Traditional Unix Scheduling

1
Traditional Unix Scheduling

• Traditional means
• No consideration on realtime processes.
• Typically 4.3BSD and SVR3
• MFQ-based
• Preemptive (time quantum based)
• Round robin in the same priority queue
• Same time quantum to different priority queues
• Dynamic priority assignment
• Priority
• Integer value between 0 and 127
• Numerically lower values correspond to higher
  priorities.
• Priorities between 0 and 49 are reserved for the
  kernel
• Priorities between 50 and 127 are for user-level
  programs

2

• PCB the most important data structure in an OS
• Process identification
• PID
• CPU state information
• The contents of CPU registers
• Program counter
• Indicates the address of the next instruction to
  be executed.
• Process control information
• Event that a process is waiting on.
• I/O status (e.g., open files)
• Scheduling info

Process Image
Process Control Block
User Program
User Data/Stack
Kernel Stack
3

• PCB contains, as process scheduling info,
• Current scheduling priority (pri)
• User mode priority (usrpri)
• Measure of recent CPU utilization (cpu)
• Nice value (user-controllable adjustment factor
  nice)
• Current scheduling priority (pri)
• Used to decide which process to schedule.
• Same as user mode priority (usrpri), when a
  process is in user mode.

• Goes down to place higher priority to kernel mode
  processing upon I/O completion.
• Down to a predefined priority (0 lt sleep
  priority lt 49).
• 10 for file access
• 20 for disk I/O
• This enables a prompt I/O completion (a prompt
  completion of system call).
• Replaced with the current user mode priority
  (usrpri) when it is back in use mode.

4
7-State Process Behavior Model
New
admitted
Dispatched to a CPU
activated
Ready,Suspended
Ready
Running
Terminated
exit
Suspended
interrupted
Event completion
I/O or event wait
Blocked
Blocked,Suspended
activated
Suspended

• Ready The process is in main memory and
  available for execution.
• Blocked The process is in main memory and
  blocked on an event.
• Blocked, suspended The process is in virtual
  memory and blocked on an event.
• Ready, suspended The process is in virtual
  memory but is available for execution as soon as
  it is loaded to main memory.

5

• User mode priority (usrpri)
• Depends two factors
• Recent CPU utilization (cpu)
• Nice value (nice)
• CPU utilization
• Initialized to 0 when a process is created.
• Incremented (for the currently-running process)
  at every clock interrupt.
• 10 msec

• Nice value
• An integer value between 0 and 39
• The default value is 20.
• Increasing this value decreases the priority.
• Only OS administrators can decrease it.
• The nice command is used to change the nice value
  for each process.
• Accepts -20 to 19 as an increment to the current
  value.

6

• Every second, user mode priority (usrpri) is
  updated as follows

• cpuj(i)
• CPU utilization by process j through interval i.
• Unit time for i 10 msec.
• DR decay rate
• cpuj is reduced by a decay rate.
• 1/2 in SVR3.
• PUSER (baseline priority) 50
• used to divide processes into fixed bands of
  priority levels.

7
An Example
DR 1/2 PUSER 50 nice 0 3 processes in
the same priority band Time quantum/process 1
sec
8
4.3BSD Scheduler
dispatched

• MFQ across RQs
• If all other ready processes are on lower RQs,
  the current process continues to run even if its
  quantum has expired.
• Priority update
• Once every second
• Process promotion and demotion across RQs
• Timing of context switches
• The current process blocks on a resource or
  exits.
• Voluntary context switch
• RR-based preemption
• Priority-based preemption
• I/O (event) completion

CPU
Run Queue 0
admitted
Run Queue 1
Run Queue 31

• 32 run (ready) queue
• A queue for 4 adjacent priorities
• RQ0 for priorities 0 to 3
• RQ1 for priorities 4 to 7
• RR in each RQ
• RR time quantum 100 msec
• The highest-priority process is always scheduled
  first in each RQ.

9
HW

• The decay factor used in 4.3BSD is

• LoadAverage
• The average number of ready processes over the
  last second.
• Discuss how this decay factor impacts process
  scheduling.
• Compare the scheduling algorithms in SVR3 and
  4.3BSD.

10
Considering Process Groups

• In traditional Unix scheduling,
• A priority is assigned per process.
• Processes compete for CPU time.
• What about a group of processes?
• Processes created by each application
• Processes created by each user
• Processes in the same session (or job)
• ps -j
• ps -Aj

11
Fair Share Scheduling (FSS)

• Group-aware priority assignment
• Group application, user or session
• Assigns the same/similar priority (share of the
  CPU) to the processes in the same group
• Give fewer CPU time to processes that have had
  more than their fair share
• Give more CPU time to processes that have had
  less than their fair share.

• Considers the CPU utilization history of a group
  of processes
• along with the CPU utilization history of
  individual processes.
• e.g., 25 share of the total CPU time, when there
  are 4 groups.
• Can assign different priorities to different
  groups
• Foreground and background processes
• OS administrators, normal users and guests
• An extension to the traditional Unix scheduler.

12

• Every second, user mode priority (usrpri) is
  updated as follows

• cpuj(i)
• CPU utilization by process j through interval i.
• Unit time for i 10 msec.
• gcpuj(i)
• CPU utilization by group k through interval i.
• DRp DRg decay rate
• cpuj and cpuk are reduced by DRp and DRg,
  respectively.
• PUSER (baseline priority)
• The default value 50
• Wk weight assigned to group k.
• 0 lt Wk lt 1 and

13
An Example
Group 1
Group 2
Process 1 usrpri cpu gcpu
Process 2 usrpri cpu gcpu
Process 3 usrpri cpu gcpu
Time
0
50 0 0 1
1 2 2 100
100
50 0 0
50 0 0
10 ms
20 ms
1 sec
50 0 0 1 1
2 2 100 100
100 50 50
50 0 0
1 2
100
DRp, DRg 1/2 PUSER 50 W1 and W2 1/2 3
processes in the same priority band P1 is in
Group 1 P2 and P3 in Group 2 Time
quantum/process 1 sec
2 sec
74 25 25 125
125
100 50 50
75 0 50
3 sec
112 62 62
74 25 25
125
62 0 25 100
125
4 sec
80 31 31 131
131
87 12 62

106 50 62
5 sec
14
Traditional Unix Scheduling
Fair Share Scheduling
Process 1 Usrpri cpu
Process 2 usrpri cpu
Process 3 usrpri cpu
Time
0
50 0 1 2 100
50 0
50 0
10 ms
20 ms
1 sec
75 50
50 0
1 100
50 0
2 sec
62 25
75 50
50 0 1 100
3 sec
75 50
56 12 13 112
62 25
4 sec
78 56
56 12 13 112
62 25