Steven for the quick response:
I have recently done multiple FI failovers between A and B while replacing my Nexus 5k to 9K/MDS, I believe I am okay with redundancy.

2 links are per IOM so 4 links per chassis
chassis 1 IOM1 - 2 links to FIA
chassis 1 IOM2 - 2 links to FIB

in this case, if I acknowledge chassis 1 IOM 2 on the subordinate (FIB) side do I need to fail the FIA to FIB to acknowledge chassis 1 IOM1?
also below is my firmware version if that makes any difference:
Boot-loader Version : v3.6.0(05/09/2012)
Kernel Version : 5.0(3)N2(4.23c)
System Version : 5.0(3)N2(4.23c)
Service Pack Version : 4.2(2)SP0(Default)
Package Version : 4.2(3d)A
Startup Kernel Version : 5.0(3)N2(4.23c)
Activate Status : Ready
Startup System Version : 5.0(3)N2(4.23c)
Activate Status : Ready
Startup Service Pack Version : 4.2(2)SP0(Default)
Activate Status : Ready

[Steven Tardy]

UCS primary/subordinate is only a management plain (which FI, of the pair, is in control of the management cluster) and has almost nothing to do with data plain traffic or who's on first.
There is no need to do FI fail over or change FI lead or primary/subordinate FI or anything like that.

Acknowledging IOM 2 or IOM 1 first doesn't matter.

After modifying the port-channel setting and acknowledge an IOM, then that IOM<>FI link will be a port-channel.

Give that IOM plenty of time to boot and all of the connections to come up and the OS/VMs to start using those connections before doing the same for the "other" IOM.
(No need to rush. Take a 30 minute break to do something else and then come back to do the "other" IOM.)

[helloworld2024]

I did it over the weekend and it worked perfectly - Thank you!