ZONES: scheduling, TS scheduling and projects and tasks, FSS scheduling without zones, FSS scheduling and FSS within zones.
1. explanation scheduling
-draw the circle, ts_tq, ts_quantum, ts_slpret
show dispadmin -l ->to see the current schedulers and their place within the scheduling circle.
# show dispadmin -c TS -g |more
-g = get
-explain the various schedulers, user priority PRI and tsuprlim
2. show the schedulers and change the User Priority (needed to calculate PRI) with TSS
# priocntl -l
show how to get the scheduling class for a PID with
# ps -e -o user,pid,comm,class,pri | more
show the current scheduling parameters for a PID with:
$ /bin/cpuhog.pl &
$ priocntl -d -i pid <pid>
PBIND ALL THE CPUHOG PROCESSES TO THE SAME CORE / CPU IN A MUTICORE / PROCESSOR ENVIRONMENT.
$ psrinfo
Show how the processes if scheduled amongst the various cores / cpu’s.
$ mpstat 3
$ /usr/sbin/pbind -b 1 <pid>
Start a process with a lower USER priority:
$ priocntl -e -c TS -p -15 /bin/cpuhog.pl &
[2] 1563
-p = average Time Slice User PRIority= TSURPRI
-m = Time Slice User PRIority LIMit = TSUPRILIM will always be >= -p
If you set -m < -p then -m will overwrite -p.
Use prstat to see the result.
LOWER the parameters of the fastest process as end – user:
$ priocntl -s -c TS -m -12 -p -20 -i pid 2563
$ priocntl -d -i pid 1563
Use prstat to see the result.
Raise the slowest running process’s user priority as root:
# priocntl -s -c TS -m 15 -i pid 1563
# priocntl -d -i pid 2563
Raise TSUPRILIM to the max as root:
# priocntl -s -c TS -m 60 -i pid 1563
Kill the perl – scripts:
# pkill -9 cpuhog.pl
3. Introduce projects and tasks
Create user “guest”:
# mkdir -p /export/home
# useradd -d /export/home/guest -m -s /bin/bash guest
# passwd guest
# vi /etc/project
lowproject:11:testing low priority projects:guest::project.cpu-shares=(privileged,1,none)
medproject:12:testing low priority projects:guest::project.cpu-shares=(privileged,5,none)
highproject:13:testing low priority projects:guest::project.cpu-shares=(privileged,10,none)
Exit as user ‘guest’ and become user guest again with ‘su – guest’. Find out the projects available for user guest:
$ ip -p → shows the default project
$ projects → shows all available projects
And start some processes in the right project as user “guest”:
$ newtask -p lowproject /bin/cpuhog.pl &
[1] 5061 &
$ newtask -p medproject /bin/cpuhog.pl &
$ newtask -p highproject /bin/cpuhog.pl &
NOW TAKE THE TIME TO PBIND THE PROCESSES TO THE SAME CORE IF NECESSARY.
$ /usr/sbin/pbind -b 3 <pid>
And check their parameters:
# priocntl -d -i pid 5061
TIME SHARING PROCESSES:
PID TSUPRILIM TSUPRI
5063 0 0
# ps -o user,pid,args,class,project,projid,taskid,pri -p 5061
USER PID COMMAND CLS PROJECT PROJID TASKID PRI
root 5061 /usr/bin/perl /bin/cpuhog.pl TS lowproject 11 163 40
While running prstat, as root change the scheduling for lowproject:
# priocntl -s -c TS -p 15 -m 22 -i projid 11
And check:
# priocntl -d -i pid 5061
TIME SHARING PROCESSES:
PID TSUPRILIM TSUPRI
5061 22 15
And the scheduling for project “lowproject”:
# priocntl -d -i projid 11
TIME SHARING PROCESSES:
PID TSUPRILIM TSUPRI
5061 22 15
This process “5061” should now be scheduled faster on the SAME CORE than the others since it is a task within the project called “lowproject”.
Notice however that this type of scheduling is hard to do on a larger system with multiple applications since it is deficult to predict the outcome of scheduling with TS.
The solution: FSS
Kill all existing perl – scripts.
$ pkill -9 cpuhog.pl
4. Introduce FSS
Change the default scheduler to FSS as root:
# dispadmin -d FSS —>> show /etc/dispadmin.conf OR move the scheduling class of running processes to FSS:
# priocntl -s -c FSS -i all or -i pid 1 –>> here “1”=pid of init for example.
The above command move all scheduled proc’s to the FSS class if they were in TS / IA before.
NOTE: default share=1
To check:
# ps -e -o user,pid,args,class,project,projid,taskid,pri
Start the user’s processes:
# su – guest
Check the default
$ id -p
Check that all proc’s are running under FSS.
$ ps -e -o user,pid,comm,class,project,projid,taskid,pri
and start a few new processes:
As root first raise the default shares (=1) on the GLOBAL ZONE:
# prctl -n zone.cpu-shares -v 20 -r -i zone global
$ newtask -p lowproject /bin/cpuhog.pl &
$ newtask -p medproject /bin/cpuhog.pl &
$ newtask -p highproject /bin/cpuhog.pl &
Take the time to bind all processess to the same core / cpu.
$ /usr/sbin/pbind -b 3 <pid>
Check:
To see the scheduling parameters:
# prctl -i pid <pid> | more
[…]
project.cpu-shares
privileged 5 – none –
[…]
Or:
# prctl -i task <taskid>
Calculate the CPU load based upon the Share distribution:
CPU%=((shares / total shares)*100%)/<number of cores or cpu’s>
Compare the outcome with prstat.
As root, change the User Priority for individual processes or projects::
# prctl -n project.cpu-shares -v 10 -r -i project lowproject OR -i pid <pid>
Check the new settings:
# prctl -i project lowproject
[…]
project.cpu-shares
privileged 10 – none –
[…]
See the influence and calculate the outcome.
Afterwards kill all perl – scripts.
$ pkill -9 cpuhog.pl
5. Perform these actions on Zones
Check the current setting of a zone:
# prctl -n zone.cpu-shares -i zone global –>> should show “1”
If using a WHOLE root – zone, as root copy the cpuhog.pl and memhog.pl scripts to all zones:
# cp -p /bin/memhog.pl /bin/cpuhog.pl /export/zones/zone1/root/usr/bin/
Start the cpuhog.pl script in a couple of zones.
To see the number of shares configured use:
GLOBAL # prctl -n zone.cpu-shares -P -i zone zone1
Bind all cpuhog.pl processess to the same core from the Global zone:
GLOBAL # pbind -b 3 12471
After a while, the load per zone should be the same on the core / cpu being used.
Supply different shares to a zone, the default for the GLOBAL zone is ”1”.
Set more shares on a none – global zone:
GLOBAL # prctl -n zone.cpu-shares -v 10 -r -i zone zone1
GLOBAL # prctl -n zone.cpu-shares -v 5 -r -i zone zone2
GLOBAL # prctl -n zone.cpu-shares -v 1 -r -i zone zone3
Check the distribution of the CPU resources per zone with prstat -Z.
# prctl -n zone.cpu-shares -i zone zone1
Now change the settings:
# prctl -n zone.cpu-shares -r -v 10 zone3 etc.
