Syed W. Ali Website

Real-time locking protocols induce priority-inversion blocking (pi-blocking) when jobs access shared resources. Lower pi-blocking durations are highly valuable as they improve safety margins and reduce the pessimism inherent in scheduability analysis. To this end, this paper presents two real-time locking protocols for globally scheduled, job-level fixed-priority systems. The first is a mutual exclusion (mutex) protocol that strategically inserts runtime delays to achieve lower average-case pi-blocking durations than prior work. The second extends this technique as a reader-writer locking protocol with substantially lower worst-case pi-blocking bounds than the state of the art. This work concludes by demonstrating the reduced pi-blocking duration of the mutex locking protocol and showing vastly improved results in a schedulability study comparing the proposed reader-writer locking protocol to the state of the art.

Reliable utilization of the radio-frequency spectrum hinges upon spectrum sensing, the detection and identification of wireless transmitters. Timely completion of sensing tasks is necessary for accuracy, regulatory compliance, and in mission-critical scenarios. While AI-enabled, GPU-accelerated sensing algorithms are seeing increasing interest, prior research has not focused on providing strong response-time guarantees for such workloads. Furthermore, existing literature on real-time scheduling and GPU-synchronization is ill-suited for the high-throughput, high-sampling-rate pipeline of real-time spectrum-sensing systems. This paper proposes multicore CPU+GPU scheduling techniques designed for GPU-accelerated spectrum-sensing workloads. A dedicated GPU-management method is proposed for GPU-reliant sensing tasks not bound by the CPU, mitigating analytical pessimism and reducing runtime overheads. This work also extends the marginal cost model, originally introduced for CPU-based signal-processing work, to analyze and optimize the execution-time behavior of batched sensing jobs on the GPU. In an empirical case study, the proposed GPU-management and batching-optimization techniques are demonstrated to improve spectrum-sensing capacity and reduce response times by up to 17.8% versus a naive batching approach.

As safety-critical systems increasingly utilize AI workloads, providing robust real-time guarantees becomes a pivotal issue for certifiably safe operation. Spatial GPU partitioning techniques have been proposed to satisfy the computational requirements of such systems. With over 500,000 ways to select partitions on modern GPUs, it becomes infeasible to test every configuration while deriving worst-case execution times (WCETs) in measurement-based timing analysis; thus, understanding how and to what extent partitions can affect execution times is critical. This work explores how NVIDIA GPU partition selection affects WCETs, demonstrating an up to 70.4% increase solely due to partition selection.

In prior work, a number of asymptotically optimal suspension-based real-time locking protocols have been presented for job-level fixed priority (JLFP) schedulers, where job priorities do not change. However, the optimality proofs for these locking protocols break down under non-JLFP scheduling, where job priorities can vary. In fact, the problem of designing an asymptotically optimal real-time locking protocol for general non-JLFP scheduling has remained open. This paper closes this problem by presenting the non-JLFP locking protocol (NJLP), the first asymptotically optimal suspension-based real-time locking protocol for non-JLFP schedulers.

Artifact evalution instructions here.
The artifact is split into three parts:

1. Graphing the difference between the analytical worst-case blocking times for the NJLP and FMLP.
2. Determining and comparing the overhead costs of the NJLP and FMLP.
3. Determining and comparing the schedulability of systems using the NJLP vs. the FMLP.

Robot autonomy is driving an ever-increasing demand for computational power, including on-board multi-core CPUs and accelerators such as GPUs, to enable fast perception, planning, control, and more. Careful scheduling of these computational tasks on the CPU cores and GPUs is important to prevent locking up the finite computational capacity in ways that hinder other critical workloads; delays in computing time-critical tasks like obstacle detection and control can have huge negative consequences for autonomous robots, potentially resulting in damage, substantial financial loss, or even loss of life. In this paper, we leverage recent advances from real-time systems research. We apply TimeWall, a component-based real-time framework, to the computational components of an autonomous drone and experimentally show that the timeliness and safe operation properties of a drone are preserved even in the presence of increasing interfering computational processes.

Fourier transforms are vital for a broad range of signal-processing applications. Accelerating FFTs with GPUs offers an orders-of-magnitude improvement vs. CPU-only FFT computation. However, two problems arise when executing FFT tasks with other GPU work. First, concurrent GPU use introduces unpredictability in the form of lengthy response times. Second, it is unclear how to best parameterize and schedule FFT tasks to meet the throughput and timeliness constraints of real-time signal processing. This work investigates how FFT and other GPU-using tasks can concurrently access a GPU while maintaining bounded response-time guarantees without sacrificing throughput. In our experiments, the techniques proposed by this work result in an up to 17% improvement in worst-case FFT response times.

This paper presents a real-time locking protocol whose design was motivated by the goal of enabling safe GPU sharing in time-sliced component-based systems. This locking protocol enables a GPU to be shared concurrently across, and utilized within, isolated components with predictable execution times. It relies on a novel resizing technique where GPU work is dimensioned on-the-fly to run on partitions of an NVIDIA GPU. This technique can be applied to any component that internally utilizes global CPU scheduling. The proposed locking protocol enables increased GPU parallelism and reduces GPU capacity loss with analytically provable benefits.

Artifact evalution instructions here.
The artifact is split into two parts:

1. Determine worst-case response times given a number of SMs.
2. A simulated schedulability study comparing using the SMLP vs. the OMLP.