Does MC/SG handle FC host bus adapter failure ?

[Prakash Achuthan]

Hi,

I am working on providing high availabilty solution for a network attached storage - HP VA7100 array. The VA array is attached to my m/c using a fibre channel host adapter.

How do I handle Host Bus Adapter (HBA) failure ?
I know MC/SG can handle network card failure, but does it address the HBA failure problem.

regards,

[Animesh Chakraborty]

Hi Prakash,
In case of HBA failure your machine(m/c) will not be able to access data so I guess it will be consider as package failure and package will switch to 2nd node.

[John Waller]

For a true HA solution, then I believe you really should have dual HBA's on your servers and setup alternate paths to access the data. I this case if one controller fails you still have a second to work on.

[Stefan Saliba]

To achieve FC host bus adapter failure you need to have multiple HBA's on the node. You will then need to configure to use PV-Links so that if an HBA fails you will use the alternate path. If you have one HBA it means that if it fails you will have a package switchover.

Good points to keep in mind are that PV-links have to be configured manually in your VG. By adding a second HBA only you will not gain anything.

Secondly it is also worth installing redundant hubs so that if your hub fails you still have alternate paths, cos it is useless having multiple hba's with only one hub.

Good luck

Stefan

[Prakash Achuthan]

Hi,

But, how does MC/SG detect HBA failure ?

Thanks

[G. Vrijhoeven]

Hi,

As far as i know, SG A.11.09 does not detect it. UNIX does and switches. A message is logged in /var/adm/syslog/syslog.log and monitoring tools like OV can send out a trap.

You will also need to have multiple HBA 's to upgrade firmware in a HA env.

Hope this will help,

Gideon

[Santosh Nair_1]

I believe LVM takes care of failing over from one HBA to another if there are PV links. Also, you could configure the disk monitoring features of EMS HA as mentioned in this URL:

http://docs.hp.com/cgi-bin/onlinedocs.py?mpn=B5736-90025&service=hpux&path=../B5736-90025/00/00/1&title=Using%20High%20Availability%20Monitors
Hope this helps.

-Santosh

[John Waller]

It looks like the bottom line is the ServiceGuard will not do anything. What you will probably have to do is either modify or write a new service script which performs some kind of check so if the HBA fails then the monitoring script wil fail , hence initiating a package switch. and a TOC of the failing node. This will help with the system running the package but unfortunatly will not help the redundent node, so it would be possible for the package to switch even if the alternate node also had a HBA failure.
This is why all recommendations go towards multiple HBA'

[melvyn burnard]

ServiceGuard is not designed for reacting to failures such as FC cards. This is handled by the SCSI/LVM subsystem, the same way MirrorDisk/UX handles a failure.
You would have dual FC cards, and yiour PV's woul d have Alternate links configured. In teh event of a failure, the LVM code would handle the problem switching all requests for te failed FC card to the remaing working FC card.
This is true whether or not you have ServiceGuard, as this is LVM functionality