Scheduling job flows efficiently and rapidly on distributed computing clusters is one of huge challenges for daily operation of data centers. In a practical scenario, a single job consists of numerous stages with complex dependency relation represented as a Directed Acyclic Graph (DAG) structure. Nowadays a data center usually equips with a cluster of heterogeneous computing servers which are different in the hardware/software configuration. From both the cost saving and environmental friendliness, the data centers could benefit a lot from optimizing the job scheduling problems in the heterogeneous environment. Thus the problem has attracted more and more attention from both the industry and academy. In this paper, we propose a task-duplication based learning algorithm, namely \lachesis \footnote{The second of the Three Fates in ancient Greek mythology, who determines destiny.}, aiming to optimize the problem. In the proposed approach, it first perceives the topological dependencies between jobs using a reinforcement learning framework and a specially designed graph neural network (GNN) to select the most promising task to be executed. Then the task is assigned to a specific executor with the consideration of duplicating all its precedent tasks according to an expert-designed rules. We have conducted extensive experiments over standard workloads to evaluate the proposed solution. The experimental results suggest that \lachesisquad can achieve at most 26.7\% reduction of makespan and 35.2\% improvement of speedup ratio over seven strong baseline algorithms, including the state-of-the-art heuristics methods and a variety of deep reinforcement learning based algorithms.
To satisfy the increasing performance needs of modern cyber-physical systems, multiprocessor architectures are increasingly utilized. To efficiently exploit their potential parallelism in hard real-time systems, appropriate task models and scheduling algorithms that allow providing timing guarantees are required. Such scheduling algorithms and the corresponding worst-case response time analyses usually suffer from resource over-provisioning due to pessimistic analyses based on worst-case assumptions. Hence, scheduling algorithms and analysis with high resource efficiency are required. A promin…
We consider global fixed-priority (G-FP) scheduling of parallel tasks, in which each task is represented as a directed acyclic graph (DAG). We summarize and highlight limitations of the state-of-the-art analyses for G-FP and propose a novel technique for bounding interfering workload, which can be applied directly to generalized DAG tasks. Our technique works by constructing optimization problems for which the optimal solution values serve as safe and tight upper bounds for interfering workloads. Using the proposed workload bounding technique, we derive a response-time analysis and show that i…
Vehicular cloud (VC) platforms integrate heterogeneous and distributed resources of moving vehicles to offer timely and cost-effective computing services. However, the dynamic nature of VCs (i.e., limited contact duration among vehicles), caused by vehicles' mobility, poses unique challenges to the execution of computation-intensive applications/tasks with directed acyclic graph (DAG) structure, where each task consists of multiple interdependent components (subtasks). In this paper, we study scheduling of DAG tasks over dynamic VCs, where multiple subtasks of a DAG task are dispersed across v…
Many HPC applications can be expressed as mixed-mode computations, in which each node of a computational DAG is itself a parallel computation that can be molded at runtime to allocate different amounts of processing resources. At the same time, modern HPC systems are becoming increasingly heterogeneous to address the requirements of energy efficiency. Effectively using heterogeneous devices is complex, requiring the developer to be aware of each DAG nodes' criticality, and the relative performance of the underlying heterogeneous resources. This paper studies how irregular mixed-mode parallel c…
Multiprocessor scheduling of hard real-time tasks modeled by directed acyclic graphs (DAGs) exploits the inherent parallelism presented by the model. For DAG tasks, a node represents a request to execute an object on one of the available processors. In one DAG task, there may be multiple execution requests for one object, each represented by a distinct node. These distinct execution requests offer an opportunity to reduce their combined cache overhead through coordinated scheduling of objects as threads within a parallel task. The goal of this work is to realize this opportunity by incorporati…
Python-written data analytics applications can be modeled as and compiled into a directed acyclic graph (DAG) based workflow, where the nodes are fine-grained tasks and the edges are task dependencies. Such analytics workflow jobs are increasingly characterized by short, fine-grained tasks with large fan-outs. These characteristics make them well-suited for a new cloud computing model called serverless computing or Function-as-a-Service (FaaS), which has become prevalent in recent years. The auto-scaling property of serverless computing platforms accommodates short tasks and bursty workloads, …
Real-time systems increasingly use multicore processors in order to satisfy thermal, power, and computational requirements. To exploit the architectural parallelism offered by the multicore processors, parallel task models, scheduling algorithms and response-time analyses with respect to real-time constraints have to be provided. In this paper, we propose a reservation-based scheduling algorithm for sporadic constrained-deadline parallel conditional DAG tasks with probabilistic execution behaviour for applications that can tolerate bounded number of deadline misses and bounded tardiness. We de…
Edge Generation Scheduling for DAG Tasks Using Deep Reinforcement Learning
TUM School of Engineering and Design, Technical University of Munich（慕尼黑工业大学TUM工程与设计学院）；Department of Electrical Engineering and Computer Sciences, University of California Berkeley（加州大学伯克利分校电气工程与计算机科学系）
This paper considers the scheduling of parallel real-time tasks with arbitrary-deadlines. Each job of a parallel task is described as a directed acyclic graph (DAG). In contrast to prior work in this area, where decomposition-based scheduling algorithms are proposed based on the DAG-structure and inter-task interference is analyzed as self-suspending behavior, this paper generalizes the federated scheduling approach. We propose a reservation-based algorithm, called reservation-based federated scheduling, that dominates federated scheduling. We provide general constraints for the design of such…
Static (offline) techniques for mapping applications given by task graphs to MPSoC systems often deliver overly pessimistic and thus suboptimal results w.r.t. exploiting time slack in order to minimize the energy consumption. This holds true in particular in case computation times of tasks may be workload-dependent and becoming known only at runtime or in case of conditionally executed tasks or scenarios. This paper studies and quantitatively evaluates different classes of algorithms for scheduling periodic applications given by task graphs (i.e., DAGs) with precedence constraints and a global…
Coflow is a recently proposed networking abstraction to help improve the communication performance of data-parallel computing jobs. In multi-stage jobs, each job consists of multiple coflows and is represented by a Directed Acyclic Graph (DAG). Efficiently scheduling coflows is critical to improve the data-parallel computing performance in data centers. Compared with hand-tuned scheduling heuristics, existing work DeepWeave [1] utilizes Reinforcement Learning (RL) framework to generate highly-efficient coflow scheduling policies automatically. It employs a graph neural network (GNN) to encode …
GPUs are readily available in cloud computing and personal devices, but their use for data processing acceleration has been slowed down by their limited integration with common programming languages such as Python or Java. Moreover, using GPUs to their full capabilities requires expert knowledge of asynchronous programming. In this work, we present a novel GPU run time scheduler for multi-task GPU computations that transparently provides asynchronous execution, space-sharing, and transfer-computation overlap without requiring in advance any information about the program dependency structure. W…
In the era of Internet of Things, there is an increasing demand for networked computing to support the requirements of the time-constrained, compute-intensive distributed applications such as multi-camera video processing and data fusion for security. We present Jupiter, an open source networked computing system that inputs a Directed Acyclic Graph (DAG)-based computational task graph to efficiently distribute the tasks among a set of networked compute nodes regardless of their geographical separations and orchestrates the execution of the DAG thereafter. This Kubernetes container-orchestratio…
Heterogeneous multi-cores utilize the strength of different architectures for executing particular types of workload, and usually offer higher performance and energy efficiency. In this paper, we study the worst-case response time (WCRT) analysis of \emph{typed} scheduling of parallel DAG tasks on heterogeneous multi-cores, where the workload of each vertex in the DAG is only allowed to execute on a particular type of cores. The only known WCRT bound for this problem is grossly pessimistic and suffers the \emph{non-self-sustainability} problem. In this paper, we propose two new WCRT bounds. Th…
Modern production data processing and machine learning pipelines on the cloud are critical components for many cloud-based companies. These pipelines are typically composed of complex workflows represented by directed acyclic graphs (DAGs). Cloud environments are attractive to these workflows due to the wide range of choice with heterogeneous instances and prices that can provide the flexibility for different cost-performance needs. However, this flexibility also leads to the complexity of selecting the right resource configuration (e.g., instance type, resource demands) for each task in the D…
Applications in data-parallel computing typically consist of multiple stages. In each stage, a set of intermediate parallel data flows (Coflow) is produced and transferred between servers to enable starting of next stage. While there has been much research on scheduling isolated coflows, the dependency between coflows in multi-stage jobs has been largely ignored. In this paper, we consider scheduling coflows of multi-stage jobs represented by general DAGs (Directed Acyclic Graphs) in a shared data center network, so as to minimize the total weighted completion time of jobs. This problem is sig…
Distributed applications, such as database queries and distributed training, consist of both compute and network tasks. DAG-based abstraction primarily targets compute tasks and has no explicit network-level scheduling. In contrast, Coflow abstraction collectively schedules network flows among compute tasks but lacks the end-to-end view of the application DAG. Because of the dependencies and interactions between these two types of tasks, it is sub-optimal to only consider one of them. We argue that co-scheduling of both compute and network tasks can help applications towards the globally optim…
With huge amounts of training data, deep learning has made great breakthroughs in many artificial intelligence (AI) applications. However, such large-scale data sets present computational challenges, requiring training to be distributed on a cluster equipped with accelerators like GPUs. With the fast increase of GPU computing power, the data communications among GPUs have become a potential bottleneck on the overall training performance. In this paper, we first propose a general directed acyclic graph (DAG) model to describe the distributed synchronous stochastic gradient descent (S-SGD) algor…
Specialized accelerators such as GPUs, TPUs, FPGAs, and custom ASICs have been increasingly deployed to train deep learning models. These accelerators exhibit heterogeneous performance behavior across model architectures. Existing schedulers for clusters of accelerators, which are used to arbitrate these expensive training resources across many users, have shown how to optimize for various multi-job, multi-user objectives, like fairness and makespan. Unfortunately, existing schedulers largely do not consider performance heterogeneity. In this paper, we propose Gavel, a heterogeneity-aware sche…
Hard real-time systems like image processing, autonomous driving, etc. require an increasing need of computational power that classical multi-core platforms can not provide, to fulfill with their timing constraints. Heterogeneous Instruction Set Architecture (ISA) platforms allow accelerating real-time workloads on application-specific cores (e.g. GPU, DSP, ASICs) etc. and are suitable for these applications. In addition, these platforms provide larger design choices as a given functionnality can be implemented onto several types of compute elements. HPC-DAG (Heterogeneous Parallel Directed Ac…
Department of Computer Science, Hong Kong Baptist University（香港浸会大学计算机科学系）
Distributed Deep Learning (DDL) has rapidly grown its popularity since it helps boost the training performance on high-performance GPU clusters. Efficient job scheduling is indispensable to maximize the overall performance of the cluster when training multiple jobs simultaneously. However, existing schedulers do not consider the communication contention of multiple communication tasks from different distributed training jobs, which could deteriorate the system performance and prolong the job completion time. In this paper, we first establish a new DDL job scheduling framework which organizes D…
Pollux improves scheduling performance in deep learning (DL) clusters by adaptively co-optimizing inter-dependent factors both at the per-job level and at the cluster-wide level. Most existing schedulers expect users to specify the number of resources for each job, often leading to inefficient resource use. Some recent schedulers choose job resources for users, but do so without awareness of how DL training can be re-optimized to better utilize the provided resources. Pollux simultaneously considers both aspects. By monitoring the status of each job during training, Pollux models how their goo…
Shenzhen Research Institute of Big Data（深圳大数据研究院）；Southern University of Science and Technology（南方科技大学）；China Academy of Information and Communications Technology（中国信息通信研究院）；The Chinese University of Hong Kong (Shenzhen)（香港中文大学（深圳））；Zhejiang University（浙江大学）
The popular federated edge learning (FEEL) framework allows privacy-preserving collaborative model training via frequent learning-updates exchange between edge devices and server. Due to the constrained bandwidth, only a subset of devices can upload their updates at each communication round. This has led to an active research area in FEEL studying the optimal device scheduling policy for minimizing communication time. However, owing to the difficulty in quantifying the exact communication time, prior work in this area can only tackle the problem partially by considering either the communicatio…
