I found that Foundry was using this approach for udp probe:

send some garbage data. If ICMP port unreachable is detected, then udp 
health check is considered fail. If no reply from server, it consider 
the port is alive and received the garbage data, and udp health check is 
considered succeed.

does this sound a qualified approach for udp?

comments from Kacheong:

You and I have talked about this using a zero length UDP
packet, instead of some garbage data.  The use of garbage
data can be a problem to some apps.  But for some stack,
zero length UDP packet may be dropped.  If you are going
to do this probe, there should be a tunable to specify
the length of the UDP packet.

And if this probe is used, I guess it should be coupled
with at least a ping probe.  No reply can just mean that
the server is completely dead.  So only if ping probe is
OK, no response to the UDP probe may make sense.

I think this should be brought up in the ilb-dev list.

Any other suggestions?

-Jan

Reply via email to