IBM RS/6000 SP, Problem Determination Manual

Page: 156 / 316

This soft copy for use by IBM employees only.
this information is essential during Estart if the Worm on the primary node is to
successfully initialize the switch in that partition.
In cases where the switch will not initialize within a partition, check that the
topology file was copied correctly into the files directory of the SDR by running
the following commands:
# cd /spdata/sys1/sdr/partitions/<IP alias>/files
# ls -l
The correct file should be listed in this directory. Take a look at the contents to
check that it contains the correct topology.
In addition, check that the SDR object class Switch_partition contains the correct
filename for the topology file that is loaded in the SDR files directory. To do this,
run the following commands:
# export SP_NAME=<syspar_name>
# SDRGetObjects Switch_partition topology_filename
The SP_NAME variable can be used to set the current partition. If this variable
is not set, then the current partition will be the
primary partition as specified in
the node
′
s /etc/SDR_dest_info file. Commands such as
SDRGetObjects
will only
provide information about the current partition. This topic will be dealt with in
more detail later in the chapter. Any discrepancy in the two filenames will cause
Estart
to fail.
In this example of an 8_8 configuration, using layout.1, the two files that will be
loaded into the SDR files directories for each partition should be:
•
/spdata/sys1/syspar_configs/topologies/any.l1.8way1.0isb
•
/spdata/sys1/syspar_configs/topologies/any.l1.8way2.0isb
Run the following commands to check that SDR contains the correct information
about the topology files for each partition:
# export SP_NAME=<IP alias>
# SDRGetObjects Switch_partition topology_filename
The output of this
SDRGetObjects
command should be:
topology_filename
any.l1.8way1.0isb
For the other partition, it should be:
topology_filename
any.l1.8way2.0isb
It does not matter that you have multiple topology files in the files directory of
the SDR (although it may be preferable to remove redundant ones), as long as
the
Switch_partition object class is pointing to the correct one.
Debugging the switch on a partitioned system is similar to that on an
unpartitioned system. The specific differences will be covered throughout this
chapter, but the fundamentals remain the same. To recap briefly some of the
actions that may be appropriate in these circumstances:
•
Run
errpt
on all nodes to look for switch or Worm errors.
• Look at the log files in /var/adm/SPlogs/css on the primary node in the
partition.
• Ensure /etc/SP exists on the primary node in the partition.
136
SP PD Guide