Time Services And Scheduling Mechanisms

Time Services: Obviously, good time services are essential to real-time applications. To provide these services and support its own activities, every system has at least one clock device. We have been calling it the system clock. In a system that supports Real-Time POSIX, this clock is called CLOCK REALTIME.

Clocks and Time. Again, a clock device contains a counter, a timer queue, and an interrupt handler. The counter is triggered to increment monotonically by a precise periodic sequence of pulses. At any time, the content of the counter gives a representation of the current time. We have been calling this counter a (hardware) clock and the triggering edge of each pulse a clock tick. The timer queue contains the pending expiration times of timers bound to the clock.

A system may have more than one clock device and uses them for different purposes. For example, a real-time monitor system may use a clock device to trigger the sampling and digitization of sensor readings and initialize the input of sensor data. A digital control system may use one or more clock devices to time the control-law computations of a rate group or commands to embedded plants.

Resolution. The resolution of a clock is the granularity of time provided by the clock. Today’s technology makes it possible to have hardware clocks with resolution on an order of nanoseconds. However, the clock resolution available to applications is usually orders of magnitude coarser, on an order of hundreds of microseconds or milliseconds.

High Resolution. A way to provide a finer resolution than hundreds of microseconds and more accurate time to an application is to map a hardware clock into the address space of the application. Then, the application can read the clock directly. On a Pentium processor, a user thread can read the Pentium time stamp counter.7 This counter starts at zero when the system is powered up and increments each processor cycle. At today’s processor speed, this means that counter increments once a few nanoseconds. By computing from the cycle count provided by the counter, an application can get more precise and higher-resolution time than that provided by the operating system.

Timers and Timer Functions. Most operating systems (including all Real-Time POSIX compliant systems, Windows NT, and Solaris) allow a thread or process to have its own timers.8 Specifically, by calling the create timer function, a thread (or process) can create a per thread (or per process) timer and, in a system containing multiple clocks, bind the timer to a specified clock. Associated with each timer is a data structure which the kernel creates in response to the create timer call. Among the information kept in the data structure is the expiration time of the timer. The data structure may also have a pointer to a handler; the handler is a routine that the calling thread wants to be execute when a timer event occurs. A thread destroys its own timer by calling the destroy timer function.

Asynchronous Timer Functions. As an example, we look at the set watchdog timer function wdStart( ) provided by VxWorks [Wind]. A watchdog (or alarm) timer is a oneshot timer that supports relative time. After a timer is created, a thread can use it by calling the wdStart( ) function and specifying as arguments the timer ID, the delay to the expiration time, the function to call at the expiration time, and an argument to be passed to the specified function. A thread can cancel a timer before its expiration by calling wdCancel( ).

Signal is the notification mechanism used by Real-Time POSIX timer function timer settime( ). When a thread calls the timer create( ) function to request the creation of a timer, it specifies the clock to which the timer is to be bound, as well as the type of signal to be delivered whenever the timer expires. If the type of signal is not specified and the clock to which the timer is bound is CLOCK REALTIME, the system will deliver a SIGARLM (alarm clock expired) signal by default.
After creating a timer, a thread can set it by calling timer settime( ). The parameters of this function include

1. the ID of the timer to be set;
2. a flag indicating whether the new expiration time is relative or absolute;
3. the new timer setting, which gives the delay to the expiration time or the absolute value of the expiration time, depending on value of the flag; and
4. the period between subsequent timer expiration times.9

Again, when the timer expires, the system delivers a signal of the specified type to the calling thread.

Synchronous Timer Functions. The set-timer functions in the previous examples are asynchronous. After being set, the timer counts down while the calling thread continues to execute. Consequently, a thread can set multiple timers to alarm at different rates or times. In contrast, after calling a synchronous timer function, the calling thread is suspended. As an example, we look at the timer sleep( ) function provided by Real-TimeMach. The function causes the calling thread to be suspended either until the specified absolute time or for the specified time interval.When the specified time is relative, the timer function is similar to the Real-Time POSIX nanosleep(t); the parameter t specifies the length of the interval the calling thread sleeps.

Timer Resolution. We measure the quality of timers by their resolution and accuracy. The term resolution of a timer usually means the granularity of the absolute time or the time interval specified as an argument of a timer function. We call this granularity the nominal timer resolution. If the nominal timer resolution is x microseconds, then the operating system will not mistake two timer events set x microseconds apart as a single timer event. However, this does not mean that the granularity of time measured by the timer, as seen by application threads, is this fine.

Periodic Timer Interrupts. To illustrate, we recall that in most operating systems, the kernel checks for the occurrences of timer events and handles the events only at time-service interrupts and these interrupts occur periodically. Now let us suppose that the nominal timer resolution is 10 microseconds and the period of time-service interrupts is 5 milliseconds. A threads sets a timer to expire twice 10 microseconds apart and reads the current time at the occurrence of each timer event. Suppose that the first timer event occurs at 5 microseconds before an interrupt and the second one at 5 microseconds after the interrupt. The kernel handles
the first one at the interrupt and the second at the next time-service interrupt. The time values read by the thread are approximately 5 milliseconds apart.

One-Shot Timer Interrupts. Alternatively, some operating systems (e.g., QNX) program the clock device to raise an interrupt at each timer expiration time. In other words, the clock interrupts in the one-shot mode. As a part of timer interrupt service, the kernel finds the next timer expiration time and sets the clock to interrupt at that time. Thus, the kernel carries out the requested action as soon as a timer expires.With this implementation, the actual timer resolution is limited only by the amount of time the kernel takes to set and service the clock device and to process each timer event, since the kernel cannot respond to timer events more frequently than once per this amount of time. This time is in order of microseconds even on today’s fast processors.

Timer Accuracy. By timer error, we mean the difference between the absolute time (or time interval) specified by the calling thread and the actual time at which the specified action starts. Timer error depends on three factors. The first is the frequency at which timer expirations are checked. This factor is the period of time-service interrupts in most operating systems since they check for timer expirations periodically.