Skip to main content
Link
Menu
Expand
(external link)
Document
Search
Copy
Copied
Search cs492-stevens
Lectures
Lecture 18 Template
18 - Scheduling (5) & Concurrency (1)
Outline
Studying Scheduling Algorithms
Scheduling in Linux
Scheduling Practice
Concurrency:
Inter-Process Communication
Announcements
Reading:
Studying Scheduling Algorithms
How are scheduling algorithms chosen?
Kernel can be compiled with different algorithms.
Old default: "Completely Fair Scheduling"
Current default: EEVDF
Extensible scheduler class
5 fully CPU-bound processes created at t = 0
Round Robin
Priority Scheduling
IPC Approaches
Message Passing
Shared Memory
FCFS
SJF
> processing time in minutes
> high weight = high priority
- Round Robin (assume Q << p)
- Priority Scheduling
- FCFS (alphabetical order)
- SJF (non-preemptive)
Determine Average Turnaround Time for the
following schedules:
Modern applications can be made up of many
processes requiring communication & coordination.
IPC must tackle:
4 common approaches:
There are a bunch of ways to do message passing:
2 main interfaces:
The same questions apply to inter-thread
communication as well (using shared memory).
Approaches:
Need to compare performance on expected load.
Deterministic Modeling
when compiled in, can't switch dynamically
implementation of Weighted Fair Queuing
"Earliest Eligible Virtual Deadline First"
sched_ext used to implement custom algorithms
based on Berkeley Packet Filters
Threads do not provide isolation, which is not
always desirable (security concerns)
IPC requires syscalls for communication
how to pass information
how to prevent undue interference
how to sequence access (producer-consumer)
work on sub-schedulers is underway
original 1996 research paper:
uses concept of lag to select eligible processes,
then computes a virtual deadline for EDF
linux history:
more reading:
fixed interval divided proportionally to threads
2016 research paper on its problems:
Default from ~2007 ~ 2024
Queuing Models
Implementation
What is the expected load?
Scheduling in Linux
Scheduling Practice
Inter-Process Communication
PA 3 instructions still pending a few things.
currently due in 2 weeks (same day as midterm)
1)
2)
3)
4)
MOS 2.5, MOS 2.4.1 - 2.4.3
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Collect behavioral traces to model expected load.
Stochastically model expected load.
Implement first, distribute, and collect user feedback
Suffers from sampling problem and traces difficult to collect
Suffers from representation/accuracy of stochastic models
Suffers from very slow iteration cycles
eurosys16
eevdf
2023 proposal
Huaicheng Li VT 2026 slides
2023 article
2026 article
2024 completion
10
2
3
2
5
1
4
6
4
8
A
B
C
D
E
Message Passing via kernel
Shared Memory via kernel
Shared File(s)
Signals
1.
2.
3.
4.
Message Passing
Shared Memory
Shared File
proc A
proc A
proc A
proc B
proc B
proc B
file
shared
kernel
kernel
kernel
msg
msg
msg
msg
-
-
-
-
-
-
-
POSIX message queues:
POSIX shared memory:
System V shared memory
System V message queues
UNIX domain sockets
Network sockets (UDP, TCP/IP, etc.)
Pipes
POSIX Example:
on Linux, the shared memory shows up as a file in
/dev/shm until unlinked
writer.c
reader.c
mqueue.h
Anonymous Pipes:
Named Pipes:
ls | wc -l
$ mkfifo pipe
$ ls -l
$ cat < pipe
$ cat > pipe
$ rm pipe
# create a named pipe pseudofile
# shows file type as `p`
# read from named pipe to stdout
# write stdin to named pipe
# delete the pipe
sys/mman.h