Both energy-efficiency and real-time performance are critical requirements in many embedded systems applications such as self-driving car, robotic system, disaster response, and security/safety control. These systems entail a myriad of real-time tasks, where each task itself is a parallel task that can utilize multiple computing units at the same time. Driven by the increasing demand for parallel tasks, multi-core embedded processors are inevitably evolving to many-core. Existing work on real-time parallel tasks mostly focused on real-time scheduling without addressing energy consumption. In this paper, we address hard real-time scheduling of parallel tasks while minimizing their CPU energy consumption on multicore embedded systems. Each task is represented as a directed acyclic graph (DAG) with nodes indicating different threads of execution and edges indicating their dependencies. Our technique is to determine the execution speeds of the nodes of the DAGs to minimize the overall energy consumption while meeting all task deadlines. It incorporates a frequency optimization engine and the dynamic voltage and frequency scaling (DVFS) scheme into the classical real-time scheduling policies (both federated and global) and makes them energy-aware. The contributions of this paper thus include the first energy-aware online federated scheduling and also the first energy-aware global scheduling of DAGs. Evaluation using synthetic workload through simulation shows that our energy-aware real-time scheduling policies can achieve up to 68% energy-saving compared to classical (energy-unaware) policies. We have also performed a proof of concept system evaluation using physical hardware demonstrating the energy efficiency through our proposed approach.
more »
« less
Cross-Layer Adaptation with Safety-Assured Proactive Task Job Skipping
During the operation of many real-time safety-critical systems, there are often strong needs for adapting to a dynamic environment or evolving mission objectives, e.g., increasing sampling and control frequencies of some functions to improve their performance under certain situations. However, a system's ability to adapt is often limited by tight resource constraints and rigid periodic execution requirements. In this work, we present a cross-layer approach to improve system adaptability by allowing proactive skipping of task executions, so that the resources can be either saved directly or re-allocated to other tasks for their performance improvement. Our approach includes three novel elements: (1) formal methods for deriving the feasible skipping choices of control tasks with safety guarantees at the functional layer, (2) a schedulability analysis method for assessing system feasibility at the architectural layer under allowed task job skippings, and (3) a runtime adaptation algorithm that efficiently explores job skipping choices and task priorities for meeting system adaptation requirements while ensuring system safety and timing correctness. Experiments demonstrate the effectiveness of our approach in meeting system adaptation needs.
more »
« less
- NSF-PAR ID:
- 10314697
- Date Published:
- Journal Name:
- ACM Transactions on Embedded Computing Systems
- Volume:
- 20
- Issue:
- 5s
- ISSN:
- 1539-9087
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
more » « less
In modern computing systems, jobs' resource requirements often vary over time. Accounting for this temporal variability during job scheduling is essential for meeting performance goals. However, theoretical understanding on how to schedule jobs with time-varying resource requirements is limited. Motivated by this gap, we propose a new setting of the stochastic bin-packing problem in service systems that allows for time-varying job resource requirements, also referred to as 'item sizes' in traditional bin-packing terms. In this setting, a job or 'item' must be dispatched to a server or 'bin' upon arrival. Its resource requirement may vary over time while in service, following a Markovian assumption. Once the job's service is complete, it departs from the system. Our goal is to minimize the expected number of active servers, or 'non-empty bins', in steady state.
Under our problem formulation, we develop a job dispatch policy, named Join-Reqesting-Server (JRS). Broadly, JRS lets each server independently evaluate its current job configuration and decide whether to accept additional jobs, balancing the competing objectives of maximizing throughput and minimizing the risk of resource capacity overruns. The JRS dispatcher then utilizes these individual evaluations to decide which server to dispatch each arriving job to. The theoretical performance guarantee of JRS is in the asymptotic regime where the job arrival rate scales large linearly with respect to a scaling factor r. We show that JRS achieves an additive optimality gap of O(√r) in the objective value, where the optimal objective value is Θ(r). When specialized to constant job resource requirements, our result improves upon the state-of-the-art o(r) optimality gap. Our technical approach highlights a novel policy conversion framework that reduces the policy design problem into a single-server problem.
-
null (Ed.)Modern latency-sensitive and real-time systems often use multi-core platforms; thus, tasks on different cores share certain hardware resources, such as the memory bus and certain cache levels. This has two undesirable consequences: (1) tasks can interfere with each other, causing high latency for the system as a whole, and (2) it becomes difficult to meet deadlines, since the worst-case timing of a given task depends on the worst task it might have to compete with. Static partitioning isolates tasks from each other by allocating a certain fraction of the resources to each; however, many tasks execute in different phases (e.g., memory-intensive and CPU-intensive) that have different requirements. Thus, system designers are left with a choice between overprovisioning, based on the most demanding phase, or suboptimal performance. In this paper, we propose a pair of techniques, called DNA and DADNA, to address the above challenge. DNA increases throughput and decreases latency, by building an execution profile of each task to identify the phases, and then dynamically allocating resources based on which task can benefit the most; DADNA further adds support for soft real-time workloads by taking deadlines into account. We have built a prototype of both techniques in the Xen hypervisor; our experimental results show that, compared to a state-of-the-art solution, DNA and DADNA can substantially improve schedulability, reduce job deadline miss ratios, and cut latencies by more than a factor of two even in extremely overloaded situations.more » « less
-
Obeid, I. ; Selesnik, I. ; Picone, J. (Ed.)The Neuronix high-performance computing cluster allows us to conduct extensive machine learning experiments on big data [1]. This heterogeneous cluster uses innovative scheduling technology, Slurm [2], that manages a network of CPUs and graphics processing units (GPUs). The GPU farm consists of a variety of processors ranging from low-end consumer grade devices such as the Nvidia GTX 970 to higher-end devices such as the GeForce RTX 2080. These GPUs are essential to our research since they allow extremely compute-intensive deep learning tasks to be executed on massive data resources such as the TUH EEG Corpus [2]. We use TensorFlow [3] as the core machine learning library for our deep learning systems, and routinely employ multiple GPUs to accelerate the training process. Reproducible results are essential to machine learning research. Reproducibility in this context means the ability to replicate an existing experiment – performance metrics such as error rates should be identical and floating-point calculations should match closely. Three examples of ways we typically expect an experiment to be replicable are: (1) The same job run on the same processor should produce the same results each time it is run. (2) A job run on a CPU and GPU should produce identical results. (3) A job should produce comparable results if the data is presented in a different order. System optimization requires an ability to directly compare error rates for algorithms evaluated under comparable operating conditions. However, it is a difficult task to exactly reproduce the results for large, complex deep learning systems that often require more than a trillion calculations per experiment [5]. This is a fairly well-known issue and one we will explore in this poster. Researchers must be able to replicate results on a specific data set to establish the integrity of an implementation. They can then use that implementation as a baseline for comparison purposes. A lack of reproducibility makes it very difficult to debug algorithms and validate changes to the system. Equally important, since many results in deep learning research are dependent on the order in which the system is exposed to the data, the specific processors used, and even the order in which those processors are accessed, it becomes a challenging problem to compare two algorithms since each system must be individually optimized for a specific data set or processor. This is extremely time-consuming for algorithm research in which a single run often taxes a computing environment to its limits. Well-known techniques such as cross-validation [5,6] can be used to mitigate these effects, but this is also computationally expensive. These issues are further compounded by the fact that most deep learning algorithms are susceptible to the way computational noise propagates through the system. GPUs are particularly notorious for this because, in a clustered environment, it becomes more difficult to control which processors are used at various points in time. Another equally frustrating issue is that upgrades to the deep learning package, such as the transition from TensorFlow v1.9 to v1.13, can also result in large fluctuations in error rates when re-running the same experiment. Since TensorFlow is constantly updating functions to support GPU use, maintaining an historical archive of experimental results that can be used to calibrate algorithm research is quite a challenge. This makes it very difficult to optimize the system or select the best configurations. The overall impact of all of these issues described above is significant as error rates can fluctuate by as much as 25% due to these types of computational issues. Cross-validation is one technique used to mitigate this, but that is expensive since you need to do multiple runs over the data, which further taxes a computing infrastructure already running at max capacity. GPUs are preferred when training a large network since these systems train at least two orders of magnitude faster than CPUs [7]. Large-scale experiments are simply not feasible without using GPUs. However, there is a tradeoff to gain this performance. Since all our GPUs use the NVIDIA CUDA® Deep Neural Network library (cuDNN) [8], a GPU-accelerated library of primitives for deep neural networks, it adds an element of randomness into the experiment. When a GPU is used to train a network in TensorFlow, it automatically searches for a cuDNN implementation. NVIDIA’s cuDNN implementation provides algorithms that increase the performance and help the model train quicker, but they are non-deterministic algorithms [9,10]. Since our networks have many complex layers, there is no easy way to avoid this randomness. Instead of comparing each epoch, we compare the average performance of the experiment because it gives us a hint of how our model is performing per experiment, and if the changes we make are efficient. In this poster, we will discuss a variety of issues related to reproducibility and introduce ways we mitigate these effects. For example, TensorFlow uses a random number generator (RNG) which is not seeded by default. TensorFlow determines the initialization point and how certain functions execute using the RNG. The solution for this is seeding all the necessary components before training the model. This forces TensorFlow to use the same initialization point and sets how certain layers work (e.g., dropout layers). However, seeding all the RNGs will not guarantee a controlled experiment. Other variables can affect the outcome of the experiment such as training using GPUs, allowing multi-threading on CPUs, using certain layers, etc. To mitigate our problems with reproducibility, we first make sure that the data is processed in the same order during training. Therefore, we save the data from the last experiment and to make sure the newer experiment follows the same order. If we allow the data to be shuffled, it can affect the performance due to how the model was exposed to the data. We also specify the float data type to be 32-bit since Python defaults to 64-bit. We try to avoid using 64-bit precision because the numbers produced by a GPU can vary significantly depending on the GPU architecture [11-13]. Controlling precision somewhat reduces differences due to computational noise even though technically it increases the amount of computational noise. We are currently developing more advanced techniques for preserving the efficiency of our training process while also maintaining the ability to reproduce models. In our poster presentation we will demonstrate these issues using some novel visualization tools, present several examples of the extent to which these issues influence research results on electroencephalography (EEG) and digital pathology experiments and introduce new ways to manage such computational issues.more » « less
-
Power grids are evolving at an unprecedented pace due to the rapid growth of distributed energy resources (DER) in communities. These resources are very different from traditional power sources as they are located closer to loads and thus can significantly reduce transmission losses and carbon emissions. However, their intermittent and variable nature often results in spikes in the overall demand on distribution system operators (DSO). To manage these challenges, there has been a surge of interest in building decentralized control schemes, where a pool of DERs combined with energy storage devices can exchange energy locally to smooth fluctuations in net demand. Building a decentralized market for transactive microgrids is challenging because even though a decentralized system provides resilience, it also must satisfy requirements like privacy, efficiency, safety, and security, which are often in conflict with each other. As such, existing implementations of decentralized markets often focus on resilience and safety but compromise on privacy. In this paper, we describe our platform, called TRANSAX, which enables participants to trade in an energy futures market, which improves efficiency by finding feasible matches for energy trades, enabling DSOs to plan their energy needs better. TRANSAX provides privacy to participants by anonymizing their trading activity using a distributed mixing service, while also enforcing constraints that limit trading activity based on safety requirements, such as keeping planned energy flow below line capacity. We show that TRANSAX can satisfy the seemingly conflicting requirements of efficiency, safety, and privacy. We also provide an analysis of how much trading efficiency is lost. Trading efficiency is improved through the problem formulation which accounts for temporal flexibility, and system efficiency is improved using a hybrid-solver architecture. Finally, we describe a testbed to run experiments and demonstrate its performance using simulation results.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1145/3477031
