[LinuxFailSafe] Question about membership_compute_and_validate:

Kashif Shaikh kshaikh@consensys.com
13 Dec 2002 14:54:18 -0500


Hello everyone,

I wish there was a failsafe-devel list, but since there isn't here it
goes:

I've been looking into Failsafe's membership algorithm(in the routine on
subject line), and what's interesting about it is that it takes into
account link failures resulting in directly and indirectly connected
nodes and generating membership combination sets, so that a biggest
membership set can be found that the majority will agree upon.

My question is this: how is it possible to have link failures resulting
in having directly/indirectly connected set of nodes?  I mean, no one
really uses token-ring or 10-base-T in ring-topology, but rather
networks are Ethernet based and nodes are usually connected to a
cascaded hub/switch setup in star-topology.  So when the link between
node and hub/switch fails, no one else can ping the node which thus
concludes we will never have a node indirectly connected!

My reason for asking this question was a result of, 'whoa...is a complex
membership algorithm to compute membership of directly connected nodes
from a set of directly/indirectly connected nodes really needed these
days? Overkill?'.  

Maybe I'm missing something, can anyone of the developers
(Paddy,Lars,etc) shed some additional insight? TIA!


Regards,

Kashif Shaikh