1. Alternate Project #4 Process and Thread Performance
CS-3013 Operating Systems Hugh C. Lauer
(Slides include materials from Slides include materials from Modern Operating Systems, 3rd ed., by Andrew Tanenbaum and from Operating System Concepts, 7th ed., by Silbershatz, Galvin, & Gagne)
CS-3013, C-Term 2012
Alt Project #4 – Performance

2. Alt Project #4 – Performance
Project #4, Kernel Message System:–
50 points
Intense kernel hacking and testing
Optional two person teams
Alternate Project #4, Thread and Processes Performance
35 points
Entirely in user space
Individual project only
CS-3013, C-Term 2012
Alt Project #4 – Performance

3. Alt Project #4 – Performance
Objective
To update Slide #36 of the Thread topic
Conduct performance measurements on Linux and Unix systems to discover how fast processes and threads can be created and destroyed
CS-3013, C-Term 2012
Alt Project #4 – Performance

4. Unix Processes vs. Threads
On a 700 Mhz Pentium running Linux
Processes:
fork()/exit(): 250 microsec
Kernel threads:
pthread_create()/pthread_join(): 90 microsec
User-level threads:
pthread_create()/pthread_join(): 5 microsec
CS-3013, C-Term 2012
Threads 4

5. Alt Project #4 – Performance
Approach
Write one or more test programs that
Forks and waits for a process enough times to determine the inherent overhead of creating each new process
Creates and joins a kernel thread enough times to determine the inherent overhead of creating each new thread
Creates and joins a user-space thread enough times to determine the inherent overhead of creating each new thread
CS-3013, C-Term 2012
Alt Project #4 – Performance

6. Alt Project #4 – Performance
Forking Performance
Your test program should fork() a new process and immediately wait for it
The child process should immediately exit with a zero return code
Do this n1 times
Measure the elapsed (wall-clock) time
Divide by n1 to obtain fork()/exit() time of slide
For your machine and operating system
CS-3013, C-Term 2012
Alt Project #4 – Performance

7. Kernel Thread Performance
Your test program should pthread_create() a new thread and immediately pthread_join() it
The newly created thread should immediately exit
Do this n2 times
Measure the elapsed (wall-clock) time
Divide by n2 to obtain pthread_create() / pthread_join()time on slide
For your machine, operating system, and kernel threads
CS-3013, C-Term 2012
Alt Project #4 – Performance

8. User-Space Thread Performance
Obtain and install GNU Portable Thread package (GNU Pth)
Also available on
Search for “pth” under version 11.4
Re-run kernel thread experiment using pth (user-space threads) instead
Obtain thread_create/join() times for performance number for slide
CS-3013, C-Term 2012
Alt Project #4 – Performance

9. Alt Project #4 – Performance
Multiple Experiments
Carry out these three experiments on as many platforms as you can manage
Your virtual machine, plus at least one other
2 points for each additional Linux platform
Must be different in speed, guest vs. host, etc.
3 points for each non-Linux platform
I.e., cygwin, MacOS, etc.
CS-3013, C-Term 2012
Alt Project #4 – Performance

10. Alt Project #4 – Performance
Reporting
For each platform, report
Times returned by the three experiments
Processor type, speed, number of cores, RAM
Type and version of operating system
Guest or host
Operating conditions (stand alone, shared, etc)
Create Excel spreadsheet and chart to report results
In Linux, processor and memory info is displayed in /proc/cpuinfo
CS-3013, C-Term 2012
Alt Project #4 – Performance

11. Submitting Your Project
Due February 26, 11:59 PM
Submit on Turnin under Alt Project 4
Include:–
Code and makefile
Write-up explaining your results and how you got them
Chart(s) and spreadsheet(s) summarizing results
CS-3013, C-Term 2012
Alt Project #4 – Performance

12. Alt Project #4 – Performance
Questions?
CS-3013, C-Term 2012
Alt Project #4 – Performance