> I have a doubt about the test Ips in the IPMP group and the daemon's
 > (mpathd) use of it to probe the target systems and detect failure.
 > 
 > Lets say i have qfe0 and qfe1 in the IPMP group group1.
 > Questions are:
 > 
 > 1. Do the test IPs on this interfaces have to be in the same subnet (e.g 
 > 10.5.30.5 and 10.5.30.6 v/s 10.5.30.5 and 10.5.31.5)

All test IP addresses should be on the same subnet -- but this need not be
the same subnet as your data addresses.  That is, assuming multiple
subnets are present on the same wire, you can use another subnet (we often
recommend RFC1918 addresses to reduce the burden on routable IPv4
addresses).  However, there must be other hosts on the link that have
addresses on whatever subnet you choose.

 > 2. The Data IP (which will float) accross this 2 interfaces and which
 > application will be using also have to be in the subnet? (e.g 10.5.30.7)

See above; no.

 > 3. Target hosts, does it have to be the hosts in the same subnet or
 > reachable via immediate router or it can be any where in the
 > network.Please provide example of configuring target hosts (in both
 > scenarios if the 2nd scenario of target host being somewhere in the
 > network table).

See above.

 > 4. Any suggestions on using Link based failure v/s probe based failure
 > detection

As per in.mpathd(1M):

     The in.mpathd daemon  can  detect  NIC  failure  and  repair
     through  two methods: by monitoring the IFF_RUNNING flag for
     each NIC (link-based failure detection), and by sending  and
     receiving  ICMP  echo  requests  and  replies  on  each  NIC
     (probe-based failure detection). Link-based  failure  detec-
     tion  requires  no explicit configuration and thus is always
     enabled (provided the  NIC  driver  supports  the  feature);
     probe-based  failure  detection  must be enabled through the
     configuration of  one  or  more  test  addresses  (described
     below),  but  has the benefit of testing the entire NIC send
     and receive path.

Is there something more you're curious about?

-- 
meem
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to