1. Event-Driven Programming Vivek Pai Dec 5, 2002

2. 2 GedankenBits  What does a raw bit cost?  IDE  40GB: $100  120GB: $180  32MB USB Pen: $38  FireWire:  60GB: $205  120GB: $280  USB:  40GB: $140  120GB: $218  What does a managed bit cost?  NAS  320GB: $3400 (Dell)  7.5TB: $300K (Zambeel)  What does a person cost?  $100K salary + $50K benefits/overhead

3. 3 Extra Project  We need to agree on it  One-page proposal  Idea, implementation, measurement  Report (~10 pages) due on Dean’s Date  Max boost of 2/3 of a letter grade

4. 4 Project 6  Several goals  Performance improvement via caching  Dynamic adjustment to load  Master/slave or symmetric programming  Due on Dean’s Date  Extra credit: 4 points (20 base)

5. 5 Another Random Aside  You may want to read  Flash: An Efficient and Portable Web Server  Available from my home page  Caveat: far more complicated than Project 5

6. 6 Official Goals  Discuss the difference between standard programming styles and event-driven programming  Show the difference between structuring servers using processes and application-level multiplexing  Discuss the benefits and drawbacks of each approach

7. 7 What Is An Event?  Some kind of notification  Interrupts  Signals  Polling (via poll/select/etc)  Callback (via function pointer)  Similarities?

8. 8 “Reactive” Environments  Windowing systems  Network-aware programs  Drivers of all sorts

9. 9 Traditional Environments  One thing going on at a time  Finish that thing, go on to next  Any step can block indefinitely  Resumption from blocking is simple – OS provided

10. 10 What Goes On In A Web Browser?  Drawing the current page  Inline images  Translating machine names to IP addresses  Launching connections  Sending requests  Getting piecemeal responses, drawing images  User clicking to next link

11. 11 Threads Versus Events  One stack versus many stacks  What happens on blocking operations  Parallelism  Shared variables  State

12. 12 Let’s Think Of States  How many possible states are there?  Take all pieces of information  Decide valid range for all pieces  Enumerate  Can we reduce states?  Some combinations invalid  Still, lots of states

13. 13 Really Reducing States  Take all major pieces of program  Add extra tags to state  What do tags look like?  Position  Count #

14. 14 State Examples  If-then-else  3 states: start, then-clause, else-clause  For loop  1 state + count  Why do we care?  Resuming at the right state

15. 15 Remember This Diagram? Read File Send Data Accept Conn Read Request Find File Send Header end

16. 16 Structure of Event-Driven Programs  Lots of state machines  Maintaining information about each one  Some way of moving through states  As few restrictions as possible on timing

17. 17 The Real Structure While (1) Get event Dispatch event  Or, while loop in library  Event handlers in main program

18. 18 Delays  Are delays possible?  Interrupt handlers – generally not  Otherwise? Depends on event rate  How to avoid delays?  More events – asynchronous operations  What happens if no async support?  Fake it

19. 19 Select( ) System Call int select(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, struct timeval *timeout); FD_SET(fd, &fdset); FD_CLR(fd, &fdset); FD_ISSET(fd, &fdset); FD_ZERO(&fdset);

20. 20

21. 21 Select Description DESCRIPTION Select() examines the I/O descriptor sets whose addresses are passed in readfds, writefds, and exceptfds to see if some of their descriptors are ready for reading, are ready for writing, or have an exceptional condition pending, respectively. The only exceptional condition detectable is out-of-band data received on a socket. The first nfds descriptors are checked in each set; i.e., the descriptors from 0 through nfds-1 in the descriptor sets are examined. On return, select() replaces the given descriptor sets with subsets consisting of those descriptors that are ready for the requested operation. Select() returns the total number of ready descriptors in all the sets. The descriptor sets are stored as bit fields in arrays of integers. The following macros are provided for manipulating such descriptor sets: FD_ZERO(&fdset) initializes a descriptor set fdset to the null set. FD_SET(fd, &fdset) includes a particular descriptor fd in fdset. FD_CLR(fd, &fdset) removes fd from fdset. FD_ISSET(fd, &fdset) is non- zero if fd is a member of fdset, zero otherwise. The behavior of these macros is undefined if a descriptor value is less than zero or greater than or equal to FD_SETSIZE, which is normally at least equal to the maximum number of descriptors supported by the system.

22. 22 Blocking Steps Read File Send Data Accept Conn Read Request Find File Send Header end Network Blocking Disk Blocking

23. 23 Overcoming Disk Blocking States Read File Send Data Accept Conn Read Request Find File Send Header end Helper

24. 24 New Architecture - AMPED Helpers are threads or processes Read File Send Data Accept Conn Read Request Find File Send Header Event Dispatcher Asymmetric Multiple Process Event Driven Helper 1Helper 2Helper N

25. 25 Caches in Flash Web Server Read File Send Data Accept Conn Read Request Find File Send Header end Pathname Translation Cache Helper Mapped File Cache Response Header Cache