Threads

Threads

1. Benefits

   • Responsiveness
     • User interaction can be run in different threads
   • Resource Sharing
     • Threads of same process share address space
       • Used shared variables to communicate (more efficient than message passing)
   • Economy
     • Smaller memory footprint
   • Scalability
     • Make use of parallel processing
   • Reduce progamming complexity

2. Concurrent and parallel execution

3. Definition

   • Thread is an execution state of a process

4. User threads

   • Thread management done by user-level thread library
     • POSIX
     • WIN32
     • Java

5. Threading modes

   • Kernel may not support multi-threaded processes
     • Can still implement threading in user threads
   • Threading models for mapping user threads to kernel threads
     • Many-to-one
       • Many user-threads mapped to single kernel thread
         • Useful if kernel doesn't support threads
         • Problem if one user thread calls blocking kernel function
           • Could block all other threads (whole process)
             • Solution: replace blocking calls with non-blocking calls and schedule threads in user space
               • Shift kernel work to user space
     • One-to-one
       • Each user-level threads maps to one kernel thread
         • Context switch overhead for switching between threads
           • Overhead not as big as process switching
             • Switching less data - page tables / memory shared between threads
     • Many-to-many
       • Many user threads mapped to many kernel threads
       • Two-level model
         • Can bind user thread directly to kernel thread

6. Thread cancellation

   • Asynchronous cancellation
     • Terminates thread immediately - useful as last resort
   • Deferred cancellation
     • Target thread allowed to periodically check if it should be cancelled
       • Cleaner - thread can perform clean up (closing files)

7. Signal handling

   • Used to notify processes that a particular event has occurred
   • Signal handler used to process signals
     • Signal generated by event
       • Signal delivered to a process
         • Signal handled by some function
     • More complex in multi-threaded processes
       • Do you act on processes or individual threads?
       • Can deliver signal to thread to which signal applies
       • Can deliver signal to every thread in the process (kill signal)
       • Can deliver signal to certain threads
       • Can assign specific thread to receive all signals for the process

8. Thread pools

   • Waste a lot of processing by creating / destroying threads
     • Solution: Pre-create a number of threads in a pool - threads await work
       • If no threads are free then block
       • Advantages
         • Slightly faster to service new request using an existing thread (reduces overhead)
         • Allows number of threads to be bound to size of the pool - ensures some level of service
           • Control degree on concurrency
             • Guarantee throughput
           • Eliminates thrashing