1. CSE 542: Operating Systems
Outline
An Empirical Study of Operating Systems Errors Andy Chou, Junfeng Yang, Benjamin Chelf, Seth Hallem, and Dawson Engler SOSP 2001
Improving the Reliability of Commodity Operating Systems Michael M. Swift, Brian N. Bershad, Henry M. Levy, SOSP 2003
20-Apr-19
CSE 542: Operating Systems

2. CSE 542: Operating Systems
Summary
Where are the errors
Driver code has three to seven times errors than kernel
How are bugs distributed
How long do bugs live
In Linux kernel, about 1.8 years
How do bugs cluster?
Clusterings when programmer ignorance of interface or system rules combines with copy-and-paste
Comparison with OpenBSD and Linux
OpenBSD has higher error rates
20-Apr-19
CSE 542: Operating Systems

3. CSE 542: Operating Systems
Size of linux tree
20-Apr-19
CSE 542: Operating Systems

4. CSE 542: Operating Systems
Errors checked
20-Apr-19
CSE 542: Operating Systems

5. CSE 542: Operating Systems
Where are the errors?
Drivers have more errors than can be accounted by the code size
Drivers are written by many developers who many not understand the kernel
Drivers are not debugged as much as kernel proper
20-Apr-19
CSE 542: Operating Systems

6. CSE 542: Operating Systems
Ages of bugs
Most bugs are fixed quickly
20-Apr-19
CSE 542: Operating Systems

7. Windows 2000 - Failure Analysis.
CSE 542/498J: Operating Systems
Saturday, April 20, 2019
Windows Failure Analysis.
32% of NT 4 faults, 27% of W2k faults
Device drivers
16%
Core NT
43%
Other third- party drivers
Anti-virus
12%
Hardware
Failure
13%
NT4
Drivers for HCL HW 7%
Drivers for NonHCL HW
20%
HW Failure
22%
Anti-Virus 4%
System Config
34%
Other 3rd Party
Kernel code
11%
MSInternalCode 2%
Other IFSDrivers 0%
Windows 2000
Changes in failure profile discuss.
Note relative size of pie much decreased.
Still need more data on all failures hence development of why boot, discuss the problems.
Note drivers due to unsigned, improvement in this area must be customer driven.
Hardware failures need to start handling.
Source: Brendan Murphy, Sample from PSS Incidents:
20-Apr-19
CSE 542: Operating Systems

8. CSE 542: Operating Systems
Nooks functions
Isolation:
Prevent extensions from damaging the kernel
Interposition:
Integrate existing extensions into the Nooks environment
Object tracking
Track all kernel resources used by extensions
Recovery
Detect and recover from a wide variety of extension faults
20-Apr-19
CSE 542: Operating Systems

9. CSE 542: Operating Systems
20-Apr-19
CSE 542: Operating Systems