Search for: All records

Due to the emergence of parallel architectures and parallel programming frameworks, modern real-time applications are often composed of parallel tasks that can occupy multiple processors at the same time. Among parallel task models, gang scheduling has received much attention in recent years due to its performance efficiency and applicability to parallel architectures such as graphics processing units. Despite this attention, the soft real-time (SRT) scheduling of gang tasks has received little attention. This paper, for the first time, considers the SRT-feasibility problem for gang tasks. Necessary and sufficient feasibility conditions are presented that relate the SRTfeasibility problem to the HRT-feasibility problem of “equivalent” task systems. Based on these conditions, intractability results for SRT gang scheduling are derived. This paper also presents server-based scheduling policies, corresponding schedulability tests, and an improved schedulability condition for the global-earlies-tdeadline-first (GEDF) scheduling of gang tasks. Moreover, GEDF is shown to be non-optimal in scheduling SRT gang tasks. 

To certify the schedulability of a system, valid per-task worst-case execution-time (WCET) estimates are almost always required. Unfortunately, on multicore machines, deriving WCET estimates through static analysis that is not highly pessimistic may never be a practical reality. The alternative is to determine WCETs via a measurement process, but such a process cannot correctly produce accurate WCET estimates with certainty. This lack of certainty necessitates the use of overrun-handling mechanisms, such as budget-enforcement techniques, to preserve temporal correctness at runtime. In many systems of interest today, tasks are interconnected to form processing graphs, which can be quite large. The simplest (and perhaps most common) approach to budget enforcement in this case is to abort an entire graph invocation whenever any node (task) overruns its budget. However, such an approach can result in a high abort rate at the graph level even when the per-node abort rate is low. To remedy this situation, this paper presents a holistic budget-management strategy for directed acyclic graphs (DAGs) that involves reallocating per-node budgets to overrunning nodes to avoid DAG-level aborts. To enable the effects of aborts to be studied analytically, a probabilistic analysis is presented to derive a DAG’s abort rate under the proposed budget-management strategy. Experimental results are also presented to demonstrate the utility of budgeting graphs holistically 

Real-time locking protocols are typically designed to reduce any priority-inversion blocking (pi9 blocking) a task may incur while waiting to access a shared resource. For the multiprocessor case, a number of such protocols have been developed that ensure asymptotically optimal pi-blocking bounds under job-level fxed-priority scheduling. Unfortunately, no optimal multiprocessor real-time locking protocols are known that ensure tight pi-blocking bounds under any scheduler. This paper presents the frst such protocols. Specifcally, protocols are presented for mutual exclusion, reader-writer synchronization, and k-exclusion that are optimal under frst-in-frst-out (FIFO) scheduling when schedulability analysis treats suspension times as computation. Experiments are presented that demonstrate the efectiveness of these protocols. 

Embedded and autonomous systems are increasingly integrating AI/ML features, often enabled by a hardware accelerator such as a GPU. As these workloads become increasingly demanding, but size, weight, power, and cost constraints remain unyielding, ways to increase GPU capacity are an urgent need. In this work, we provide a means by which to spatially partition the computing units of NVIDIA GPUs transparently, allowing oft-idled capacity to be reclaimed via safe and effcient GPU sharing. Our approach works on any NVIDIA GPU since 2013, and can be applied via our easy-to-use, user-space library titled libsmctrl. We back the design of our system with deep investigations into the hardware scheduling pipeline of NVIDIA GPUs. We provide guidelines for the use of our system, and demonstrate it via an object detection case study using YOLOv2. 

Real-time locking protocols are typically designed to reduce any priority-inversion blocking (pi-blocking) a task may incur while waiting to access a shared resource. For the multiprocessor case, a number of such protocols have been developed that ensure asymptotically optimal pi-blocking bounds under job-level fixed-priority scheduling. Unfortunately, no optimal multiprocessor real-time locking protocols are known that ensure tight pi-blocking bounds under any scheduler. This paper presents the first such protocols. Specifically, protocols are presented for mutual exclusion, reader-writer synchronization, and k-exclusion that are optimal under first-in-first-out (FIFO) scheduling when schedulability analysis treats suspension times as computation. Experiments are presented that demonstrate the effectiveness of these protocols.