1000v on UCS

[jbain44]

What's been the experience with running the 1000v VSM on a UCS blade where the Blade is also running a VEM? Thought I had a solid configuration with HA failover but something happened and it all blew up... and all 6 of my hosts disappeared from the VSM. I see in the Cisco Doc it is a supported config but I'm concerned in the event of an outage to both the Primary and Standby VSM or other changes within the UCS domain that the configuration may not be stable. I'm looking for any input from Cisco on weather or not this is a valid config to go into Production with or if it's "recommended" that we run the VSM's on hosts outside the UCS domain? Please advise.

Thanks,

Jason

[Robert Burns]

Jason,

We have quire a large install base of UCS with 1000v currently and most customers are running their VSMs on the VEMs which are DVS.members.

If you're using both UCS adapters for VEM uplinks (in the same Uplink Port Profile), ensure you're NOT using "Enable Failover" on the vNICs in on your VEM's UCS Service Profiles.  Reason being is for failover to occur the 1000v must detect a failure.  With the enable failover option the link never goes down.  As best practice we advise people to disable this in this situation.

Something must have triggered the failure.  Was it an HA event or were you testing anything?  Did you check the VEM-Support logs, VSM logs?  Any detail will help us determine the cause and advise on steps to avoid it in the future.

Regards,

Robert

[jbain44]

Robert,

Yeah there was a bunch of process crashes in the log but the buffer wasn't long enough and I forgot to capture the information before rebooting the VSM's to see if the envrionment would come back. The only thing I'm aware of that happened was that we enabled NTP on all 6 of our ESXi hosts. I'm wondering if the time stamp had something to do with this. Do the VSM and VEM's use timestamps in their communication process? I'm also now getting errors in the logs now that the VEM licenses can't be obtained. We're still operating in eval mode but it hasn't been 60 days so I'm really thinking enabling NTP was the clulprit here. Have you seen this before? I have tac case 613805501 open btw.

Thanks,

Jason

[Lukas Mazur]

hi Jason, Robert,

i observed a similar behavior on my ESX servers. The Problem was that immediatelly after i configured the NTP and set the correct time the eval license expired. (before configuring NTP the system time on the ESX was 2001-xx something also on UCS Blades). Therefore might it be in your case to that the eval license expired and the Nexus1000v (Enterprise Plus) wasnt "licensed" anymore and generated your problem?

its just a hint ...

BR

Lukas

[jbain44]

Thanks Lucas. Yeah this is somethin Cisco needs to include in the Getting Started Guide. Enable NTP before you install the VEM's. Assuming this would not be an issue with fully licensed VEM's. But for now I guess I'm left with reinstalling the VEM's to get everything back to normal....

[Robert Burns]

Jason/Luka,

We can add a reminder to our Getting Started Guide about this, but its expected that when you get to the point of installing 1000v, your vSphere setup should be complete - this includes setting accurate system and/or NTP time.  These requirements should be part of setting up your VMware hosts and will be included in the VMware Quick Start Guides also.

I'll take the feedback to our Docs team.

Regards,

Robert