IBM RS/6000 SP, Problem Determination Manual

Page: 124 / 316

This soft copy for use by IBM employees only.
token, it will verify if it is still the backup before it takes over for the primary.
The backup daemon cancels this thread when it is no longer the backup.
4.4.3 Secondary Node Behavior
Every node, including the primary and the backup primary, executes the
following steps in order to start up the switch:
1. Builds the “ i d e a l ” database based on the topology file
2. Modifies the “ i d e a l ” database based on “ d e l t a s ” received from the primary
node
3. Calls the route table generator to build the routes from this node to all other
nodes
4. Tells the switch protocols (IP and User Space) which destinations are (or are
not) available
5. Loads the routes down to the adapter
The deltas are messages indicating uninitialized/fenced nodes and/or
uninitialized/fenced links received from the primary node.
For the most part, the secondary node repeats steps 2 through 5. The first step
applies only on IPL or when the topology file is changed (such as with switch
partitioning).
Secondary nodes send/receive service packets to/from the primary node only:
The secondary node acknowledges the primary node for each database delta
message it receives from the primary node.
4.4.4 Recovery from a Switch Failure
If a switch fabric or node failure occurs, the primary node must log the incident,
and decide whether the active configuration must be adjusted to render the “ill”
component(s) inactive. If the failing component is a switch link, the route table
must be updated so the faulty link is no longer used. If a non-primary node is no
longer reachable, the primary will fence that node.
If the primary node fails, the primary-backup must “take over.” It must:
•
Fence the primary
•
Assume the primary personality
•
Choose a new primary
• Run switch initialization
4.5 Switch Commands
Eannotator
- Annotates the connection labels onto the topology file.
Figure 47 on page 98 shows some of the connection labels in a topology file.
Below is an example of a connection label:
L01-S00-BH-J18 to L01-N1
104
SP PD Guide