Bound thread scheduling

The semantics of bound thread scheduling differs considerably from that for multiplexed threads.

• Bound threads are permanently attached to their LWPs; consequently, they are exempt from that level of scheduling by the Threads Library.

• The Threads Library supports the concept of ``scheduling policy'' as well as ``priority level'' for bound threads. When the programmer specifies these characteristics, the Threads Library applies them to the LWP holding the thread. See priocntl(S) for more information.

  These characteristics can be modified with the thr_setscheduler(S) function.

• The available scheduling policies for bound threads are:

  
  SCHED_TS or SCHED_OTHER

  The two values are synonymous. The bound thread is run by an LWP of the kernel time-sharing scheduling class.

  ---

  NOTE: Technically, multiplexed threads are also categorized as having the SCHED_TS policy even though they are not necessarily run by LWPs in the kernel time sharing class. The Threads Library algorithm for scheduling multiplexed threads (round robin) bears a closer resemblance to the kernel's fixed priority class than the kernel's time sharing class.

  ---

  
  SCHED_FIFO or SCHED_RR

  The thread will be run on an LWP of the fixed-priority scheduling class. SCHED_FIFO means that the LWP will have an infinite time quantum (not preempted) whereas SCHED_RR (round-robin) uses a fixed priority with a finite time slice.

  The SCHED_FIFO and SCHED_RR policies can be used only by bound threads.

  ---

  NOTE: Appropriate privilege is required to set the policy of a thread to SCHED_FIFO or SCHED_RR. See priocntl(S).

  ---

• When an executing bound thread calls thr_yield(S), it deliberately surrenders the processor to another LWP if one is available to run on that processor.

A bound thread with real-time constraints can further improve response time by using processor_bind(S) to bind its LWP to a processor. It can use _lwp_self(S) to find the ID of the LWP to which it is bound, and pass that as an argument to processor_bind.