Fixed-task-priority (FTP) scheduling and earliest-deadline-first (EDF) scheduling policies are alike in fixing the priority of each job of a task at the time the job is released. This common feature of FTP and EDF scheduling permits a unified analysis of scheduling failures, to derive new sufficient tests for meeting all deadlines of a set of independent sporadic tasks under a global preemptive FTP or EDF scheduling policy. The performance of the new tests has been evaluated in comparison to prior schedulability tests, by simulation. The new tests are able to verify some schedulable task systems that could not be verified by prior tests, but also fail to verify some systems that can be verified by prior analysis techniques. The biggest gain appears to be for fixed-task-priority scheduling, especially with post-period deadlines.

Now that multicore microprocessors have become a commodity, it is natural to think about employing them in all kinds of computing, including high-reliability embedded real-time systems. Appealing aspects of this development include the ability to process more instructions per second and more instructions per watt. However, not all problems are amenable to parallel decomposition, and for those that are, designing a correct scalable solution can be difficult. If there are deadlines or other hard timing constraints the difficulty becomes much greater.This paper reviews some of what is known about multiprocessor scheduling of task systems with deadlines, including recent advances in the analysis of arbitrary sporadic task systems under fixed-priority and earliest-deadline first scheduling polices. It also examines critically the foundations of these theoretical results, including assumptions about task independence and worst-case execution time estimates, with a view toward their practical applicability.

The specification of the sporadic server real-time scheduling policy in the IEEE POSIX standard is defective, and needs to be corrected. Via experiments using a POSIX sporadic server implementation under Linux, as well as simulations, we have shown and confirmed previously unreported defects. We propose and demonstrate a corrected sporadic server formulation that eliminates these defects without changes to the syntax of the API or any significant increase in implementation complexity.

A scheduling policy or a schedulability test is defined to be sustainable with respect to a particular workload model if any task system represented in that model that is determined to be schedulable remains so if it behaves "better" than mandated by its specifications. We investigate the sustainability properties of global scheduling algorithms when applied to systems represented using the sporadic task model. We show that Fixed-Priority (FP) scheduling of sporadic task sets is sustainable under a variety of scheduling parameter relaxations, including decreased execution requirements, later arrivals, and deadline relaxations. It follows that all sufficient tests of global FP schedulability are sustainable for sporadic task systems. We show that the Earliest Deadline First (EDF) and Earliest-Deadline with Zero Laxity scheduling policies are sustainable with respect to decreased execution requirements and later arrivals. We also introduce a notion of self-sustainability, and show that many widely-used EDF schedulability tests are not self-sustainable but one is.

Many contemporary disk drives have built-in queues and schedulers. These features can improve I/O performance, by offloading work from the system's main processor, avoiding disk idle time, and taking advantage of vendor-specific disk characteristics. At the same time, they pose challenges for scheduling requests that have real-time requirements, since the operating system has less visibility and control over service times. While it may be possible for an operating system to obtain more predictable real-time performance by bypassing the on-disk queue and scheduler, the diversity and continuing evolution of disk drives make it difficult to extract the necessary detailed timing characteristics of a specific disk, and to generalize that approach to all hard drives. This paper demonstrates three techniques we developed in the Linux operating system to bound real-time request response times for disks with internal queues and schedulers. The first technique is to use the disk's built-in starvation prevention scheme. The second is to prevent requests from being sent to the disk when real-time requests are waiting to be served. The third is to limit the length of the on-disk queue in addition to the second technique. Our results show the ability to guarantee a wide range of desired response times while still allowing the disk to perform scheduling optimizations. These techniques can be generalized to disks from different vendors.

A schedulability test is derived for the global earliest deadline zero laxity (EDZL) scheduling algorithm on a platform with multiple identical processors. The test is sufficient, but not necessary, to guarantee that a system of independent sporadic tasks with arbitrary deadlines will be successfully scheduled, with no missed deadlines, by the multiprocessor EDZL algorithm. Global EDZL is known to be at least as effective as global earliest-deadline-first (EDF) in scheduling task sets to meet deadlines. It is shown, by testing on large numbers of pseudo-randomly generated task sets, that the combination of EDZL and the new schedulability test is able to guarantee that far more task sets meet deadlines than the combination of EDF and known EDF schedulability tests.

This report describes a necessary and sufficient test for the schedulability of a set of sporadic hard-deadline tasks on a multiprocessor platform, using any of a variety of scheduling policies including global fixed task-priority and earliest-deadline-first (EDF). The contribution is to establish an upper bound on the computational complexity of this problem, for which no algorithm has yet been described. The compute time and storage complexity of the algorithm, which performs an exhaustive search of a very large state space, make it practical only for tasks sets with very small integer periods. However, as a research tool, it can provide a clearer picture than has been previously available of the real success rates of global preemptive priority scheduling policies and low-complexity sufficient tests of schedulability.

The Office of Graduate Studies has verified and approved the above named committee members. iii To my father Weiming Qian and my mother Jianping Cai iv ACKNOWLEDGEMENTS I would like to thank my advisor, Dr. Andy Wang, for his guidance and support.

API extensions and performance improvements to the Linux operating system now enable it to serve as a platform for a range of embedded real-time applications, using fixed-priority preemptive scheduling. Powerful techniques exist for analytical verification of application timing constraints under this scheduling model. However, when the application is layered over an operating system the operating system must be included in the analysis. In particular, the computational workloads due to device drivers and other internal components of the operating system, and the ways they are scheduled, need to match abstract workload models and scheduling polices that are amenable to analysis. This paper assesses the degree to which the effects of device drivers in Linux can now be modeled adequately to admit fixed-priority preemptive schedulability analysis, and what remains to be done to reach that goal.

The multiprocessor edf scheduling of sporadic task systems is studied. A new sufficient schedulability test is presented and proved correct. It is shown that this test generalizes the previously-known exact uniprocessor edf -schedulability test, and that it offers non-trivial quantitative guarantees (including a resource augmentation bound) on multiprocessors.

Device drivers are integral components of operating systems. The computational workloads imposed by device drivers tend to be aperiodic and unpredictable because they are triggered in response to events that occur in the device, and may arbitrarily block or preempt other time-critical tasks. This characteristic poses significant challenges in real-time systems, where schedulability analysis is essential to guarantee system-wide timing constraints. At the same time, device driver workloads cannot be ignored. Demand-based schedulability analysis is a technique that has been successful in validating the timing constraints in both single and multiprocessor systems. In this paper we present two approaches to demand-based schedulability analysis of systems that include device drivers. First, we derive load-bound functions using empirical measurement techniques. Second, we modify the scheduling of network device driver tasks in Linux to implement an algorithm for which a load-bound function can be derived analytically. We demonstrate the practicality of our approach through detailed experiments with a network device under Linux. Our results show that, even though the network device driver does not conform to conventional periodic or sporadic task models, it can be successfully modeled using hyperbolic load-bound functions that are fitted to empirical performance measurements

A new feasibility test for preemptive scheduling of periodic or sporadic real-time tasks on a single-queue m-server system allows for arbitrary fixed task priorities and arbitrary deadlines. For the special case when deadline equals period and priorities are rate monotonic, any set of tasks with maximum individual task utilization umax and minimum individual task utilization umin is feasible if the total utilization does not exceed \(m(1-u_{\rm max})/2 + u_{\rm min}\).

The load parameter of a sporadic task system is defined to be the largest possible cumulative execution requirement that can be generated by jobs of the task system over any time interval, normalized by the length of the interval. This parameter is known to play a very important role in the uniprocessor feasibility analysis of sporadic task systems. In this paper, it is shown that the load of a sporadic task system may be used as an accurate indicator of its feasibility upon preemptive multiprocessors as well. Exact algorithms, and approximate ones that can be guaranteed to be accurate to within an arbitrary additive error > 0, for computing a task system's load are presented and proven correct. The performance of these algorithms is evaluated by simulation over randomly generated task systems

This report describes a necessary condition for feasibility of scheduling a set of sporadic hard-deadline tasks on identical multiprocessor platforms, which is also a sufficient condition if there is only a single processor.The key contribution is the characterization of the maximum, over all time intervals of a given length, of the amount of computation that must be completed to meet all deadlines, and a method of computing this function efficiently to any desired degree of accuracy.Empirical data are provided to verify that the new infeasibility test can be computed efficiently and is an improvement over previously known checks for infeasibility.

A polynomial-time algorithm is presented for partitioning a collection of sporadic tasks among the processors of an identical multiprocessor platform with static-priority scheduling on each individual processor. Since the partitioning problem is easily seen to be NP-hard in the strong sense, this algorithm is not optimal. A quantitative characterization of its worst-case performance is provided in terms of sufficient conditions and resource augmentation approximation bounds. The partitioning algorithm is also evaluated over randomly generated task systems

This paper describes the application-level interfaces for POSIX signal handling supported by the GNu Ada'95 Runtime Library (GNARL), and their implementation. GNARL maps some signals to exceptions, and allows others to be handled via protected procedures, task entries, or the POSIX.5b synchronous signal-awaiting operations. Measurements of the relative performance of the latter three mechanisms are also reported.