Will san booted alpha survive switch failure?

[javelina]

Currently we have two es80's with a single hba connected to a san switch with eva6100 array. They boot off the san. I would like to add a second hba connection to a second san switch. What I'm not clear on is whether that will allow the alpha's to survive the loss of a san switch or if it simply allows a quick reboot option.

What makes me think that it might not survive the loss of a switch is that when bootdef_dev is set to one unit on the san via multiple paths, the boot will try each path in turn until it finds one that is connected. So really when it boots it only uses one path. If that path is lost while the os is running, will it search the other paths specified in bootdef_dev and be able to keep things running?

It seems like it would be engineered to survive but I don't want to push for this implementation to only find out later that it really doesn't work that way and the os can access data via multiple paths but the console can only really connect to the boot disk via one path at a time which is determined at boot time.

Thanks,
Jeff

[Kevin Raven (UK) · Accepted solution]

It will survive.
When you boot and do a show device/multi you will see all available path to the disk.
We have production clusters that boot from EMC mulit-pathed system disk.
We have lost HBA on servers running OpenVMS and after 5.5 seconds it will switch to use the other HBA card. Proven and works.
Of course if you have a cluster and the correct params set vias sysgen, then you can pull all links to the HBA cards i.e. Server has no valid HBA cards linking to disks and it will access disks via abnother server in the cluster, that then uses it's HBA cards to access disks. How cool is that.
Give it another 20 years and this OpenVMS O/S will catch on ...lol

[Kevin Raven (UK)]

p.s. We run in a production environment where 7 seconds is a mjor outage.
Clusters run 24*7 ....6 days per week.
In 24 months ...outages = 1.
Outage was not OpenVMS issue!
If i had to trust my life support system to an OS. Then I would opt for OpenVMS in a cluster.
Who would go for Windows ?
or Veritas clustering ?

[Ivan Ferreira]

For sure, it will survive. The bootdef_dev is used at boot, but when the OS is up, you have all paths to the VDISKs available.

[javelina]

Thanks guys. Good to have confirmation.