Read the notes in this section after you invoke
the cluster_config utility.
Update the LSF Scheduler After Running the spconfig Command |
 |
The SLURM spconfig command
updates the available resources that are known to SLURM. When this
command is run, SLURM reports more accurate information. The HP XC System Software Installation Guide instructs
you to run spconfig when you perform SLURM postconfiguration
tasks.
The LSF batch scheduler integrated with SLURM
relies on SLURM to gather resource information. When you run the spconfig command to update resource information in SLURM,
you must also update LSF.
Enter the following command to ensure that LSF
is updated with the current resource information reported by SLURM
so that its scheduling decisions fully use all cluster resources:
# badmin reconfig
Checking configuration files ...
No errors found.
Reconfiguration initiated |
Hardware Configurations with Mixed Opteron and Xeon Hardware
and InfiniBand Require an Imaging Override |
|
For the InfiniBand interconnect to function properly,
you must configure an imaging override for hardware configurations
with mixed architecture Opteron and Xeon architectures.
Use the procedure in the HP XC System Software Administration Guide, which describes
how to use file overrides to the golden image, to edit the /etc/modprobe.conf.vapi file to have different values for
nodes with an architecture that is different from the head node.
For Opteron nodes in a
hardware configuration with a Xeon head node, edit the /etc/modprobe.conf.vapi file to set the value of the disable_msix parameter
to 1, similar to the following:
options mod_thh disable_msix=1
options mod_rhh disable_msix=1 |
For Xeon nodes in a hardware
configuration with an Opteron head node, edit the /etc/modprobe.conf.vapi file to set thedisable_msix parameter to 0, similar to the following:
options mod_thh disable_msix=0
options mod_rhh disable_msix=0 |
snmptrapd Might Not Start on Systems Discovered With the ic=AdminNet
Option |
|
This release
note only applies to systems that were discovered with the --ic=AdminNet option and systems with hardware configurations
that contain HP blade servers, Modular Cooling System (MCS) devices,
or both. You must apply this release note now as well as any time
you rerun the cluster_config utility.
If you specified the --ic=AdminNet option on the discover command line, the IP address
for the Administration Network and Interconnect Network devices are
the same, causing duplicate addresses to be written to the /etc/init.d/snmptrapd file. Duplicate addresses cause snmptrapd to fail to start.
Resolve this issue by removing the duplicate address
from the /etc/init.d/snmptrapd file and restarting
the snmptrapd service:
Use the text editor of your
choice to open the /etc/init.d/snmptrapd file,
look for the OPTIONS= line, and remove the duplicate
IP address (in this example 172.31.15.254) in the last line:
# snmp_traps service edit: The OPTIONS= line below will supercede those above
OPTIONS="-Lsd -p /var/run/snmptrapd.pid -c /etc/snmp/snmptrapd.conf -C -On
127.0.0.1 172.31.15.254 172.31.15.254" |
Save your change and exit
the text editor.
Restart the snmptrapd service:
# service snmptrapd restart |
Configure Additional Boot Option on ProLiant DL145 Generation
2 Client Nodes With PCI-X InfiniBand Cards |
|
This note
applies only to systems ProLiant DL145 Generation 2 Client Nodes
With PCI-X InfiniBand Cards. Skip this task if the system uses other
networking or interface cards, such as Quadrics, Myrinet or InfiniBand
PCI-Express single data rate (SDR) or double data rate (DDR).
An additional kernel boot option is required on
ProLiant DL145 Generation 2 Client Nodes with PCI-X InfiniBand cards.
Without this boot option, the card will not come up properly, even
though it is seen with the lspci command.
Depending upon the system configuration, follow
one of these procedures to add the boot option:
If the system is homogeneous
and all nodes are ProLiant DL145 Generation 2 nodes, add the additional
kernel boot option (noapic), to the grub.conf file in the golden image, as follows:
Use the text editor of your
choice to edit the golden image version of the grub.conf file:
/var/lib/systemimager/images/base_image/boot/grub/grub.conf |
Add the noapic option to
the end of the kernel boot options as follows:
#boot=/dev/cciss/c0d0
default=0
timeout=5
hiddenmenu
title Linux for High Performance Computing 4 (Savant) (2.6.9-34.4hp.XCsmp)
root (hd0,0)
kernel /vmlinuz-2.6.9-34.4hp.XCsmp ro root=LABEL=/ console=ttyS2 noapic
initrd /initrd-2.6.9-34.4hp.XCsmp.img |
Save your changes and exit
the text editor.
If the hardware configuration
contains one or more HP ProLiant DL145 Generation 2 nodes, add the
additional kernel boot option noapic to the grub.conf file as an override to the system image. The HP XC System Software Administration Guide describes
how to overwrite files delivered in the golden image.
Increase the Imaging Time Out Limit on Systems With Large Disks |
|
If
the system contains disks that are 250 GB or larger in size, you may
need to increase the 45 minute default allowed for imaging by the startsys command.
For example, on systems with disks larger than
250 GB, the --image_timeout=n option of startsys is set to 60 to increase the image time out limit.
You might use a different value depending upon the size of the disks
in the system.
# startsys --image_and_boot --image_timeout=60 |
Printable version
