Measuring the Latency Time ofReal-Time Unix-like Operating SystemsNewton FallerTR-92-037June 1992AbstractWith the advent of continuous-media applications, real-time operating systems, once confined toprocess control and other specialized applications, are coming to the desktop. The popularity ofUNIX made this operating system the first choice for use with such real-time desktop applica-tions. However, since UNIX kernel does not provide real-time responsiveness, some softwaredevelopers have been trying to adapt it to respond to this new requirements, while others havebeen proposing its total redesign. Though the evaluation of the performance of a real-timeoperating system depends on many factors, a predictable small latency time in responding toexternal events is always essential. In this paper, after a discussion about the probable sources oflatency, it is presented a method for collecting information about context-switching andinterrupt-acknowledge times in UNIX-like operating systems without requiring external measur-ing tools. It is also proposed, a form of presentation of these data aimed at facilitating the com-parison with previously collected data obtained from the same or from other systems. The paperis illustrated with actual results obtained by the application of the method to TROPIX, a real-timeUNIX-like operating system, running on a Motorola 68010-based computer. The impact of ker-nel preemption and some practical measurement interference considerations due to dynamicmemory refresh, DMA operation and disk multiblock access are also discussed.____________________________________This research was supported by CNPq - Conselho Nacional de Desenvolvimento Cient´ifico e Tecnolo´gico(Brasil) and also by the National Science Foundation and the Defense Advanced Research Projects Agency(DARPA) under Cooperative Agreement NCR-8919038 with the Corporation for National Research Initia-tives, by AT&T Bell Laboratories, Hitachi, Ltd., Hitachi America, Ltd., Pacific Bell, the University of Cali-fornia under a MICRO grant, and the International Computer Science Institute. The views and conclusionscontained in this document are those of the authors, and should not be interpreted as representing officialpolicies, either expressed or implied, of the U.S. Government or any of the sponsoring organizations.Measuring the Latency Time of Real-Time Unix-like Operating SystemsNewton FallerInternational Computer Science InstituteBerkeley, CA, [email protected] Federal do Rio de JaneiroRio de Janeiro, RJ, [email protected] or [email protected] the advent of continuous-media applications, real-time operating systems, onceconfined to process control and other specialized applications, are coming to the desktop.The popularity of UNIX made this operating system the first choice for use with suchreal-time desktop applications. However, since UNIX kernel does not provide real-timeresponsiveness, some software developers have been trying to adapt it to respond to thisnew requirements, while others have been proposing its total redesign. Though theevaluation of the performance of a real-time operating system depends on many factors, apredictable small latency time in responding to external events is always essential. Inthis paper, after a discussion about the probable sources of latency, it is presented amethod for collecting information about context-switching and interrupt-acknowledgetimes in UNIX-like operating systems without requiring external measuring tools. It isalso proposed, a form of presentation of these data aimed at facilitating the comparisonwith previously collected data obtained from the same or from other systems. The paperis illustrated with actual results obtained by the application of the method to TROPIX, areal-time UNIX-like operating system, running on a Motorola 68010-based computer.The impact of kernel preemption and some practical measurement interference con-siderations due to dynamic memory refresh, DMA operation and disk multiblock accessare also discussed.1 - IntroductionA real-time operating system may be defined as a system such that, for a definedmaximum workload, it is always capable of acknowledging a specific set of asynchro-nous external events and execute all the processing required to respond to them within afinite and predictable amount of time compatible with the application. Operating systemlatency time is the real elapsed time between the occurrence of a specific external eventand the appropriate system response to it.Complex computer-based real-time systems have been controlled by real-timeoperating systems. These systems, once confined to process control and other specialapplications, are replacing time-sharing operating systems due to the stringent require-ments of continuous-media applications. In fact, in the near future, it is expect that theywill become commonplace in desktop workstations.Real-time systems have been divided in hard and soft [StRa88]. They basicallydiffer in the way scheduling is done. While in hard real-time systems a set of tasks withtheir deadlines are specified, in soft real-time systems the highest priority task should beexecuted as fast as possible. In this paper we will be concerned with soft real-time sys-tems only.With the UNIX system being the most popular operating system for workstations, itis easy to understand that many software developers have been trying to fit UNIX to runappropriately with real-time applications. This has been done through adaptation of itskernel, for instance [KhSZ92] [Fual91], or even, sometimes, through the design of a newkernel [FaSa90].There are many ways of estimating the performance of a real-time operating sys-tems. In [Fual91], for instance, four metrics are described: the Rhealstone metric, theProcess Dispatch Latency Time, the Tri-Dimensional Measure, and the Real/StoneBenchmark. In all of them, latency time has always been one of the main factors.In this paper we suggest a method for obtaining interrupt response and contextswitching latency times in Unix-like operating systems requiring no external measuringtools. It is assumed, however, that the hardware has a real-time clock. The operating sys-tem is not required to be specifically designed for real-time applications but it shouldhave some similarity to the UNIX system. A description of the procedures to be exe-cuted and the software instrumentation required for the measurements to be