The Unix Programming Environment

Here are the major Unix topics you should understand:

processes and fork
the shell: a user-mode command interpreter
the current directory and name resolution
arguments
environment variables
$PATH and command execution
inheritance
"everything is a file"
file descriptors
open/read/write/lseek/close
stdio/stdout/stderr
pipes and redirection
dup/exec
foreground and background
signals and kill
uid and groups
user/group/other * rwx file permissions
the setuid bit

Exercises

These may be useful in helping you to become familiar with Unix, and for understanding what we might be asking you to do in a real OS class.

lmbench

[lmbench] lmbench: Portable Tools for Performance Analysis. By Larry McVoy and Carl Staelin. In the USENIX Technical Conference, 1996.

The lmbench distribution, built for our linux systems, can be found in the course directory /usr/project/courses/spring06/cps196 on CS department systems (e.g., login.cs.duke.edu). There are some other simple Unix programs in toys and ctoys that may be interesting or useful. lmbench is also available at /usr/research/playground/cps196/lmbench, accessible from all departmental and research machines.

Read the following sections of [lmbench] to understand some basic issues of benchmarking methodology: introduction, sizing the benchmarks, and timing issues. Read section 5 well enough to understand the bandwidth measures and why they are important. Read 6.3-6.9: syscall overhead, signal handling overhead, process creation overhead, context switching, IPC latency, file system latency, and disk latency.

Run the lmbench latency numbers on your Xen virtual machines. Consider the following questions:

How much better are the latencies than they are on the mid-1990s machines reported in the paper?
How much faster does Moore's Law predict they should be?
What does this test say about Moore's Law and operating systems?
Is this a fair test? What are some likely sources of error?
lmbench is a microbenchmark suite. You should understand how microbenchmarks differ from other kinds of benchmarks. Why are they useful? How can they help predict application performance? What other information is required?

Why aren't operating systems getting faster as fast as hardware?

Running lmbench

You can find the lmbench distribution compiled and ready to go at /usr/research/playground/cps196/lmbench/lmbench. Copy this directory over to your own playground space, or anywhere else you prefer that is accessible from your virtual machine. Read the "README" file to find out how to run the benchmarks.

You can NOT run the benchmarks where they are located now. Copy the directory over to your own space
You CAN use the lmbench3.tar.gz file if you want, but if you do, please see below:
- Go through the Makefile and remove all dependencies to "bk.ver" that you find.
  OR
- Copy my Makefile over the one you get from the tarball.