Forum Discussion

Sean_-'s avatar
Sean_-
Level 5
19 years ago

Is I/O Fencing applicable in this scenario?

Current Scenario -
2-nodes VCS cluster version 4.0 and Solaris 8 OS
2 units of SAN Storage (mirrored across each other using VERITAS Volume Manager)
The 2-nodes cluster are connected via ISL
The 2 units of SAN storage are connected via FC ISL

My question is whether I could configure I/O Fencing and coordinator disk using the SAN storage in the mentioned scenario ? (because there are 2 SAN storage)

If not what other method to prevent split brain ?

Thank you.
business continuity
Cluster Server
  • Gene_Henriksen's avatar
    Gene_Henriksen
    Level 6
    19 years ago
    Yes you could configure with 1 coordinator disk on one array and 2 on the other.

    Other methods of avoiding splitbrain include:
    1) the use of multiple heartbeats (add low priority links over public links),
    2) the use of the preonline_ipc script Note: it is located in the sample_triggers directory, move to the triggers directory and turn on Preonline for service groups to cause the trigger to ping for the IP in the SG prior to online, (if you understand Perl, read the script, it also checks to see if it is isolated (no public IP links working, etc)
    3) insuring that the LLT links are over totally independent hardware infrastructure (no shared lines or switches).

    IF you lose the array with 2 of the coordinator disks and one system goes down, the other will not be able to get a majority of the coor disk and will also panic, on reboot the system seeing only one coord disk cannot start without intervention. In this case you must weigh the value of saving your data from corruption via split brain against the cost of intervention in this case.
  • Sean_-'s avatar
    Sean_-
    Level 5
    19 years ago
    Thank you Gene for your reply, it is helpful.

    No other alternative for I/O fencing for my scenario ? (doesn't require user intervention, for example)

    How about having coordinator disks on the mirror volumes of SAN storage A and SAN storage B ? (so that even Site B encounter with disaster, the Cluster Node A able to access the coordinator disks (mirror volumes) at SAN storage A)

    SAN storage A and SAN storage B having same data, however located at different location. Cluster Node A is connected to SAN storage A and Cluster Node B is connected to SAN storage B.

    Alternative to I/O fencing, disk heartbeat is it recommended to be used for preventing split brain ?