Training machine learning (ML) models with large datasets can incur significant resource contention on shared clusters. This training typically involves many iterations that continually improve the quality of the model. Yet in exploratory settings, better models can be obtained faster by directing resources to jobs with the most potential for improvement. We describe SLAQ, a cluster scheduling system for approximate ML training jobs that aims to maximize the overall job quality. When allocating cluster resources, SLAQ explores the quality-runtime trade-offs across multiple jobs to maximize system-wide quality improvement. To do so, SLAQ leverages the iterative nature of ML training algorithms, by collecting quality and resource usage information from concurrent jobs, and then generating highly-tailored quality-improvement predictions for future iterations. Experiments show that SLAQ achieves an average quality improvement of up to 73% and an average delay reduction of up to 44% on a large set of ML training jobs, compared to resource fairness schedulers.
Loop scheduling techniques aim to achieve load-balanced executions of scientific applications. Dynamic loop self-scheduling (DLS) libraries for distributed-memory systems are typically MPI-based and employ a centralized chunk calculation approach (CCA) to assign variably-sized chunks of loop iterations. We present a distributed chunk calculation approach (DCA) that supports various types of DLS techniques. Using both CCA and DCA, twelve DLS techniques are implemented and evaluated in different CPU slowdown scenarios. The results show that the DLS techniques implemented using DCA outperform the…
The essence of distributed computing systems is how to schedule incoming requests and how to allocate all computing nodes to minimize both time and computation costs. In this paper, we propose a cost-aware optimal scheduling and allocation strategy for distributed computing systems while minimizing the cost function including response time and service cost. First, based on the proposed cost function, we derive the optimal request scheduling policy and the optimal resource allocation policy synchronously. Second, considering the effects of incoming requests on the scheduling policy, the additiv…
Computational Grids, emerging as an infrastructure for next generation computing, enable the sharing, selection, and aggregation of geographically distributed resources for solving large-scale problems in science, engineering, and commerce. As the resources in the Grid are heterogeneous and geographically distributed with varying availability and a variety of usage and cost policies for diverse users at different times and, priorities as well as goals that vary with time. The management of resources and application scheduling in such a large and distributed environment is a complex task. This …
Edge computing is an emerging technology which places computing at the edge of the network to provide an ultra-low latency. Computation offloading, a paradigm that migrates computing from mobile devices to remote servers, can now use the power of edge computing by offloading computation to cloudlets in edge-clouds. However, the task scheduling of computation offloading in edge-clouds faces a two-fold challenge. First, as cloudlets are geographically distributed, it is difficult for each cloudlet to perform load balancing without centralized control. Second, as tasks of computation offloading h…
The existing distributed TDMA-scheduling techniques can be classified as either static or dynamic. The primary purpose of static TDMA-scheduling algorithms is to improve the channel utilization by generating a schedule of shorter length. But, they usually take a longer time to schedule, and hence, are not suitable for WSNs, in which the network topology changes dynamically. On the other hand, dynamic TDMA-scheduling algorithms generate a schedule quickly, but they are not efficient in terms of generated schedule length. In this paper, we propose a new approach to TDMA scheduling for WSNs, that…
Training Graph Neural Networks (GNN) on large graphs is resource-intensive and time-consuming, mainly due to the large graph data that cannot be fit into the memory of a single machine, but have to be fetched from distributed graph storage and processed on the go. Unlike distributed deep neural network (DNN) training, the bottleneck in distributed GNN training lies largely in large graph data transmission for constructing mini-batches of training samples. Existing solutions often advocate data-computation colocation, and do not work well with limited resources where the colocation is infeasibl…
The aim of this paper is to provide a description of deep-learning-based scheduling approach for academic-purpose high-performance computing systems. The share of academic-purpose distributed computing systems (DCS) reaches 17.4 percents amongst TOP500 supercomputer sites (15.6 percents in performance scale) that makes them a valuable object of research. The core of this approach is to predict the future workflow of the system depending on the previously submitted tasks using deep learning algorithm. Information on predicted tasks is used by the resource management system (RMS) to perform effi…
Clusters, grids, and peer-to-peer (P2P) networks have emerged as popular paradigms for next generation parallel and distributed computing. The management of resources and scheduling of applications in such large-scale distributed systems is a complex undertaking. In order to prove the effectiveness of resource brokers and associated scheduling algorithms, their performance needs to be evaluated under different scenarios such as varying number of resources and users with different requirements. In a grid environment, it is hard and even impossible to perform scheduler performance evaluation in …
Scheduling the maintenance based on the condition, respectively the degradation level of the system leads to improved system's reliability while minimizing the maintenance cost. Since the degradation level changes dynamically during the system's operation, we face a dynamic maintenance scheduling problem. In this paper, we address the dynamic maintenance scheduling of manufacturing systems based on their degradation level. The manufacturing system consists of several units with a defined capacity and an individual dynamic degradation model, seeking to optimize their reward. The units sell thei…
Distributed Stream Processing frameworks are being commonly used with the evolution of Internet of Things(IoT). These frameworks are designed to adapt to the dynamic input message rate by scaling in/out.Apache Storm, originally developed by Twitter is a widely used stream processing engine while others includes Flink, Spark streaming. For running the streaming applications successfully there is need to know the optimal resource requirement, as over-estimation of resources adds extra cost.So we need some strategy to come up with the optimal resource requirement for a given streaming application…
The workflow is a general notion representing the automated processes along with the flow of data. The automation ensures the processes being executed in the order. Therefore, this feature attracts users from various background to build the workflow. However, the computational requirements are enormous and investing for a dedicated infrastructure for these workflows is not always feasible. To cater to the broader needs, multi-tenant platforms for executing workflows were began to be built. In this paper, we identify the problems and challenges in the multiple workflows scheduling that adhere t…
In this paper we analyze the problem of optimal task scheduling for data centers. Given the available resources and tasks, we propose a fast distributed iterative algorithm which operates over a large scale network of nodes and allows each of the interconnected nodes to reach agreement to an optimal solution in a finite number of time steps. More specifically, the algorithm (i) is guaranteed to converge to the exact optimal scheduling plan in a finite number of time steps and, (ii) once the goal of task scheduling is achieved, it exhibits distributed stopping capabilities (i.e., it allows the …
Recent years have witnessed a rapid growth of distributed machine learning (ML) frameworks, which exploit the massive parallelism of computing clusters to expedite ML training. However, the proliferation of distributed ML frameworks also introduces many unique technical challenges in computing system design and optimization. In a networked computing cluster that supports a large number of training jobs, a key question is how to design efficient scheduling algorithms to allocate workers and parameter servers across different machines to minimize the overall training time. Toward this end, in th…
Nowadays large-scale distributed machine learning systems have been deployed to support various analytics and intelligence services in IT firms. To train a large dataset and derive the prediction/inference model, e.g., a deep neural network, multiple workers are run in parallel to train partitions of the input dataset, and update shared model parameters. In a shared cluster handling multiple training jobs, a fundamental issue is how to efficiently schedule jobs and set the number of concurrent workers to run for each job, such that server resources are maximally utilized and model training can…
We consider a parallel system of $m$ identical machines prone to unpredictable crashes and restarts, trying to cope with the continuous arrival of tasks to be executed. Tasks have different computational requirements (i.e., processing time or size). The flow of tasks, their size, and the crash and restart of the machines are assumed to be controlled by an adversary. Then, we focus on the study of online distributed algorithms for the efficient scheduling of the tasks. We use competitive analysis, considering as efficiency metric the completed-load, i.e., the aggregated size of the completed ta…
Scheduling is essentially a decision-making process that enables resource sharing among a number of activities by determining their execution order on the set of available resources. The emergence of distributed systems brought new challenges on scheduling in computer systems, including clusters, grids, and more recently clouds. On the other hand, the plethora of research makes it hard for both newcomers researchers to understand the relationship among different scheduling problems and strategies proposed in the literature, which hampers the identification of new and relevant research avenues.…
In end-to-end distributed real time systems, a task may be executed sequentially on different processors. The end-toend task response time must not exceed the end-to-end task deadline to consider the task a schedulable task. In transient over load periods, deadlines may be missed or processors may saturate. The imprecise computation technique is a way to overcome the mentioned problems by trading off precision and timeliness. We developed an imprecise integrated framework for scheduling fixed priority end-to-end tasks in distributed real time systems by extending an existing integrated framewo…
Fueled by advances in distributed deep learning (DDL), recent years have witnessed a rapidly growing demand for resource-intensive distributed/parallel computing to process DDL computing jobs. To resolve network communication bottleneck and load balancing issues in distributed computing, the so-called ``ring-all-reduce'' decentralized architecture has been increasingly adopted to remove the need for dedicated parameter servers. To date, however, there remains a lack of theoretical understanding on how to design resource optimization algorithms for efficiently scheduling ring-all-reduce DDL job…
This work addresses the problem of scheduling user-defined analytic applications, which we define as high-level compositions of frameworks, their components, and the logic necessary to carry out work. The key idea in our application definition, is to distinguish classes of components, including rigid and elastic types: the first being required for an application to make progress, the latter contributing to reduced execution times. We show that the problem of scheduling such applications poses new challenges, which existing approaches address inefficiently. Thus, we present the design and evalu…
High-throughput computing projects require the solution of large numbers of problems. In many cases, these problems can be solved on desktop PCs, or can be broken down into independent "PC-solvable" sub-problems. In such cases, the projects are high-performance computing projects, but only because of the sheer number of the needed calculations. We briefly describe our efforts to increase the throughput of one such project. We then explain how to easily set up a distributed computing facility composed of standard networked PCs running Windows 95, 98, 2000, or NT. The facility requires no specia…
Task graphs provide a simple way to describe scientific workflows (sets of tasks with dependencies) that can be executed on both HPC clusters and in the cloud. An important aspect of executing such graphs is the used scheduling algorithm. Many scheduling heuristics have been proposed in existing works; nevertheless, they are often tested in oversimplified environments. We provide an extensible simulation environment designed for prototyping and benchmarking task schedulers, which contains implementations of various scheduling algorithms and is open-sourced, in order to be fully reproducible. W…
The use of meta-schedulers for resource management in large-scale distributed systems often leads to a hierarchy of schedulers. In this paper, we discuss why existing meta-scheduling hierarchies are sometimes not sufficient for Grid systems due to their inability to re-organise jobs already scheduled locally. Such a job re-organisation is required to adapt to evolving loads which are common in heavily used Grid infrastructures. We propose a peer-to-peer scheduling model and evaluate it using case studies and mathematical modelling. We detail the DIANA (Data Intensive and Network Aware) schedul…
Resource Allocation and Workload Scheduling for Large-Scale Distributed Deep Learning: A Survey
Artificial Intelligence Research Institute, Shenzhen MSU-BIT University（深圳北理莫斯科大学人工智能研究院）；Guangdong-Hong Kong-Macao Joint Laboratory for Emotional Intelligence and Pervasive Computing, Shenzhen MSU-BIT University（深圳北理莫斯科大学粤港澳情感智能与普适计算联合实验室）；School of Information Science and Engineering, Lanzhou University（兰州大学信息科学与工程学院）；Department of Electrical and Computer Engineering, The University of British Columbia（不列颠哥伦比亚大学电气与计算机工程系）
Computationally-intensive loops are the primary source of parallelism in scientific applications. Such loops are often irregular and a balanced execution of their loop iterations is critical for achieving high performance. However, several factors may lead to an imbalanced load execution, such as problem characteristics, algorithmic, and systemic variations. Dynamic loop self-scheduling (DLS) techniques are devised to mitigate these factors, and consequently, improve application performance. On distributed-memory systems, DLS techniques can be implemented using a hierarchical master-worker exe…
Cloud computing is an emerging technology in distributed computing which facilitates pay per model as per user demand and requirement.Cloud consist of a collection of virtual machine which includes both computational and storage facility. The primary aim of cloud computing is to provide efficient access to remote and geographically distributed resources. Cloud is developing day by day and faces many challenges, one of them is scheduling. Scheduling refers to a set of policies to control the order of work to be performed by a computer system. A good scheduler adapts its scheduling strategy acco…
The availability of powerful microprocessors and high-speed networks as commodity components has enabled high performance computing on distributed systems (wide-area cluster computing). In this environment, as the resources are usually distributed geographically at various levels (department, enterprise, or worldwide) there is a great challenge in integrating, coordinating and presenting them as a single resource to the user; thus forming a computational grid. Another challenge comes from the distributed ownership of resources with each resource having its own access policy, cost, and mechanis…
The next generation of scientific experiments and studies, popularly called as e-Science, is carried out by large collaborations of researchers distributed around the world engaged in analysis of huge collections of data generated by scientific instruments. Grid computing has emerged as an enabler for e-Science as it permits the creation of virtual organizations that bring together communities with common objectives. Within a community, data collections are stored or replicated on distributed resources to enhance storage capability or efficiency of access. In such an environment, scientists ne…
A customary solution to reduce the energy consumption of wireless communication devices is to periodically put the radio into low-power sleep mode. A relevant problem is to schedule the wake-up of nodes in such a way as to ensure proper coordination among devices, respecting delay constraints while still saving energy. In this paper, we introduce a simple algebraic characterization of the problem of periodic wake-up scheduling under both energy consumption and delay constraints. We demonstrate that the general problem of wake-up times coordination is equivalent to integer factorization and dis…
Wireless sensor networks are normally characterized by resource challenged nodes. Since communication costs the most in terms of energy in these networks, minimizing this overhead is important. We consider minimum length node scheduling in regular multi-hop wireless sensor networks. We present collision-free decentralized scheduling algorithms based on TDMA with spatial reuse that do not use message passing, this saving communication overhead. We develop the algorithms using graph-based k-hop interference model and show that the schedule complexity in regular networks is independent of the num…
A fundamental problem in wireless networks is the maximum link scheduling problem: given a set $L$ of links, compute the largest possible subset $L'\subseteq L$ of links that can be scheduled simultaneously without interference. This problem is particularly challenging in the physical interference model based on SINR constraints (referred to as the SINR model), which has gained a lot of interest in recent years. Constant factor approximation algorithms have been developed for this problem, but low complexity distributed algorithms that give the same approximation guarantee in the SINR model ar…
Computational Grids are a new trend in distributed computing systems. They allow the sharing of geographically distributed resources in an efficient way, extending the boundaries of what we perceive as distributed computing. Various sciences can benefit from the use of grids to solve CPU-intensive problems, creating potential benefits to the entire society. Job scheduling is an integrated part of parallel and distributed computing. It allows selecting correct match of resource for a particular job and thus increases the job throughput and utilization of resources. Job should be scheduled in an…
The overall performance of a distributed system is highly dependent on the communication efficiency of the system. Although network resources (links, bandwidth) are becoming increasingly more available, the communication performance of data transfers involving large volumes of data does not necessarily improve at the same rate. This is due to the inefficient usage of the available network resources. A solution to this problem consists of data transfer scheduling techniques, which manage and allocate the network resources in an efficient manner. In this paper we present several online and offli…
Distributed asynchronous offline training has received widespread attention in recent years because of its high performance on large-scale data and complex models. As data are distributed from cloud-centric to edge nodes, a big challenge for distributed machine learning systems is how to handle native and natural non-independent and identically distributed (non-IID) data for training. Previous asynchronous training methods do not have a satisfying performance on non-IID data because it would result in that the training process fluctuates greatly which leads to an abnormal convergence. We propo…
We propose an asynchronous iterative scheme that allows a set of interconnected nodes to distributively reach an agreement within a pre-specified bound in a finite number of steps. While this scheme could be adopted in a wide variety of applications, we discuss it within the context of task scheduling for data centers. In this context, the algorithm is guaranteed to approximately converge to the optimal scheduling plan, given the available resources, in a finite number of steps. Furthermore, by being asynchronous, the proposed scheme is able to take into account the uncertainty that can be int…
The method of choice for parameter aggregation in Deep Neural Network (DNN) training, a network-intensive task, is shifting from the Parameter Server model to decentralized aggregation schemes (AllReduce) inspired by theoretical guarantees of better performance. However, current implementations of AllReduce overlook the interdependence of communication and computation, resulting in significant performance degradation. In this paper, we develop Caramel, a system that accelerates decentralized distributed deep learning through model-aware computation scheduling and communication optimizations fo…
We consider the classical problem of scheduling task graphs corresponding to complex applications on distributed computing systems. A number of heuristics have been previously proposed to optimize task scheduling with respect to metrics such as makespan and throughput. However, they tend to be slow to run, particularly for larger problem instances, limiting their applicability in more dynamic systems. Motivated by the goal of solving these problems more rapidly, we propose, for the first time, a graph convolutional network-based scheduler (GCNScheduler). By carefully integrating an inter-task …
Hunan Normal University（湖南师范大学）；Hunan University of Commerce（湖南商学院）；Northumbria University（诺森比亚大学）；University of Essex（埃塞克斯大学）；Queen Mary University of London（伦敦玛丽女王大学）
We consider the optimization of distributed resource scheduling to minimize the sum of task latency and energy consumption for all the Internet of things devices (IoTDs) in a large-scale mobile edge computing (MEC) system. To address this problem, we propose a distributed intelligent resource scheduling (DIRS) framework, which includes centralized training relying on the global information and distributed decision making by each agent deployed in each MEC server. More specifically, we first introduce a novel multi-agent ensemble-assisted distributed deep reinforcement learning (DRL) architectu…
The problem of attaining energy efficiency in distributed systems is of importance, but a general, non-domain-specific theory of energy-minimal scheduling is far from developed. In this paper, we classify the problems of energy-minimal scheduling and present theoretical foundations of the same. We derive results concerning energy-minimal scheduling of independent jobs in a distributed system with functionally similar machines with different working and idle power ratings. The machines considered in our system can have identical as well as different speeds. If the jobs can be divided into arbit…
Institute of Automation and Applied Informatics (IAI)（自动化与应用信息研究所（IAI））；Karlsruhe Institute of Technology (KIT)（卡尔斯鲁厄理工学院（KIT））
Recent years have seen an increasing integration of distributed renewable energy resources into existing electric power grids. Due to the uncertain nature of renewable energy resources, network operators are faced with new challenges in balancing load and generation. In order to meet the new requirements, intelligent distributed energy resource plants can be used which provide as virtual power plants e.g. demand side management or flexible generation. However, the calculation of an adequate schedule for the unit commitment of such distributed energy resources is a complex optimization problem …
