UCS Boot from SAN

[dani_bosch]

Hello,

Provided that we boot from SAN, is UCS capable of automatically failing-over the OS boot path, should a fabric failure occur? If the answer is NO, isn't it a big serious issue of booting from SAN?

Thanks a lot,

[Walter Dey]

Hallo Dani

The boot policy can be setup with redundancy; vhba0, primary, connected to fabric A, vhba1, secondary, connected to fabric B. Both paths have visibility to the same boot lun.

Gruss Walter.

[dani_bosch]

Hello Gruss,

But is this redundancy valid for a failover DURING PRODUCTION? I think this redundancy only applies when the service profile boots for the first time, but once already booted, it doesn't automatically failover, should a fabric failure occur. Could you confirm/deny this point?

[Walter Dey]

The policy applies only during the boot process anyway; if a fabric would fail, the policy stays identical of course, it will not change ; therefore if a blade must boot, it tries first the failing fabric path, and after a timeout, it will switch to the redundant one.

This has been tested and works in production.

[dani_bosch]

Sorry Gruss, but are you therefore recognizing that if the fabric the OS is booting from goes down, then the server goes down indeed?? Does that mean that if a fabric goes down in a UCS boot from SAN environment, statistically half of the servers go down????

[Walter Dey]

UCS consists of a dual fabric for Ethernet and FC; therefore if a fabric fails, and you have setup properly NIC Teaming (Ethernet) and Multipathing (fibrechannel), your servers are alive.

I still don't understand your arguments ? are you concerned, that a

- fabric could fail during the boot process

- fabric fails after the server is up and running

Please clarify ?

[dani_bosch]

Gruss, I'm concerned about the path to the BOOT LUN (not the data LUN) after it is up and running...in case this path goes down (remember we're speaking about the BOOT LUN, not the data LUN) are you telling me that there is an automatic failover of this path as well?? (provided that we have MPIO or similar)

[Walter Dey · Accepted solution]

Multipathing applies to boot and data lun and is independant of boot policy

Sorry I don't understand where I misused the term automatic.

The boot policy is a static, very simple construct: first use the first path for booting; it it fails, try the second one ?

Regarding FC multipathing, once the server is up and running: before the fabric failure you have 2 paths, after only one path to the lun (data and boot).

If all uplinks to the Northbound switch fail, or a whole fabric , UCSM will signal link down to all vnics/vhba's of this fabric. (default setting of Network Control Policy -> Action on Uplink failure -> link down)

[dani_bosch]

Alright,I thought that the path to the BOOT LUN was never handled by the MPIO...in this case, everything's fine and there are no high availability drawbacks in using boot from SAN.

Thanks Gruss