Cisco UCS Blade Failover- State Less Computing

[Gonuguntla Gopi]

Hi ,

One of my customer is looking for blade failover so that the service profile & OS will be moved onto the blade seeamlessly without any manual intervention.

I tried my best & Created the server pool of four blades &  service profile  for the same. Out of four blades service profile is assigned to the three blades. Rest one is spare...Looks good

Now the testing started....As per my knowledge i showed the failover.

I decommissioned the server on the UCS Manager & the service profile started moving on to the other blade .....It looks good.

But customer is not happy with this scenario.....

Customer removed one of the blade from the pool & Now the service profile didn't moved onto another spare blade inthe pool.

Is this possible ?? I never heard about this.....up on any cisco events  or docs....etc.

Please advise.....Looking forward to hear ASAP.

Regards,

Gopi G 

+91-9818122788

[Walter Dey]

Hi Gopi

Officially, this is not supported.

There are in fact cases that work, as you mention.

The critical point however is: when do you consider a blade dead, and therefore starting failover. If this is a positive false signal, you are in trouble.

Don't forget, that UCS philosophy doesn't rely on OS specific plugin's.

The failover is clearly disruptive !

Walter.

[Gonuguntla Gopi]

Hi Walter,

Thanks for the your time...& info...

What is the best recommended possible way ?

Even with the possible failover scenario i explained ...I don't think so that is also a full pledge one.

For example...If i am running the Esxi 5.5 of enterprise edition is running on one of the blades...the vswitch configuration of each blade would be different? 

LUN cloning to the spare blade is mandate right ? then only i can move the particular blade & its profile on to spare blade along with OS ?

I hope there are alot of limitations with this failover point ? I hope UCS won't give full redundancy?

Please advise ?

[Walter Dey · Accepted solution]

Hi

If you move a SP from one blade to another the following has to be taken into account

- is this within one UCS domain, or cross UCS domain (compare e.g. global pools, global SP in the context of UCS central)

- Blade hardware has to compliant (number of I/O adaptors, processor familiy (VMware EVC !), PCI bus structure); no problem, if the target blade has e.g. more memory......

- FC zoning is no problem, will be automatically moved

In summary, moving a SP works within a UCS domain, as well as across UCS domain, if carefully planned and implemented.