This thesis studies the problem of experimentally evaluating the scaling behaviors of existing multicore real-time task scheduling algorithms on large-scale multicore platforms. As chip manufacturers rapidly increase the core count of processors, it becomes imperative that multicore real-time scheduling algorithms keep pace. Thus, it must be determined if existing algorithms can scale to these new high core-count platforms. Significant research exists on the theoretical performance of multicore real-time scheduling algorithms, but the vast majority of this research ignores the effects of scalability. It has been demonstrated that multicore real-time scheduling algorithms are feasible for small core-count systems (e.g. 8-core or less), but thus far the majority of the algorithmic research has never been tested on high core-count systems (e.g. 48-core or more). We present an experimental analysis of the scalability of 16 multicore real-time scheduling algorithms. These algorithms include global, clustered, and partitioned algorithms. We cover a broad range of algorithms, including deadline-based and utility accrual scheduling algorithms. These algorithms are compared under metrics including schedulability, tardiness, deadline satisfaction ratio, and utility accrual ratio. We consider multicore platforms ranging from 8 to 48 cores. The algorithms are implemented in a real-time Linux kernel we create called ChronOS. ChronOS is based on the Linux kernel's PREEMPT RT patch, which provides the underlying operating system kernel with real-time capabilities such as full kernel preemptibility and priority inheritance for kernel locking primitives. ChronOS extends these capabilities with a flexible, scalable real-time scheduling framework. Our study shows that it is possible to implement global fixed and dynamic priority and simple global utility accrual real-time scheduling algorithms which will scale to large-scale multicore platforms. Interestingly, and in contrast to the conclusion of prior research, our results reveal that some global scheduling algorithms (e.g. G-NP-EDF) is actually scalable on large core counts (e.g. 48). In our implementation, scalability is restricted by lock contention over the global schedule and the cost of inter-processor communication, rather than the global task queue implementation. We also demonstrate that certain classes of utility accrual algorithms such as the GUA class are inherently not scalable. We show that algorithms implemented with scalability as a first-order implementation goal are able to provide real-time guarantees on our 48-core platform. / Master of Science

Utility Accrual (UA) scheduling is a method of dynamic real-time scheduling that is designed to respond to overload conditions by producing a feasible schedule that heuristically maximizes a pre-defined metric of utility. Whereas utility accrual schedulers have traditionally focused on CPU overload, this dissertation explores memory overload conditions during which the aggregate memory demand exceeds a system's available memory bandwidth. Real-time systems are typically implemented in C or other languages that use explicit dynamic memory management. Taking advantage of modern type-safe languages, such as Java, necessitates the use of garbage collection (GC). The timeliness requirements of real-time systems, however, impose specific demands on the garbage collector. Garbage collection introduces a significant source of unpredictability in the execution timeline of a task because it unexpectedly interjects pauses of arbitrary length, at arbitrary points in time, with an arbitrary frequency. To construct a feasible schedule, a real-time scheduler must have the ability to predict the collector's activities and plan for them accordingly. We have devised CADUS (Collector-Aware Dynamic Utility Scheduler), a utility accrual algorithm that tightly links CPU scheduling with the memory requirements -and the corresponding garbage collection activities - of real-time tasks. By constructing and storing memory time allocation profiles, we address the problem of GC activation strategy. We estimate GC latency by using a real-time collector and modeling its behavior. We project GC frequency by planning, at schedule construction time, the memory bandwidth available to the collector. CADUS can point the collector's activities to any specific task in the system. The runtime system provides this ability by maintaining separate logical heaps for all tasks. We demonstrate the viability of CADUS through extensive simulation studies. We evaluated the behavior of CADUS under a wide range of CPU and memory load conditions and utility distributions. We compared its performance against an existing GC-unaware UA scheduler and found that CADUS consistently outperformed its GC-unaware counterpart. We investigated and identified the reasons for the superior performance of CADUS and quantified our results. Most significantly, we found that in an overloaded dynamic soft real-time system, a scheduler's preemption decisions have a highly significant impact on GC latency. A dynamic real-time scheduler therefore must predict the impact of its preemption decisions on GC latency in order to construct time-feasible schedules. / Ph. D.

Scheduling decisions in soft real-time environments are based on a utility function. The goal of such schedulers is to use a best-effort approach to maximize the utility function and ensure graceful degradation at overloads. Utility Accrual (UA) schedulers use heuristics to maximize the accrued utility. Heuristic-based scheduling do not always yield the optimal schedule even if there exists one because they do not explore the entire search space of task orderings. In distributed systems, local UA schedulers use the same heuristics along with deadline decomposition for task segments. At present, there has been no evaluation and analysis of the degree to which these polynomial-time, heuristic algorithms succeed in maximizing the total utility accrued. We implemented a preemptive, off-line static scheduling algorithm that performs an exhaustive search of all the possible task orderings to yield the optimal schedules. We simulated two important online dynamic UA schedulers, DASA-ND and LBESA for different system loads, task models, utility and load distribution patterns, and compared their performance with their corresponding optimal schedules. Our experimental analysis indicates that for most scenarios, both DASA-ND and LBESA create optimal schedules. When task utilities are equal or form a geometric sequence with an order of magnitude difference in their utility values, UA schedulers show more than 90% probability of being optimal for single-node workloads. Even though deadline decomposition substantially improves the optimality of both DASA-ND and LBESA under different scenarios for distributed workloads, it can adversely affect the scheduling decisions for some task sets we considered. / Master of Science

We present a utility accrual real-time scheduling algorithm called CIC-VCUA, for tasks whose execution times are functions of their starting times. We model such variable execution times employing variable cost functions (or VCFs). The algorithm considers application activities that are subject to time/utility function time constraints (or TUFs), execution times described using VCFs, and concurrent, mutually exclusive sharing of non-CPU resources. We consider the multi-criteria scheduling objective of (1) assuring that the maximum interval between any two consecutive, successful completions of jobs of a task must not exceed a specified upper bound, and (2) maximizing the system's total accrued utility, while satisfying mutual exclusion resource constraints. Since the scheduling problem is intractable, CIC-VCUA statically computes worst-case sojourn times of tasks, selects tasks for execution based on their potential utility density, and completes them at specific times, in polynomial-time. We establish that CIC-VCUA achieves optimal timeliness during under-loads. Further, we identify the conditions under which timeliness assurances hold. Our simulation experiments illustrate CIC-VCUA's effectiveness and superiority. / Master of Science

In this dissertation, we consider the problem of scheduling and synchronization of distributable real-time threads --- Real-Time CORBA's first-class abstraction for programming real-time, multi-node sequential behaviors. Distributable real-time threads can be scheduled, broadly, using two paradigms: node independent scheduling, in which nodes independently construct thread schedules, based on node-level decomposition of distributable thread (or DT) scheduling parameters, and collaborative scheduling, in which nodes collaborate to construct system-wide thread schedules, which may or may not involve scheduling parameter decomposition. While significant literature exists on node independent scheduling, little is known about collaborative scheduling and its concomitant tradeoffs. We design three collaborative scheduling algorithms, called ACUA, QBUA, and DQBUA. ACUA uses theory of consensus and QBUA uses theory of quorums for distributable thread schedule construction. DQBUA extends QBUA with lock-based, local and distributed concurrency control. The algorithms consider a model where distributable threads arrive arbitrarily, have time/utility function time constraints, access resources in an arbitrary way (e.g., arbitrary lock acquire/release order, arbitrary nestings), and are subject to arbitrary node crash failures and message losses. We analytically establish several properties of the algorithms including probabilistic end-to-end termination time satisfactions, timeliness optimality during underloads, bounded exception handling time, and correctness of the algorithms in partially synchronous systems. We implement distributable real-time threads in the Linux kernel as a first-class programming and scheduling abstraction. The resulting kernel, called ChronOS, provides application interfaces for creating and manipulating distributable threads, as well as kernel interfaces and mechanisms for scheduling them (using both independent and collaborative approaches). ChronOS also has failure detector mechanisms for detecting and recovering from distributable thread failures. We implement the proposed scheduling algorithms and their competitors in ChronOS and compare their behavior. Our studies reveal that the collaborative scheduling algorithms are superior to independent scheduling algorithms for certain thread sets, in particular, when thread sections have significantly varying execution time. This variability, especially if the variability is not consistent among the threads, may cause each node to make conflicting decisions in the absence of global information. We observe that collaborative schedulers outperform independent schedulers (e.g., EDF augmented with PIP) in terms of accrued utility by as much as 75%. We identify distributed dependencies as one of the major sources of overhead in collaborative scheduling. In particular, the cost of distributed lock-based concurrency control (e.g., lock management, distributed deadlock detection/resolution) can significantly reduce the problem space for which collaborative scheduling is beneficial. To mitigate this, we consider the use of software transactional memory (or STM), an optimistic, non-blocking synchronization alternative to lock-based concurrency control which has been extensively studied in non real-time contexts. We consider distributable real-time threads with STM concurrency control, and develop techniques for analyzing and bounding their end-to-end response times on distributed single-processor and distributed multiprocessor systems. We also develop contention management techniques, a key component of STM, which are driven by threads' real-time scheduling parameters, and establish their tradeoffs against non-real-time contention managers. / Ph. D.

This dissertation presents a class of utility accrual scheduling and synchronization algorithms for dynamic, single and multiprocessor real-time systems. Dynamic real-time systems operate in environments with run-time uncertainties including those on activity execution times and arrival behaviors. We consider the time/utility function (or TUF) timing model for specifying application time constraints, and the utility accrual (or UA) timeliness optimality criteria of satisfying lower bounds on accrued activity utility, and maximizing the total accrued utility. Efficient TUF/UA scheduling algorithms exist for single processors---e.g., the Resource-constrained Utility Accrual scheduling algorithm (RUA), and the Dependent Activity Scheduling Algorithm (DASA). However, they all use lock-based synchronization. To overcome shortcomings of lock-based (e.g., serialized object access, increased run-time overhead, deadlocks), we consider non-blocking synchronization including wait-free and lock-free synchronization. We present a buffer-optimal, scheduler-independent wait-free synchronization protocol (the first such), and develop wait-free versions of RUA and DASA. We also develop their lock-free versions, and upper bound their retries under the unimodal arbitrary arrival model. The tradeoff between wait-free, lock-free, and lock-based is fundamentally about their space and time costs. Wait-free sacrifices space efficiency in return for no additional time cost, as opposed to the blocking time of lock-based and the retry time of lock-free. We show that wait-free RUA/DASA outperform lock-based RUA/DASA when the object access times of both approaches are the same, e.g., when the shared data size is so large that the data copying process dominates the object access time of two approaches. We derive lower bounds on the maximum accrued utility that is possible with wait-free over lock-based. Further, we show that when maximum sojourn times under lock-free RUA/DASA is shorter than under lock-based, it is a necessary condition that the object access time of lock-free is shorter than that of lock-based. We also establish the maximum increase in activity utility that is possible under lock-free and lock-based. Multiprocessor TUF/UA scheduling has not been studied in the past. For step TUFs, periodic arrivals, and under-loads, we first present a non-quantum-based, optimal scheduling algorithm called Largest Local Remaining Execution time-tasks First (or LLREF) that yields the optimum total utility. We then develop another algorithm for non-step TUFs, arbitrary arrivals, and overloads, called the global Multiprocessor Utility Accrual scheduling algorithm (or gMUA). We show that gMUA lower bounds each activity's accrued utility, as well as the system-wide, total accrued utility. We consider lock-based, lock-free, and wait-free synchronization under LLREF and gMUA. We derive LLREF's and gMUA's minimum-required space cost for wait-free synchronization using our space-optimal wait-free algorithm, which also applies for multiprocessors. We also develop lock-free versions of LLREF and gMUA with bounded retries. While the tradeoff between wait-free LLREF/gMUA versus lock-based LLREF/gMUA is similar to that for the single processor case, that between lock-free LLREF/gMUA and lock-based LLREF/gMUA hinges on the cost of the lock-free retry, blocking time under lock-based, and the operating system overhead. / Ph. D.

This dissertation first presents an uniprocessor real-time scheduling algorithm called the Generic Benefit Scheduling algorithm (or GBS). GBS solves a previously open real-time scheduling problem: scheduling activities subject to arbitrarily shaped, time/utility function (TUF) time constraints and mutual exclusion resource constraints. A TUF specifies the utility of completing an application activity as an application- or situation-specific function of when that activity completes. GBS considers the scheduling objective of maximizing system-wide, total accrued utility, while respecting mutual exclusion constraints. Since this problem is NP-hard, GBS heuristically computes schedules in polynomial-time. The performance of the GBS algorithm is evaluated through simulation and through an implementation on a Portable Operating System Interface (POSIX)-compliant real-time operating system. The simulation studies and implementation measurements reveal that GBS performs close to, if not better than existing algorithms for the cases that they apply. Further, the results verify the effectiveness of GBS for its unique model. We also analytically establish timeliness and non-timeliness properties of GBS including bounds on activity utilities and mutual exclusion. GBS targets real-time systems that are subject to significant non-determinism inherent in their operating environments e.g., completely unknown activity arrivals. When system uncertainties can be stochastically characterized (e.g., stochastic activity arrivals and execution times), it is possible to provide stochastic assurances on timeliness behavior. The dissertation also presents algorithmic solutions to fundamental assurance problems in TUF-driven real-time systems, including stochastically satisfying individual, activity utility lower bounds and system-wide, total utility lower bounds. The algorithmic solutions include algorithms for processor bandwidth allocation and TUF scheduling. While bandwidth allocation algorithms allocate processor bandwidth share to activities to satisfy utility lower bounds, TUF scheduling algorithms schedule activities to maximize accrued utility. The algorithmic solutions and analysis are extended with a class of lock-free and lock-based resource access protocols to satisfy mutual exclusion constraints. We show that satisfying utility lower bounds with lock-based resource access protocols does not imply doing so with the lock-free scheme, and vice versa. Finally, the dissertation presents a rule-based framework for trading off assurance requirements on utility lower bound satisfaction. / Ph. D.

We consider the problem of scheduling real-time tasks on a multiprocessor system. Our primary focus is scheduling on multiprocessor systems where the total task utilization demand, U, is greater than m, the number of processors on a multiprocessor system---i.e., the total available processing capacity of the system. When U > m, the system is said to be overloaded; otherwise, the system is said to be underloaded. While significant literature exists on multiprocessor real-time scheduling during underloads, little is known about scheduling during overloads, in particular, in the presence of task dependencies---e.g., due to synchronization constraints. We consider real-time tasks that are subject to time/utility function (or TUF) time constraints, which allow task urgency to be expressed independently of task importance---e.g., the most urgent task being the least important. The urgency/importance decoupling allowed by TUFs is especially important during overloads, when not all tasks can be optimally completed. We consider the timeliness optimization objective of maximizing the total accrued utility and the number of deadlines satisfied during overloads, while ensuring task mutual exclusion constraints and freedom from deadlocks. This problem is NP-hard. We develop a class of polynomial-time heuristic algorithms, called the Global Utility Accrual (or GUA) class of algorithms. The algorithms construct a directed acyclic graph representation of the task dependency relationship, and build a global multiprocessor schedule of the zero in-degree tasks to heuristically maximize the total accrued utility and ensure mutual exclusion. Potential deadlocks are detected through a cycle-detection algorithm, and resolved by aborting a task in the deadlock cycle. The GUA class of algorithms include two algorithms, namely, the Non-Greedy Global Utility Accrual (or NG-GUA) and Greedy Global Utility Accrual (or G-GUA) algorithms. NG-GUA and G-GUA differ in the way schedules are constructed towards meeting all task deadlines, when possible to do so. We establish several properties of the algorithms including conditions under which all task deadlines are met, satisfaction of mutual exclusion constraints, and deadlock-freedom. We create a Linux-based real-time kernel called ChronOS for multiprocessors. ChronOS is extended from the PREEMPT_RT real-time Linux patch, which provides optimized interrupt service latencies and real-time locking primitives. ChronOS provides a scheduling framework for the implementation of a broad range of real-time scheduling algorithms, including utility accrual, non-utility accrual, global, and partitioned scheduling algorithms. We implement the GUA class of algorithms and their competitors in ChronOS and conduct experimental studies. The competitors include G-EDF, G-NP-EDF, G-FIFO, gMUA, P-EDF and P-DASA. Our study reveals that the GUA class of algorithms accrue higher utility and satisfy greater number of deadlines than the deadline-based scheduling algorithms by as much as 750% and 600%, respectively. In addition, we observe that G-GUA accrues higher utility than NG-GUA during overloads by as much as 25% while NG-GUA satisfies greater number of deadlines than G-GUA by as much as 5% during underloads. / Master of Science

In this dissertation, we consider scheduling distributed soft real-time tasks in unreliable (e.g., those with arbitrary node and network failures) and untrustworthy systems (e.g., those with Byzantine node behaviors). We present a distributed real-time scheduling algorithm called Gamma. Gamma considers a distributed (i.e., multi-node) task model where tasks are subject to Time/Utility Function (or TUF) end-to-end time constraints, and the scheduling optimality criterion of maximizing the total accrued utility. The algorithm makes three novel contributions. First, Gamma uses gossip for reliably propagating task scheduling parameters and for discovering task execution nodes. Second, Gamma achieves distributed real-time mutual exclusion in unreliable environments. Third, the algorithm guards against potential disruption of message propagation due to Byzantine attacks using a mechanism called Launcher-Attacker-Infective-Susceptible-Immunized-Removed-Consumer (or LAISIRC). By doing so, the algorithm schedules tasks with probabilistic termination-time satisfactions, despite system unreliability and untrustworthiness. We analytically establish several timeliness and non-timeliness properties of the algorithm including probabilistic end-to-end task termination time satisfactions, optimality of message overheads, mutual exclusion guarantees, and the mathematical model of the LAISIRC mechanism. We conducted simulation-based experimental studies and compared Gamma with its competitors. Our experimental studies reveal that Gamma's scheduling algorithm accrues greater utility and satisfies a greater number of deadlines than do competitor algorithms (e.g., HVDF) by as much as 47% and 45%, respectively. LAISIRC is more tolerant to Byzantine attacks than competitor protocols (e.g., Path Verification) by obtaining as much as 28% higher correctness ratio. Gamma's mutual exclusion algorithm accrues greater utility than do competitor algorithms (e.g., EDF-Sigma) by as much as 25%. Further, we implemented the basic Gamma algorithm in the Emulab/ChronOS 250-node testbed, and measured the algorithm's performance. Our implementation measurements validate our theoretical analysis and the algorithm's effectiveness and robustness. / Ph. D.

In this dissertation, we consider timeliness and energy optimization in battery-powered, mobile embedded real-time systems. We focus on real-time systems that operate in environments with dynamically uncertain properties, including context-dependent activity execution times and arbitrary activity arrival patterns. We consider an application model where activities are subject to time/utility function (or TUF) time constraints, mutual exclusion constraints on concurrent sharing of non-CPU resources, timeliness requirements including assurances on individual activity timeliness behavior, and system-level energy consumption requirements including a non-exhaustable energy budget. To account for uncertainties in activity properties in dynamic systems, we stochastically describe activity execution demands, and describe activity arrival behaviors using the unimodal arbitrary arrival model, which allows unbounded arrival frequencies. We consider the scheduling optimality criteria of: (1) probabilistically satisfying lower bounds on individual activities' maximal timeliness utilities, and (2) maximizing system-level energy efficiency, while ensuring that the system's energy consumption never exhausts the energy budget and resource mutual exclusion constraints are satisfied. For this multi-criteria scheduling problem, we present a DVS (dynamic voltage scaling)-based, real-time scheduling algorithm called the Energy-Bounded Utility Accrual Algorithm (or EBUA). Since the scheduling problem is NP-hard, EBUA heuristically (and dynamically) allocates CPU cycles to activities, computes activity schedules, and scales CPU voltage and frequency with a polynomial-time cost. If activities' cumulative execution demands exceed the available CPU time or may exhaust the system's energy budget, the algorithm defers and rejects jobs in a controlled fashion, minimizing system-level energy consumption and maximizing total accrued utility. We analytically establish several properties of EBUA. We prove that the algorithm never exhausts the specified energy budget. Further, we establish EBUA's timeliness optimality during under-loads, freedom from deadlocks, and correctness in mutually exclusive resource sharing. In particular, we prove that the algorithm's timeliness behavior subsumes the optimal timeliness behavior of deadline scheduling as a special case, and identify the conditions under which lower bounds on individual activity utilities are satisfied. In addition, we upper bound the time needed for mutually exclusively accessing shared resources under EBUA. We conduct experimental studies by simulating the algorithm on the DVS-enabled AMD k6 processor model, and by implementing it on QNX Neutrino 6.2.1 RTOS. Our experimental results validate our analytical results. Further, they confirm EBUA's superiority over other energy-efficient real-time scheduling algorithms on timeliness and energy consumption behaviors. / Ph. D.