Lecture 10 - Scheduling

We talked a lot about running along until we block, then pick who gets to run, but how is this done?

Process States

Remember at any time we have one of 3 process states:

• Running
• Blocked
• Ready

We have the difference between a policy and a mechanism:

• policy: How we want things to behave (graduate students should finish within 7 years)
• mechanism: How we're going to make it happen (cut off funding vs. throwing them out)

Process Types

There's two types of process types:

• IO Bound: characterized by short bursts of computation before blocking IO
  • Might want to give priority because they can get done and go back to sleep (hide IO latency)
  • Also, more likely to be interactive
• Compute Bound: characterized by long bursts of computation before blocking on IO (or a semaphore).

Scheduling is mandatory in two cases:

1. When a process exits
2. When a process blocks

Scheduling Criteria

It might be desirable under a few other conditions:

1. When a new process is created
2. When an IO interrupt occurs
3. When a timer interrupt occurs

What makes a good algorithm?

• fairness: make sure each process can get its fair share

• efficiency/utilization: keep the CPU busy as close to 100% of the time (don't have busy waits)

• response time: minimize response time for interactive users

• turnaround: minimize turnaround time for batch users.

• throughput: maximize the number of jobs processed per time

Examples of Non-Preemptive Scheduling

For instance FCFS; first-come-first served. You do the shortest job first:

• provably optimal
• Problem: Starvation

This is the bubble-sort method of organizing jobs.

Another example is round-robin.