IBM RS/6000 SP, Руководство по определению проблем

Страница: 128 / 316

This soft copy for use by IBM employees only.
4.5.1 The rc.switch Script
The rc.switch is run from the /etc/inittab during boot.
Whenever there is a problem initializing any of the nodes onto the switch, check
that the Worm daemon is running on the relevant nodes:
# ps -ef | grep Worm
If it is not running, then start it by running rc.switch script and check to ensure
the daemon stays up. If the Worm daemon is not running on the primary, then
none of the nodes will be initialized.
Always check the
/var/adm/SPlogs/css/rc.switch.log
to see any abnormal
behavior.
The content of various switch log files will be explained later in this section.
The
/usr/lpp/ssp/css/ifconfig
command configures or displays the network IP
interface parameters for the switch.
Always use this command in conjunction with the switch (css0 device) as
modifications had to made over standard TCP/IP implementation of AIX for this
network. Run the following command on the nodes to check that the
css0
interface is up and all the nodes have the same attributes.
# dsh -a /usr/lpp/ssp/css/ifconfig css0
Ensure that there no inconsistencies. For example, when a node does not have
ARP enabled when others do, it may not be possible to even
ping that node.
Run the following command to resolve this on that node:
# /usr/lpp/ssp/css/ifconfig css0 <switch IP address> arp
A detailed flow chart of this script can be found on Appendix A, “RS/6000 SP
Script Files” on page 229 under A.4, “The rc.switch Script” on page 262.
4.5.2 Switch Initialization
/usr/lpp/ssp/css/Estart_sw is called by /usr/lpp/ssp/bin/Estart script, and is
run only on the primary node (oncoming for SP Switch). It is useful to know that
the timeout variable
LIMIT for the Estart command is located in this file. This
may have to be adjusted for systems that take longer than the default (180 sec).
Related Problem That Has Been Experienced:
The daemon must have enough time to go around and touch all of the switch
resources in the system. Due to heavy load on the primary node, there is a
possibility that the daemon will not finish the job within a timeout window. If the
timeout is reached, the switch network may experience switch faulting. The
problem has been experienced on systems that have primary nodes with heavy
Micro Channel loading. Also, the larger the system, the more time it takes to get
around the network. In this case, the probability increases that the timeout will
be reached.
Later in this section we will discuss how to diagnose this kind of problem by
analyzing the switch log files.
108
SP PD Guide