Sun SPARC(R) Enterprise M3000/M4000/M5000/M8000/M9000 (OPL) Servers: How To Disable CPU's Across The System Reboot

Asset ID:	1-71-1564303.1
Update Date:	2018-03-05
Keywords:

Solution Type Technical Instruction Sure

Solution 1564303.1 : Sun SPARC(R) Enterprise M3000/M4000/M5000/M8000/M9000 (OPL) Servers: How To Disable CPU's Across The System Reboot

Applies to:

Sun SPARC Enterprise M5000 Server - Version All Versions to All Versions [Release All Releases]
Sun SPARC Enterprise M9000-64 Server - Version All Versions to All Versions [Release All Releases]
Sun SPARC Enterprise M9000-32 Server - Version All Versions to All Versions [Release All Releases]
Sun SPARC Enterprise M3000 Server - Version All Versions to All Versions [Release All Releases]
Sun SPARC Enterprise M4000 Server - Version All Versions to All Versions [Release All Releases]
Information in this document applies to any platform.

Goal

How to disable CPU's permanently on Mx000 servers ?

Solution

The short answer is: The M-Series does not support a blacklisting mechanism. Hence you can't disable CPU's.

But there are ways to manage CPU resources:

How to manage CPU resources

Physically
- physical method : removal of CPU boards
on XSCF
- dynamic reconfiguration of quad XSB boards ( XSB's containing CPU + memory and optionally IO )
on Solaris
- psradm, optionally executed in a start script
- another alternate method may using NON global zones with appropriate CPU resources

Example 1 : Commands to achieve CPU reduction with dynamic reconfiguration of a quad XSB board

(4 CPU's in Solaris = 2 physical cores = 1 CPU chip, but also the associated memory will be deconfigured )

initial prtdiag output before DR operation:
[ snip ]
==================================== CPUs ====================================

      CPU                 CPU                         Run    L2$    CPU   CPU
LSB   Chip                 ID                         MHz     MB    Impl. Mask
---   ---- ---------------------------------------- ----   ---    ----- ----
00     0      0,   1,   2,   3                       2150   5.0        6 144
01     1     40, 41, 42, 43                       2150   5.0        6 144

============================ Memory Configuration ============================
[ snip ]

XSCF> showdomainstatus -a
DID         Domain Status
00          Running

XSCF> showboards -va
XSB R DID(LSB) Assignment Pwr Conn Conf Test    Fault    COD
---- - -------- ----------- ---- ---- ---- ------- -------- ----
00-0   00(00)   Assigned    y    y    y    Passed Normal   n
00-1   00(01)   Assigned    y    y    y    Passed Normal   n
[ snip ]

XSCF> deleteboard -yc unassign 00-1
XSB#00-1 will be unconfigured from domain immediately. Continue?[y|n] :y
Start unconfiguring XSB from domain.
Unconfigured XSB from domain.
XSB power off sequence started. [1200sec]
0..end
Operation has completed.

XSCF> showboards -va
XSB R DID(LSB) Assignment Pwr Conn Conf Test    Fault    COD
---- - -------- ----------- ---- ---- ---- ------- -------- ----
00-0 * 00(00)   Assigned    n    n    n    Passed Normal   n
00-1   SP       Available   n    n    n    Passed Normal   n
[ snip ]

XSCF> showdomainstatus -a
DID         Domain Status
00          Running

Verification of the result after DR operation took place:

final prtdiag output and psrinfo:

[ snip ]
==================================== CPUs ====================================

      CPU                 CPU                         Run    L2$    CPU   CPU
LSB   Chip                 ID                         MHz     MB    Impl. Mask
---   ---- ---------------------------------------- ----   ---    ----- ----
00     0      0,   1,   2,   3                       2150   5.0        6 144

============================ Memory Configuration ============================
[ snip ]

bash-3.2# psrinfo
0       on-line   since 07/19/2013 19:57:54
1       on-line   since 07/19/2013 19:57:54
2       on-line   since 07/19/2013 19:57:54
3       on-line   since 07/19/2013 19:57:54

Example 2 : Commands to achieve CPU offlining with psradm (optionally executed in a start script)

pasradm syntax for offlining CPU's:
psradm -f <cpu#>

initial prtdiag and psrinfo output before rc3 script applied and Solaris rebooted:
[ snip ]
==================================== CPUs ====================================

      CPU                 CPU                         Run    L2$    CPU   CPU
LSB   Chip                 ID                         MHz     MB    Impl. Mask
---   ---- ---------------------------------------- ----   ---    ----- ----
00     0      0,   1,   2,   3                       2150   5.0        6 144
01     1     40, 41, 42, 43                       2150   5.0        6 144

============================ Memory Configuration ============================
[ snip ]

bash-3.2# psrinfo
0       on-line   since 07/02/2013 21:22:12
1       on-line   since 07/02/2013 21:22:13
2       on-line   since 07/02/2013 21:22:13
3       on-line   since 07/02/2013 21:22:13
40      on-line   since 07/02/2013 21:22:13
41      on-line   since 07/02/2013 21:22:13
42      on-line   since 07/02/2013 21:22:13
43      on-line   since 07/02/2013 21:22:13

example script:

e.g. saved as /etc/rc3.d/S98cpu_disable which must be hardlink from /etc/init.d/cpu_disable

bash-3.2# ln /etc/init.d/cpu_disable /etc/rc3.d/S98cpu_disable
bash-3.2# ls -li /etc/init.d/cpu_disable /etc/rc3.d/S98cpu_disable
201272 -rwxr-xr-x 2 root root 232 Jul 23 21:09 /etc/init.d/cpu_disable
201272 -rwxr-xr-x 2 root root 232 Jul 23 21:09 /etc/rc3.d/S98cpu_disable

example content of /etc/init.d/cpu_disable

#!/bin/sh
#
# script to disable dedicated cpu's ( every second CPU )
#
index=0
for cpu in `/usr/sbin/psrinfo|/usr/bin/awk '{print $1;}'`; do
if [ $index -eq 1 ]
then
psradm -f $cpu;index=0
else index=1
fi
done
exit 0

Verification of the result:

bash-3.2# last|egrep "psradm|reboot"
psradm    043 off                      Tue Jul 23 22:32   still logged in
psradm    041 off                      Tue Jul 23 22:32   still logged in
psradm    003 off                      Tue Jul 23 22:32   still logged in
psradm    001 off                      Tue Jul 23 22:32   still logged in
reboot    system boot                   Tue Jul 23 22:32
reboot    system down                   Tue Jul 23 22:30
[ snip ]

prtdiag ( unchanged, all CPU's visible, but not all active ) and psrinfo
[ snip ]
==================================== CPUs ====================================

      CPU                 CPU                         Run    L2$    CPU   CPU
LSB   Chip                 ID                         MHz     MB    Impl. Mask
---   ---- ---------------------------------------- ----   ---    ----- ----
00     0      0,   1,   2,   3                       2150   5.0        6 144
01     1     40, 41, 42, 43                       2150   5.0        6 144

============================ Memory Configuration ============================
[ snip]

bash-3.2# psrinfo
0       on-line   since 07/23/2013 22:32:32
1       off-line since 07/23/2013 22:32:56
2       on-line   since 07/23/2013 22:32:33
3       off-line since 07/23/2013 22:32:56
40      on-line   since 07/23/2013 22:32:33
41      off-line since 07/23/2013 22:32:56
42      on-line   since 07/23/2013 22:32:33
43      off-line since 07/23/2013 22:32:56

Errors for rc3 scripts can be found in /var/svc/log/milestone-multi-user-server:default.log

For more information on lecacy rc scripts usage and trouble shooting please refer to:
"troubleshooting of SMF legacy rc init scripts in Solaris 10 (Doc ID 1438926.1)"

If you referring to Oracle database licensing for a solaris-zone, the following thread in the
My Oracle Support Community may be interesting too

(follow the link to the community and search for the title of the the thread "Capped-CPU for Oracle database licensing in a solaris-zone" using the top right search box)

Capped-CPU for Oracle database licensing in a solaris-zone

To discuss this information further with Oracle experts and industry peers, we encourage you to review, join or start a discussion in an appropriate
My Oracle Support Community - Oracle Sun Technologies Community.

Moving to internal due to Jive migration: URL for thread redirection changes

re-edit when Jive migration is done after 31 Jan 2014

Capped-CPU for Oracle database licensing in a solaris-zone

once migrated we can update the URL and remove the search recommendation

References

<NOTE:1438926.1> - Troubleshooting of SMF Legacy rc init scripts in Solaris 10

Attachments

This solution has no attachment