José Fonseca

A multi-DAG model for real-time parallel applications with conditional execution

Owing to the current trends for higher performance and the ever growing availability of multiprocessors in the embedded computing (EC) domain, there is nowadays a strong push towards the parallelization of modern embedded applications. Several real-time task models have recently been proposed to capture different forms of parallelism. However, they do not deal explicitly with control flow information as they assume that all the threads of a parallel task must execute every time the task is activated. In contrast, in this paper, we present a multi-DAG model where each task is characterized by a set of execution flows, each of which represents a different execution path throughout the task code and is modeled as a DAG of sub-tasks. We propose a two-step solution that computes a single synchronous DAG of servers for a task modeled by a multi-DAG and show that these servers are able to supply every execution flow of that task with the required cpu-budget so that the task can execute entirely, irrespective of the execution flow taken at run-time, while satisfying its precedence constraints. As a result, each task can be modeled by its single DAG of servers, which facilitates in leveraging the existing single-DAG schedulability analyses techniques for analyzing the schedulability of parallel tasks with multiple execution flows.

The Challenge of Time-Predictability in Modern Many-Core Architectures

The recent technological advancements and market trends are causing an interesting phenomenon towards the convergence of High-Performance Computing (HPC) and Embedded Computing (EC) domains. Many recent HPC applications require huge amounts of information to be processed within a bounded amount of time while EC systems are increasingly concerned with providing higher performance in real-time. The convergence of these two domains towards systems requiring both high performance and a predictable time-behavior challenges the capabilities of current hardware architectures. Fortunately, the advent of next-generation many-core embedded platforms has the chance of intercepting this converging need for predictability and high-performance, allowing HPC and EC applications to be executed on efficient and powerful heterogeneous architectures integrating general-purpose processors with many-core computing fabrics. However, addressing this mixed set of requirements is not without its own challenges and it is now of paramount importance to develop new techniques to exploit the massively parallel computation capabilities of many-core platforms in a predictable way.

Dynamic Global Scheduling of Parallel Real-Time Tasks

High-level parallel languages offer a simple way for application programmers to specify parallelism in a form that easily scales with problem size, leaving the scheduling of the tasks onto processors to be performed at runtime. Therefore, if the underlying system cannot efficiently execute those applications on the available cores, the benefits will be lost. In this paper, we consider how to schedule highly heterogenous parallel applications that require real-time performance guarantees on multicore processors. The paper proposes a novel scheduling approach that combines the global Earliest Deadline First (EDF) scheduler with a priority-aware work-stealing load balancing scheme, which enables parallel real-time tasks to be executed on more than one processor at a given time instant. Experimental results demonstrate the better scalability and lower scheduling overhead of the proposed approach comparatively to an existing real-time deadline-oriented scheduling class for the Linux kernel.

Response time analysis of sporadic DAG tasks under partitioned scheduling

Several schedulability analyses have been proposed for a variety of parallel task systems with real-time constraints. However, these analyses are mostly restricted to global scheduling policies. The problem with global scheduling is that it adds uncertainty to the lower-level timing analysis which on multicore systems are heavily context-dependent. As parallel tasks typically exhibit intense communication and concurrency among their sequential computational units, this problem is further exacerbated. This paper considers instead the schedulability of partitioned parallel tasks. More precisely, we present a response time analysis for sporadic DAG tasks atop multiprocessors under partitioned fixed-priority scheduling. We assume the partitioning to be given. We show that a partitioned DAG task can be modeled as a set of self-suspending tasks. We then propose an algorithm to traverse a DAG and characterize such worst-case scheduling scenario. With minor modifications, any state-of-the-art technique for sporadic self-suspending tasks can thus be used to derived the worst-case response time of a partitioned DAG task. Experiments show that the proposed approach significantly tightens the worst-case response time of partitioned parallel tasks comparatively to the state-of-the-art when the most accurate technique is chosen.

Improved response time analysis of sporadic DAG tasks for global FP scheduling

One of the major sources of pessimism in the response time analysis of globally scheduled real-time tasks is the computation of the upper-bound on the inter-task interference. This problem is further exacerbated when intra-task parallelism is permitted, because of the complex internal structure of parallel tasks. This paper considers the global fixed-priority scheduling (G-FP) of sporadic real-time tasks, each one modeled by a directed acyclic graph (DAG) of parallel subtasks. We present a response time analysis (RTA) technique based on the concept of problem window. We propose two novel techniques to derive less pessimistic upper-bounds on the workload produced by the carry-in and carry-out jobs of the interfering tasks, by taking into account the precedence constraints between their subtasks. We show that with these new upper-bounds, the proposed schedulability test does not only theoretically dominate state-of-the-art techniques but also offers significant improvements on the schedulability of DAG tasks for randomly generated task sets.