How do u increase the time between KA packets


Env: AIX 3.2.5; HACMP 3.1.1; 2 node cascading cluster where each node
is an RISC System/6000 Model R24.

The customer would like instructions on how to increase the time
between KA packets in HACMP 3.1.1 in order to prevent a dms timeout.

See item AX1437 for reference.

The first step is to adjust the Failure Detection Rate from Normal to
Slow.  If this is ineffective, then, you can manually adjust the
hbrate of the NIM that you are using:

 1)  cd /etc/objrepos
 2)  odmget -q name=ether HACMPnim > /tmp/nim
 3)  "edit the file"

           FROM:  HACMPnim:
        name = "ether"
        desc = "Ethernet Protocol"
        addrtype = 0
        path = "/usr/sbin/cluster/nims/nim_ether"
        para = ""
        grace = 30
        hbrate = 500000
        cycle = 12

           TO:  HACMPnim:
        name = "ether"
        desc = "Ethernet Protocol"
        addrtype = 0
        path = "/usr/sbin/cluster/nims/nim_ether"
        para = ""
        grace = 30
        hbrate = 1000000
        cycle = 12

  4) odmdelete -o HACMPnim -q name=ether
  5) odmadd /tmp/nim

The hbrate is defined as follows:

    hbrate is the number of microseconds between keep-alives.  The
default value for ethernet is 500000 (which translates to .5 seconds
between keep-alives, or 2 per second)  It must be realized that this 
will slow down the cluster manager, causing it to take longer to 
start and process events.


Act:  We discussed this at length with the customer.  We agreed that
if they are willing to have the cluster failure detection that slow,
then they are free to set the desired values.

Basically, the time that it takes before a dms timeout occurs is

  network failure detection = (hbrate)*(cycles)-1

of the slowest network.

Thus, if you modify the hbrate to be 2 seconds and the cycles to 91 of
the ethernet network, the dms timeout should occur in about 3 minutes.

Note that the failure detection of each network is determined in
parallel.  Thus, the slowest network determines how long it takes for
a dms timeout to occur.

Note also that you would rather keep the hbrate down around 2 to 3
seconds and adjust the cycles to reach your 3 minute goal.  Otherwise,
you will significantly increase the time required for HACMP to start
and stop.

NextAct:  Closing call with customer agreement.

Support Line: How do u increase the time between KA packets ITEM: BJ0970L
Dated: July 1996 Category: N/A
This HTML file was generated 99/06/24~13:30:21
Comments or suggestions? Contact us