Scheduling

1
Scheduling

• Scheduling is policy context switch is mechanism
• Scheduling of multiple processes on single CPU
  machine
• Can also be applied to multi-CPU system
• Can also be applied to scheduling threads
• What does scheduler need to consider?
• Type of process
• Compute-bound spends most of time computing
• I/O bound spends most of time reading I/O
• What process is waiting on what other processes?
• What processes are blocking and why?

2
Scheduling Criteria

• Meets policy goal
• Real-time
• Low response time (low turn around time average
  time it takes for job to be completed after
  submission)
• High throughput number of jobs completed per
  hour
• Fair share equal time
• Gang parallel processes (1-to-1 mapping of
  process to CPU)
• Low overhead keep CPU doing useful work as
  opposed to context switching
• Freedom from starvation
• Stability at highest loads

3
Scheduling Algorithms

• Batch off-line, jobs come with resource
  estimates
• Goal high throughput, low turnaround time, high
  CPU utilization
• Timeshare (interactive in Tannenbaum book)
• Goal low response time, meet users expectation
• Real-time
• Goal avoid losing data, avoid quality
  degradation
• Gang for multi-processors

4
Batch Scheduling

• Know how much time is needed to complete job
  ahead of time
• Some intuitive algorithms
• FIFO
• Shortest job first
• Shortest remaining time next
• Shortest job first is provably optimal for
  minimizing run-time proof?
• If resource needs are sufficiently well defined,
  can also ensure no deadlock
• Can use non-preemptive algorithms since this
  reduces process switching

5
Time Sharing (Interactive)

• Scheduler that we most often encounter multiple
  applications running on your PC
• Typical system
• Priority queues with round-robin in each priority
  level
• Priorities of processes change
• Scheduling invariant at all times, highest
  priority process is running

6
Priority Queues with Round-Robin

• Round-robin each process is allowed to run
  specified time quantum
• Quantum a time interval, e.g., 100 msec
• Each priority level is allowed to run for some
  time quantum
• Highest priority processes allowed few quantums
• Lower priority processes allowed more quantums
• Why?

7
How to Assign/Change Priorities?

• Statically based on process type
• Priority 1/f
• f size of quantum used last
• More a process ran, lower its priority
• Least run priority gets highest priority to run
  next

8
Priority Queue (Corbato et al.)

• High priority given 1 quantum lower priority
  given more quantums
• Whenever process completed its allowed quanta at
  a priority level, it moved down one priority
  level
• Round-robin within a level
• Result
• Interactive processes with few instructions gets
  completed quickly
• Long CPU-bound processes moved down in priority
  but were given longer time interval to run

9
Interactive Systems Other Algs.

• Shortest Process Next
• Also used in batch systems
• Interactive jobs tend to take little time and
  give users greater sense of interactivity
• Need estimate of running time for a job
• Guaranteed Scheduling
• Example N jobs, each will get 1/N CPU time
• Keep guarantee by keeping track of how much CPU
  time each process has and how much it is entitled
• Fair-share
• Equal share of CPU to each user (NOT process)

10
Real-time Scheduling

• Categories
• Soft desirable to meet deadlines
• Hard required to meet deadlines
• Periodic vs. aperiodic occurs at regular or
  irregular intervals
• Scheduler may need to know required resources to
  meet deadline ahead of time
• Examples
• Should a real-time video server be hard or soft?
• How about a medical monitor?

11
Determining Schedulability of Real-time Process

• m periodic events
• p period of process
• c CPU time of event I
• Load can be handled if .
• Examples of dynamic real-time system
• Video server
• Number of people connected and getting streamed
  video changes
• Server needs to balance load

i
i
12
Scheduling and Synchronization

• Synchronization is a form of scheduling (among a
  restricted set of processes)
• Common bad interactions
• Priority inversion
• Convoy phenomenon
• Results from harmful coupling between schedulers
  decisions and how access to locked data is
  synchronized

13
Priority Inversion Problem

• Occurs when low-priority process holds lock and
  is descheduled while high-priority process waits
  for lock
• Occurs when processes have priorities and
  scheduler invariant is highest priority ready
  process always runs immediately
• Example
• Low priority process A has lock
• A preempted by high priority process B
• B tries to get lock and blocks
• Medium priority process C runs before A can run
  again

14
Priority Inversion Solution

• Preventing preemption inside critical section is
  not acceptable
• Critical section could be long
• Violates scheduler invariant that highest
  priority process always runs next
• Priority inheritance process in critical
  section inherits (for duration of critical
  section only) the priority of the
  highest-priority process that is blocked
• Higher priority process is blocked, but only for
  duration of critical section
• Critical section can be long, but process can be
  preempted inside of critical section by
  non-blocking high priority process
• Deadlock possible