Compute (part 2)

0. Pitch

• demo: a random process: wall time, system time, user time

 perl -MPOSIX -l -0777 -ne '@f = /\(.*\)|\S+/gs;
    printf "utime: %.2f\nstime: %.2f\ncutime: %.2f\ncstime: %.2f\n",
      map {$_/POSIX::sysconf( &POSIX::_SC_CLK_TCK )}@f[13..16]' "/proc/55789/stat"
utime: 12.16
stime: 1.95
cutime: 0.00
cstime: 0.00

razvan@bifrost:~$ ps -o cputime,etime 55789
    TIME     ELAPSED
00:00:14  2-17:03:00

• demo: a process runs and prints system time and user time

1. Recap: Process Creation and Termination

• demo: posix_spawn()
• demo: orphan and zombie processes
• demo: fork() + exec()

2. Copy-on-Write

• using multiprocessing for parallel work
• diagram: fork and creation of a new address space, copy-on-write
• demo: copy-on-write + access + page faults

3. Overview of Scheduling

• fewer CPUs than threads
• diagram: scheduler distributes threads to CPUs, time multiplexing
• scheduler is a function
• scheduler requirements: list of all threads, list of running threads, accounting of thread running time, list of threads ready to run

4. Thread States

• diagram: the 3 most important states: running, ready, waiting
  • queues: ready queue, blocked queue
• demo: ps + state
• demo: thread is running, thread is sleeping ("rabid" CPU + wait for input)

5. Context Switch

• a thread is a context of execution: IP and SP
• execution context switch: switch a thread with another thread
  • not related to domain switch: user-mode to kernel-mode
• diagram: one thread leaves, another thread enters
  • condition to enter: be in the ready queue
  • conditions to leave: thread time slice expires, thread blocks, a high priority thread appears, thread yields the CPU
• table: voluntary / non-voluntary; go to blocked / go to ready
• demo: list context switches with ps; destination state (blocked, ready)
• demo: process using the buffer cache and not using the buffer cache, number of context switches

6. State Transitions

• diagram: state transitions and causes of each
• demo: measure time in ready queue
• demo load average: number of running / runnable threads per CPU
• context switch overhead

7. Scheduling Attributes / Nuts and Bolts

• system input: thread types: CPU / IO-intensive
• admin input: nice
• demo: nice
• diagram: scheduler input: thread behaviour, niceness; scheduler output: priority, time slice
• metrics: fairness / interactivity vs throughput

Conclusion and Takeaways

• The OS is responsible for scheduling more threads to the existing CPUs.
• The scheduler is a function that selects a thread to run on the CPU.
• The scheduler is invoked voluntarily or non-voluntarily.
• A thread has 3 main states: running, ready, blocked.
• Thread behavior and admin input affects the thread priority while in the ready state and the time slice while in the running case.