Technical Women

039
Figure 1. Maria Klawe

"Leave a Trace" by CHVRCHES

Today

  • Scheduling goals.

  • Human-computer expectations:

    • Responsiveness

    • Continuity

    • Throughput

  • Simple scheduling algorithms.

Three Cheers for Carl!

Many thanks to Carl for filling in last week.

ASST2 Checkpoint

At this point:

  • If you have not finished ASST2.1, you’re behind.

  • If you’re working on sys_{write,open,close,lseek}…​, you’re OK.

Today’s PSA

  • Nobody wants to work with a jerk—​no matter how talented you are.

Questions from Last Week?

Thread States

We talk about threads—and sometimes the processes containing them—as being in several different states:

  • Running: executing instructions on a CPU core.

  • Ready: not executing instructions but capable of being restarted.

  • Waiting, Blocked or Sleeping: not executing instructions and not able to be restarted until some event occurs.

Thread State Transitions

  • Running → Ready: a thread was descheduled.

  • Running → Waiting: a thread performed a blocking system call.

  • Waiting → Ready: the event the thread was waiting for happened.

  • Ready → Running: a thread was scheduled.

  • Running → Terminated: a thread exited or hit a fatal exception.

Operating systems have data structures to organize threads into these groups which you encountered during ASST1.

Scheduling: What?

What is scheduling?

  • Scheduling is the process of choosing the next thread (or threads) to run on the CPU (or CPUs).

  • We will primarily discuss single core scheduling for most of the week but return to multi-core scheduling issues later.

Scheduling: Why?

Why schedule threads?

  • CPU multiplexing: we have more threads that cores to run them on.

  • Kernel privilege: we are in charge of allocating the CPU and must try to make good decisions. Applications rely on it.

Scheduling: When?

When does scheduling happen?

  1. When a thread voluntarily gives up the CPU by calling yield().

  2. When a thread makes a blocking system call and must sleep until the call completes.

  3. When a thread exits.

  4. When the kernel decides that a thread has run for long enough.

  • #4 is what makes a scheduling policy preemptive, as opposed to cooperative: the kernel can preempt (or stop) a thread that has not requested to be stopped.

Why yield()?

  • We have not discussed yield().

What is the rationale behind having a way for threads to voluntarily give up the CPU?

  • yield() can be a useful way of allowing a well-behaved thread to tell the CPU that it has no more useful work to do.

  • yield() is inherently cooperative. "Let me get out of the way so that another, more useful, thread can run."

Scheduling: How?

Two separate questions here:

  • Mechanism: how do we switch between threads?

  • Policy: how do we choose the next thread to run?

How do we switch between threads?

  • Perform a context switch and move threads between the ready, running, and waiting queues.

How do we choose the next thread to run

  • Nice of you to ask. That’s our focus this week.

Policy v. Mechanism

Scheduling is as example of useful separation between policy and mechanism:

  • P: deciding what thread to run.

  • M: context switch.

  • M: maintaining the running, ready and waiting queues.

  • P: giving preference to interactive tasks.

  • M: using timer interrupts to stop running threads.

  • P: choosing a thread to run at random.

Scheduling Matters

How the CPU is scheduled impacts every other part of the system.

Why?

  • Using other system resources requires the CPU!

  • Intelligent scheduling makes a modestly-powered system seem fast and responsive.

  • Stupid scheduling makes a powerful system seem sluggish and laggy.

Human-Computer Interaction (and Expectations)

What do you expect from your machine?

  • Respond (Click)

  • Continue (Watch, or active waiting)

  • Finish (Expect, or passive waiting)

Respond (Click)

Responsiveness: when you give the computer and instruction, or input, it responds in a timely manner.

  • It may not finish, but at least you know it has started (or understood).

Most of what we do with computers consists of responsive tasks. This is using a computer, and what makes computers different from television.

Examples of responsive tasks:

  • Web browsing: when a link is clicked, retrieve the web page.

  • Editing: when I enter text at the keyboard, place it at the cursor.

  • Chatting: when I hit send, transmit the text to my chat partner.

Continue (Watch)

Continuity: when you ask the computer to perform a continuous task it does so smoothly.

  • Continuity implies active waiting: you are not interacting with the computer, but you are expecting it to continue to perform a task you have initiated.

As computers have started to deliver media, this function is increasingly important.

Examples of continuous tasks:

  • Blinking a cursor.

  • Playing music or a movie.

  • Stupid (!) web animations.

Finish (Expect)

Completion: when we ask to the computer to perform a task—or it performs one on our behalf—that we expect to take a long time, we want it to complete eventually.

  • Completion implies passive waiting: you are asking the computer to continue to deliver interactive performance while working on your long-running task. (We also consider these background tasks.)

  • Unlike responsive and continuous task, background tasks may not be user initiated.

Examples of background tasks:

  • Performing a system backup.

  • Indexing files on my computer.

Click, Watch, Expect

Many applications combine all three system expectations.

Music player:

  • Click: change tracks.

  • Watch: play the selected track.

  • Finish: update album artwork.

Web browser:

  • Click: follow a link.

  • Watch: play web video.

  • Finish: index search history.

Conflicting Goals

Scheduling is a balance between meeting deadlines and optimizing resource allocation:

  • Optimal resource allocation: carefully allocate tasks so that all resources are constantly in used.

  • Meeting deadlines: drop everything and do this now!

Responsiveness and continuity require meeting deadlines—unpredictable or predictable:

  • Responsiveness → unpredictable deadlines. "When the user moves the mouse I need to be ready to redraw the cursor."

  • Continuity → predictable deadlines. "Every 5 ms I need to write more data to the sound card buffer."

Throughput requires careful resource allocation:

  • Throughput → optimal resource allocation. "I should really give the backup process more resources so that it can finish overnight."

Deadlines Win

Humans are sensitive to responsiveness and continuity.

  • We don’t notice resource allocation (as much).

  • Heard: "My computer feels slow."

  • Unheard: "My computer is not using all of its RAM."

Why

  • Poor responsiveness or continuity wastes our time! ("The mouse jumped all over and I couldn’t click anywhere.", "The movie kept stalling and I couldn’t watch it.")

  • Poor throughput usually just wastes computer time. ("The backup took 12 hours but I was sleeping.")

Scheduling Goals

(Or, how to evaluate schedulers.)

  • How well does it meet deadlines—unpredictable or predictable?

  • How completely does it allocate system resources?

    • No point having idle CPU, memory, or disk bandwidth when something useful could be happening.

On human-facing systems, deadlines (or interactivity) usually wins. Why?

  • Your time is more valuable than your computer’s.

(Aside) Realtime Scheduling

We have established that deadlines are important to human-facing systems. This is mainly because systems that don’t meet deadlines are annoying. ("Buffering…​", "Buffering…​", etc.)

There are other classes of systems where the failure to meet deadlines can be fatal.

  • "I meant to get around to running the motion_stop task 1 s ago, but I didn’t quite make it. And…​the robot rolled off of the cliff.

Scheduling Principles: Questions?

Next Time

  • Simple scheduling algorithms.

← →