HP UX cluster boot from SAN

[sameerK]

Hi I am at the designing phase. I will have one Superdome SD32 64 way.I will be partitioning it 4 partitions. I will be using service guard to cluster 2 partitions.I understand that these servers do not have local drives, so I will have to boot from storage. I have eva 4000 storage. I have also read in HP CA EVA document that HP UX boot from SAN is not supported under cluster. Does that mean I will not be able to boot the cluster partitions from SAN? if no what is thw rok around

[Torsten.]

Hi,

1) building a cluster in only one box may be not the best idea from high availability point of view.

2) Don't know all relevant docs of course, but HP CA EVA is different compared to an EVA without continuous access. I can imagine that CA is not supported as boot device, but I guess you don't really need CA for this.

[sameerK]

Hi Torsten
I was CA EVA cos I also have a DR site for the storage replication. Actually I am looking at configuring Metro cluster where there will be two more partitions at the DR site wich will participate with the two local partitions in one cluster.Assuming the network infrastructure & links are taken care of, I would like to boot these clustered partitions from SAN.I actually did not get the meaning of not SAN boot not supported in HPUX cluster. Would you know why?

[Torsten.]

IMHO, there are some strong reasons against this.

If you want to replicate one servers boot disk to another location this assumes you want to boot the other server from the same disk. Booting 2 different servers from the same disk? Very unlikely that this will work (needless to talk about "supported" here)! Next: if this would be possible anyhow, how to tell the second server to boot right now?
The basic cluster concept is to have at least 2 nodes up and running, one serves the application and in a case of failure the second node will take over and run the package. Because of this, all nodes have to be up and running all the time.

How to "transfer" the application and data from one node to another is a totally different topic. A possible solution here is to use CA.