Real-time scheduling of parallel tasks in the Linux Kernel

This paper proposes a global multiprocessor scheduling algorithm for the Linux kernel that combines the global EDF scheduler with a priority-aware work-stealing load balancing scheme, enabling parallel real-time tasks to be executed on more than one processor at a given time instant. We state that some priority inversion may actually be acceptable, provided it helps reduce contention, communication, synchronisation and coordination between parallel threads, while still guaranteeing the expected system’s predictability. Experimental results demonstrate the low scheduling overhead of the proposed approach comparatively to an existing real-time deadline-oriented scheduling class for the Linux kernel.

Towards the Combination of Work-Stealing and Semi-Partitioned Scheduling for Parallel Tasks

Paper/Poster presented in Work in Progress Session, 28th GI/ITG International Conference on Architecture of Computing Systems (ARCS 2015). 24 to 26, Mar, 2015. Porto, Portugal.

Towards the Combination of Work-Stealing and Semi-Partitioned Scheduling for Parallel Tasks

Poster presented in Work in Progress Session, 28th GI/ITG International Conference on Architecture of Computing Systems (ARCS 2015). 24 to 26, Mar, 2015. Porto, Portugal.

Server-based scheduling of parallel real-time tasks

Multicore platforms have transformed parallelism into a main concern. Parallel programming models are being put forward to provide a better approach for application programmers to expose the opportunities for parallelism by pointing out potentially parallel regions within tasks, leaving the actual and dynamic scheduling of these regions onto processors to be performed at runtime, exploiting the maximum amount of parallelism. It is in this context that this paper proposes a scheduling approach that combines the constant-bandwidth server abstraction with a priority-aware work-stealing load balancing scheme which, while ensuring isolation among tasks, enables parallel tasks to be executed on more than one processor at a given time instant.

Allocation of Parallel Real-Time Tasks in Distributed Multi-core Architectures supported by an FTT-SE Network

Distributed real-time systems such as automotive applications are becoming larger and more complex, thus, requiring the use of more powerful hardware and software architectures. Furthermore, those distributed applications commonly have stringent real-time constraints. This implies that such applications would gain in flexibility if they were parallelized and distributed over the system. In this paper, we consider the problem of allocating fixed-priority fork-join Parallel/Distributed real-time tasks onto distributed multi-core nodes connected through a Flexible Time Triggered Switched Ethernet network. We analyze the system requirements and present a set of formulations based on a constraint programming approach. Constraint programming allows us to express the relations between variables in the form of constraints. Our approach is guaranteed to find a feasible solution, if one exists, in contrast to other approaches based on heuristics. Furthermore, approaches based on constraint programming have shown to obtain solutions for these type of formulations in reasonable time.

A Multi-DAG Model for Real-Time Parallel Applications with Conditional Execution

The 30th ACM/SIGAPP Symposium On Applied Computing (SAC 2015). 13 to 17, Apr, 2015, Embedded Systems. Salamanca, Spain.

Semi-Partitioned Scheduling of Fork-Join Tasks using Work-Stealing

13th IEEE/IFIP International Conference on Embedded and Ubiquitous Computing (EUC 2015). 21 to 23, Oct, 2015, Session W1-A: Multiprocessing and Multicore Architectures. Porto, Portugal.

Task partitioning and priority assignment for distributed hard real-time systems

In this paper, we propose the Distributed using Optimal Priority Assignment (DOPA) heuristic that finds a feasible partitioning and priority assignment for distributed applications based on the linear transactional model. DOPA partitions the tasks and messages in the distributed system, and makes use of the Optimal Priority Assignment (OPA) algorithm known as Audsley’s algorithm, to find the priorities for that partition. The experimental results show how the use of the OPA algorithm increases in average the number of schedulable tasks and messages in a distributed system when compared to the use of Deadline Monotonic (DM) usually favoured in other works. Afterwards, we extend these results to the assignment of Parallel/Distributed applications and present a second heuristic named Parallel-DOPA (P-DOPA). In that case, we show how the partitioning process can be simplified by using the Distributed Stretch Transformation (DST), a parallel transaction transformation algorithm introduced in [1].

Modeling the effects of node heterogeneity on the performance of a real grid application

The performance benefit when using Grid systems comes from different strategies, among which partitioning the applications into parallel tasks is the most important. However, in most cases the enhancement coming from partitioning is smoothed by the effect of the synchronization overhead, mainly due to the high variability of completion times of the different tasks, which, in turn, is due to the large heterogeneity of Grid nodes. For this reason, it is important to have models which capture the performance of such systems. In this paper we describe a queueing-network-based performance model able to accurately analyze Grid architectures, and we use the model to study a real parallel application executed in a Grid. The proposed model improves the classical modelling techniques and highlights the impact of resource heterogeneity and network latency on the application performance.

Modeling the effects of node heterogeneity on the performance of grid applications

The performance benefit when using grid systems comes from different strategies, among which partitioning the applications into parallel tasks is the most important. However, in most cases the enhancement coming from partitioning is smoothed by the effects of synchronization overheads, mainly due to the high variability in the execution times of the different tasks, which, in turn, is accentuated by the large heterogeneity of grid nodes. In this paper we design hierarchical, queuing network performance models able to accurately analyze grid architectures and applications. Thanks to the model results, we introduce a new allocation policy based on a combination between task partitioning and task replication. The models are used to study two real applications and to evaluate the performance benefits obtained with allocation policies based on task replication.

Implementing intelligent cores using processor virtualization for fault tolerance

Processor virtualization for process migration in distributed parallel computing systems has formed a significant component of research on load balancing. In contrast, the potential of processor virtualization for fault tolerance has been addressed minimally. The work reported in this paper is motivated towards extending concepts of processor virtualization towards ‘intelligent cores’ as a means to achieve fault tolerance in distributed parallel computing systems. Intelligent cores are an abstraction of the hardware processing cores, with the incorporation of cognitive capabilities, on which parallel tasks can be executed and migrated. When a processing core executing a task is predicted to fail the task being executed is proactively transferred onto another core. A parallel reduction algorithm incorporating concepts of intelligent cores is implemented on a computer cluster using Adaptive MPI and Charm ++. Preliminary results confirm the feasibility of the approach.

Module property verification : A method to plan and perform quality verifications in modular architectures

Modular product architectures have generated numerous benefits for companies in terms of cost, lead-time and quality. The defined interfaces and the module’s properties decrease the effort to develop new product variants, and provide an opportunity to perform parallel tasks in design, manufacturing and assembly. The background of this thesis is that companies perform verifications (tests, inspections and controls) of products late, when most of the parts have been assembled. This extends the lead-time to delivery and ruins benefits from a modular product architecture; specifically when the verifications are extensive and the frequency of detected defects is high. Due to the number of product variants obtained from the modular product architecture, verifications must handle a wide range of equipment, instructions and goal values to ensure that high quality products can be delivered. As a result, the total benefits from a modular product architecture are difficult to achieve. This thesis describes a method for planning and performing verifications within a modular product architecture. The method supports companies by utilizing the defined modules for verifications already at module level, so called MPV (Module Property Verification). With MPV, defects are detected at an earlier point, compared to verification of a complete product, and the number of verifications is decreased. The MPV method is built up of three phases. In Phase A, candidate modules are evaluated on the basis of costs and lead-time of the verifications and the repair of defects. An MPV-index is obtained which quantifies the module and indicates if the module should be verified at product level or by MPV. In Phase B, the interface interaction between the modules is evaluated, as well as the distribution of properties among the modules. The purpose is to evaluate the extent to which supplementary verifications at product level is needed. Phase C supports a selection of the final verification strategy. The cost and lead-time for the supplementary verifications are considered together with the results from Phase A and B. The MPV method is based on a set of qualitative and quantitative measures and tools which provide an overview and support the achievement of cost and time efficient company specific verifications. A practical application in industry shows how the MPV method can be used, and the subsequent benefits

A methodology for granularity-based control of parallelism in logic programs

Several types of parallelism can be exploited in logic programs while preserving correctness and efficiency, i.e. ensuring that the parallel execution obtains the same results as the sequential one and the amount of work performed is not greater. However, such results do not take into account a number of overheads which appear in practice, such as process creation and scheduling, which can induce a slow-down, or, at least, limit speedup, if they are not controlled in some way. This paper describes a methodology whereby the granularity of parallel tasks, i.e. the work available under them, is efficiently estimated and used to limit parallelism so that the effect of such overheads is controlled. The run-time overhead associated with the approach is usually quite small, since as much work is done at compile time as possible. Also,a number of run-time optimizations are proposed. Moreover, a static analysis of the overhead associated with the granularity control process is performed in order to decide its convenience. The performance improvements resulting from the incorporation of grain size control are shown to be quite good, specially for systems with medium to large parallel execution overheads.

Towards granularity based control of parallelism in logic programs

Several types of parallelism can be exploited in logic programs while preserving correctness and efficiency, i.e. ensuring that the parallel execution obtains the same results as the sequential one and the amount of work performed is not greater. However, such results do not take into account a number of overheads which appear in practice, such as process creation and scheduling, which can induce a slow-down, or, at least, limit speedup, if they are not controlled in some way. This paper describes a methodology whereby the granularity of parallel tasks, i.e. the work available under them, is efficiently estimated and used to limit parallelism so that the effect of such overheads is controlled. The run-time overhead associated with the approach is usually quite small, since as much work is done at compile time as possible. Also, a number of run-time optimizations are proposed. Moreover, a static analysis of the overhead associated with the granularity control process is performed in order to decide its convenience. The performance improvements resulting from the incorporation of grain size control are shown to be quite good, specially for systems with médium to large parallel execution overheads.

GPRM: a high performance programming framework for manycore processors

Processors with large numbers of cores are becoming commonplace. In order to utilise the available resources in such systems, the programming paradigm has to move towards increased parallelism. However, increased parallelism does not necessarily lead to better performance. Parallel programming models have to provide not only flexible ways of defining parallel tasks, but also efficient methods to manage the created tasks. Moreover, in a general-purpose system, applications residing in the system compete for the shared resources. Thread and task scheduling in such a multiprogrammed multithreaded environment is a significant challenge. In this thesis, we introduce a new task-based parallel reduction model, called the Glasgow Parallel Reduction Machine (GPRM). Our main objective is to provide high performance while maintaining ease of programming. GPRM supports native parallelism; it provides a modular way of expressing parallel tasks and the communication patterns between them. Compiling a GPRM program results in an Intermediate Representation (IR) containing useful information about tasks, their dependencies, as well as the initial mapping information. This compile-time information helps reduce the overhead of runtime task scheduling and is key to high performance. Generally speaking, the granularity and the number of tasks are major factors in achieving high performance. These factors are even more important in the case of GPRM, as it is highly dependent on tasks, rather than threads. We use three basic benchmarks to provide a detailed comparison of GPRM with Intel OpenMP, Cilk Plus, and Threading Building Blocks (TBB) on the Intel Xeon Phi, and with GNU OpenMP on the Tilera TILEPro64. GPRM shows superior performance in almost all cases, only by controlling the number of tasks. GPRM also provides a low-overhead mechanism, called “Global Sharing”, which improves performance in multiprogramming situations. We use OpenMP, as the most popular model for shared-memory parallel programming as the main GPRM competitor for solving three well-known problems on both platforms: LU factorisation of Sparse Matrices, Image Convolution, and Linked List Processing. We focus on proposing solutions that best fit into the GPRM’s model of execution. GPRM outperforms OpenMP in all cases on the TILEPro64. On the Xeon Phi, our solution for the LU Factorisation results in notable performance improvement for sparse matrices with large numbers of small blocks. We investigate the overhead of GPRM’s task creation and distribution for very short computations using the Image Convolution benchmark. We show that this overhead can be mitigated by combining smaller tasks into larger ones. As a result, GPRM can outperform OpenMP for convolving large 2D matrices on the Xeon Phi. Finally, we demonstrate that our parallel worksharing construct provides an efficient solution for Linked List processing and performs better than OpenMP implementations on the Xeon Phi. The results are very promising, as they verify that our parallel programming framework for manycore processors is flexible and scalable, and can provide high performance without sacrificing productivity.