CSE350 Software Design and Engineering

1. CSE350 Software Design and Engineering University of Pennsylvania Professor Jonathan M. Smith http://www.cis.upenn.edu/~jms Office: 254 Moore GRW, Phone: 8-9509 January 9th, 2001

2. Administrative • First group assignment Tuesday, 1/15 • Groups will be formed today • We will use http://www.cis.upenn.edu/~cse350 as the class web site • It’s currently outdated – will fix

3. What is an operating system? • An operating system manages hardware resources • Typically, for multiple users • Protection, services such as file systems (structure), concurrent execution, buffering, etc. • In UNIX (for example), OS functions are privileged – behind a protection boundary

4. The Unix Time-sharing System(slides from J.R. Davin) • D. M. Ritchie and K. Thompson. • The Unix time-sharing system. • BSTJ, 57:6 (July-August, 1978), 1905-1929. • K. Thompson. • Unix implementation. • BSTJ, 57:6 (July-August, 1978), 1931-1946.

5. Features • A hierarchical filesystem incorporating demountable volumes • Compatible file, device, interprocess communication • The ability to initiate asynchronous processes • System command language selectable on a per-user basis • Over 100 subsystems including a dozen languages • High degree of portability

6. Economics of the 1970s • Hardware cost: $40,000 • Software cost: two man-years • Cost recovery for disk space

7. PDP-11/70 Platform • 16-bit word (8-bit byte) • 768 KBytes core memory • System kernel 90 KBytes • Minimal system 96 Kbytes • Two 200 MByte moving-head disks • 20 variable speed (300 to 1200 baud) communication interfaces • 12 communication lines (9600 baud) • Synchronous interfaces (2400 and 4800 baud) for inter-machine file transfer • Nine-track tape, line printer, phototypesetter • Voice synthesizer, digital switching network, chess machine

8. A Unix System in 1978 • User population: 125 • Maximum simultaneous users: 33 • Directories: 1630 • Files: 28300 • Disk blocks (512-byte) used: 301,700 • Daily command invocations: 13500 • Daily (non-idle) CPU hours: 9.6 • Daily connect hours: 230

9. The C Programming Language • System re-coded in C in summer, 1973 • 1/3 bigger than assembler language version

10. Observation "The most important role of the system is to provide a file system." (Page 1907)

11. Filesystem Properties • Internal file structure controlled by application programs • Internal file structure not imposed by system • Hierarchy • Directories as files • "." and ".." convention • Links • Restrictions on namespace topology • Mountable volumes

12. File Types • Ordinary • Directory • Special • Character • Block

13. Filesystem Representation • i-Number • i-List • i-Node

14. File Properties • Owner user-id • Owner group-id • Protection bits • Physical disk (or tape) address of contents • Size • Time of creation • Time of last use • Time of last modification • Number of links • File type

15. File Protection • User (owner) read, write, execute • Group read, write, execute • Others read, write, execute • Set-user-id --drwxwrxrwx • "Execute" permission on directories

16. Filesystem API • filep = open (name, flag) • filep = creat (name) • n = read (filep, buffer, count) • n = write (filep, buffer, count) • location = lseek (filep, offset, base)

17. Why file descriptors? • Consider open(“/home/jms/cse350.txt”) • Consider read(“/home/jms/cse350.txt”) • open() happens once, read() many times • Pay cost of name lookup, permissions checks (existence!) at open(), reuse results, repetitively, at read() • open() returns a descriptor, used by read()

18. Example: copy a file from one place to another main() { char c; int r_fd = 0, w_fd = 1; while( read(r_fd, &c, 1) > 0 ) write(w_fd, &c, 1); }

19. Crossing protection boundary • How does a program access services? • It does so through entry points called system calls • These include read(), write(), open(),close() and creat() • Appear to be subroutines calls in “C”

20. File Space Management • Blocks 0 through 9 indicated in i-node • Blocks 10 through 137 indicated indirectly • Blocks 138 through 16521 indicated doubly indirectly • Blocks 16522 and higher indicated triply indirectly Performance Techniques: caching and read-ahead

21. Special File Naming • Example Name: /dev/foo • Major Device Number: selects driver code • Minor Device Number: selects device instance within class

22. File System Data Structure

23. Process Management API • processid = fork () • execute (file, arg1, ..., argN) • processid = wait (& status) • exit (status) • filep = pipe () • Traps, Signals • Minimalist, integrated process synchronization

24. Process Control Data Structure

25. The Shell and Reusability • Redirection • Stdin, stdout, stderr • Pipes and filters • Argument parsing and globbing • Multitasking • Basic control structures

26. Perspective "The success of the Unix system is largely due to the fact that it was not designed to meet any predefined objectives.“ (Page 1926) Motivation: dissatisfaction with existing facilities.

27. Retrospective Design Considerations • Easy to write, test, run programs • Interactive use • Constraints on size • Self-maintenance • Available source code

28. Easy Programming • Device-independent file abstraction • No "access methods" • Few system constraints on program

29. Influences • fork () from GENIE time-sharing system • I/O API from Multics • Shell concept from Multics • Implementing "system" code as user primitives from Multics

30. Unix Implementation • 10,000 lines of C code • 1,000 lines of assembler code • 800 lines not possible in C • 200 lines for efficiency

31. Observation "Throughout, simplicity has been substituted for efficiency." (Page 1932)

32. Observation "The UNIX kernel is an I/O multiplexer more than a complete operating system. This is as it should be." (Page 1945)

33. Unsupported (unwanted?) features: • File access methods • File disposition • File formats • File maximum size • Spooling • Command language • Logical records • Physical records • Logical file names • Multiple character sets • Operator and operator console • Login and logout