On real-time and non real-time distributed computing

Abstract

Abstract. In this paper, taking an algorithmic viewpoint, we explore the differences existing between the class of non real-time computing problems (R~) versus the class of real-time computing problems (~). We show how a problem in class RN can be transformed into its counterpart in class ~. Claims of real-time behavior made for solutions to problems in class ~ are examined. Ah example of a distributed computingproblem arising m class Is studmd, along with its solutmn. It is shown why off-line strategies or scheduling algorithms that are not driven by real-time/timeliness requirements ~ are incorrect for class ~. Finally, a unified approach to conceiving and measuring the efficiency of solutions to problems in classes R~ and ~ is proposed and illustrated with a few examples. INTRODUCTIONOver the last 20 years, the distributed algorithms community has spent considerable effort solving problems in the areas of serializable or linearizable concurrent computing. These problems belong to a class denoted R~, the class of non reMtime computing problems. Separately, over the last 30 years, the real-time algorithms community has spent considerable effort solving scheduling problems in the area of centralized computing, considering preemptable and non-preemptable resources. Only recently has distributed computing received some attention. These problems belong to a class denoted ~, the class of real-time computing problems.In this paper, taking an algorithmic viewpoint, we embark on exploring the differences between both classes. The scope of this paper is restricted to that of deterministic algorithms. Besides intellectual interest, investigation of these issues is expected to bring the two communities closer, via the clarification of a few essential concepts. In particular, we have observed that the qualifiers "realtime" and "distributed" may not carry the same meaning in both communities. Broadly speaking, with few exceptions, the distributed algorithms community does not consider problem specifications that include real-time requirements. Conversely, the real-time algorithms community often fails to understand what is implied with considering a distributed computing problem.The remainder of this paper is organized as follows. In section 2, we introduce the distinctive attributes of class ~. In section 3, we examine claims of real-time

Failure Detection with Booting in Partially Synchronous Systems

“…Note carefully that the detection time is always as good as provided by the underlying system/network, i.e., the FD timing properties "emerge" naturally from the system/network capabilities [1,26]. Moreover, if the system/network load returns to expected behavior, our algorithm is still as exact and fast as predicted, while algorithms that adapt their timeouts would have larger detection latencies then.…”

“…The coverage of such time-free solutions is hence necessarily higher than that of a solution involving some timing assumptions. The apparent contradiction between time-free algorithms and timeliness properties can be resolved by following the design immersion principle, which was introduced in [26] and referred to as the late binding principle in [1]. Design immersion permits to consider time-free algorithms for implementing system or application level services in real-time systems, by enforcing that timing-related conditions (like "delay for time X") are expressed as time-free logical conditions (like "delay for x round-trips" or "delay for x events").…”

“…Nevertheless, following the design immersion principle [1,26], a bound on the detection time can be computed when our algorithm is immersed in some real system. By conducting a worst-case schedulability analysis of our FD, a bound for τ + can be established.…”

Fast asynchronous uniform consensus in real-time distributed systems

“…The idea of deferring the consideration of some MðSÞ until after having devised and proved some design Á in some MðÁÞ has been stated first in [16], echoed in [9] and [12], and detailed in [17], under the name of "design immersion" (in a computational model). This is equivalent to the concept of "late binding" (of a design to some computational model), a well-known concept in the field of programming languages.…”

Asynchrony and real-time dependable computing

“…The idea of deferring the consideration of some M(S) until after having devised and proved some design in some M( ) less "restrictive" than M(S) has been stated first in [12], echoed in [9] and [11], and detailed in [13], under the name of "design immersion" (in a computational model). This is equivalent to the concept of late binding (of a design to some computational model), a well known concept in the areas of programming languages and compilation.…”