CS 3733 Operating Systems Notes: CPU Scheduling Examples

VAX VMS Scheduling

The VAX uses a multi-level feedback queue system with 32 priority levels:

16-31 are static real-time queues.
They are scheduled according to priorities on a pre-emptive basis.
A process is assigned a priority when it starts and that determines its queue.
0-15 are dynamic queues.
All processes from a given queue are executed before processes from any lower number queue is considered.
This is non -preemptive except for the quantum.
Each system call has a priority increment.
When an event occurs which causes a process to become executable, its priority is computed as the sum of the base and the increment.
Each time a process is scheduled its priority is decremented, but not below the base.
Processes also age if they spend a long time in a queue.
This allows them to increase their priority.

Windows NT Scheduling (also W2000, XP, maybe Vista)

Similar to the VAX (since it was designed by the VMS designers)

Processes waiting for keyboard or mouse get largest priority increment.
The current window also gets a priority boost.
Scheduling is preemptive even on the dynamic queues.

Traditional UNIX Scheduling Algorithm

UNIX 4.3 uses multilevel feedback queues.

Each runnable process is assigned a number that determines which queue they are placed in.
Lower numbers mean higher priority.
Negative numbers are for system processes which cannot be killed by signals.
First process in lowest nonempty queue is run.
You can nice to reduce your priority.
Time quantum of 0.1 second is used, and priorities are recalculated once a second.
No preemption except for quantum expiration.
Under some flavors of Unix, an interactive process running in a window that has the focus is given a higher priority.
The process's user-mode priority is:
p_usrpri = P_USER + .25 p_cpu + 2 p_nice
here p_cpu is incremented each time the system clock ticks and the process is found to be executing.
It is also adjusted once per second for ready processes using a digital decay filter.

where load is the sampled average length of the run queue over the previous 1-minute interval of system operation.
When a process is blocked for an event, it cannot accumulate CPU time. When a process has been sleeping for longer than 1 second

where p_slptime is an estimate of how long its been blocked and load is the current system load
System does not preempt processes executing in kernel mode. This means it is not good for real-time system.

Linux Scheduling Algorithm

Uses two separate process scheduling algorithms:
one for time-sharing which concentrates on fairness, and
one for real-time tasks where absolute priorities are more important.

Time sharing processes based on a credit system.

When a new task must be chosen, the one with the most credits is run.
Every time the timer interrupt occurs, the running process loses one credit and is removed from the CPU when its credits run out.
If no runnable process has any credits, the credits of all processes in the system are adjusted using the formula:
credits = credits/2 + priority
What does this do if left to run?
Can this starve out processes?

Real-time scheduling, each process has a priority and a scheduling class. The scheduling classes are FIFO=FCFS and RR.

the highest priority real-time task is run first
only if there are no runnable real-time tasks will a time-sharing task run.
for tasks of the same priority it is FCFS
if scheduling class is FCFS, will run until it requests I/O
real-time processes do not preempt other processes