Threading by Appointment

We propose a concurrent programming model called threading-by-appointment (TAP). Unlike traditional threads, TAP threads can only communicate with system components or other threads by appointment. For example, a TAP thread cannot simply try to access some shared resource. Instead, a TAP thread must make an appointment with the shared resource in advance. Only at the time of the appointment, the thread can actually access the shared resource. The desired start time and duration of an appointment may either be provided by the thread or computed by the TAP runtime system. The duration of an appointment must be finite. Instantaneous appointments are possible but only allow a single atomic operation per appointment. TAP threads cannot deadlock because the duration of appointments is finite. Race conditions may occur for making appointments but can be avoided by making appointments independently of the system’s progress. Threading-byappointment makes the process of determining the time instants for system interaction explicit and may therefore help to increase the determinism of concurrent applications as well as the efficient use of resources. Interesting questions arise such as how to schedule appointments as opposed to how to multiplex resources such as the CPU, and how to define structured programming elements that support threading-by-appointment.