[Info-vax] reboot due to network failure
Keith Parris
keithparris_deletethis at yahoo.com
Thu Dec 22 13:10:33 EST 2011
On 12/13/2011 5:56 AM, Phillip Helbig---undress to reply wrote:
> All the nodes have the same number of votes. However, the one which
> survived does not have the highest SCSSYSTEMID.
The code which compares SCSSYSTEMID values uses the 6-byte (48-bit)
DECnet Phase IV-style MAC address format of the SCS System ID, with the
two low-order bytes being the SCS System ID value, but with the two
bytes swapped because the Ethernet standard was apparently developed on
Big-endian hardware, and then the comparison is a character-string
comparison in Little-endian order, low-order byte first. So while
deterministic, the decision isn't easy to predict.
If you need a particular node to be the one to survive, give it more votes.
Note that if connectivity loss is partial, in the case of a tie in
numbers of votes between two possible subsets of the cluster, the number
of nodes in a subset is the next factor used is selecting the optimal
subset to survive. If both votes and nodes are equal, then the
comparison of SCS System ID values (as described above) for the set of
nodes running the latest version of cluster software is done, with the
subset having the node with the "highest" value (using that
character-string comparison) surviving.
More information about the Info-vax
mailing list