Modern GPU clusters, particularly those built on NVIDIA's Multi-Instance GPU (MIG) architecture, often suffer from inefficiencies because jobs are treated as rigid, indivisible blocks that occupy a fixed slice until completion. The reliance on static peak memory estimates exacerbates fragmentation, underutilization, and job rejections. We propose Scheduler-Driven Job Atomization (SJA), a new paradigm that establishes a bidirectional interaction between scheduler and jobs. In SJA, the scheduler advertises available execution gaps, and jobs respond by signaling interest if they can potentially generate a subjob that fits the offered time-capacity window. The scheduler may collect multiple signals for the same slot and, based on its allocation policy (e.g., fairness, efficiency, or SLA priorities), selects which job is granted the slot. Only then does the chosen job materialize a safe, self-contained subjob tailored to that opportunity. Unlike migration or preemption, SJA proactively shapes workloads before execution, thereby avoiding costly state transfers and unpredictable interruptions. It aims to increase GPU utilization, reduce wait times, and minimize migration overhead by aligning jobs with opportunities in real time, ensuring that each admitted subjob is correct by construction. This paper is presented as a concept paper: it introduces the paradigm, defines its building blocks, and outlines future research directions, rather than offering a full experimental evaluation.
Job schedulers are a key component of scalable computing infrastructures. They orchestrate all of the work executed on the computing infrastructure and directly impact the effectiveness of the system. Recently, job workloads have diversified from long-running, synchronously-parallel simulations to include short-duration, independently parallel high performance data analysis (HPDA) jobs. Each of these job types requires different features and scheduler tuning to run efficiently. A number of schedulers have been developed to address both job workload and computing system heterogeneity. High perf…
Distributed data processing systems like MapReduce, Spark, and Flink are popular tools for analysis of large datasets with cluster resources. Yet, users often overprovision resources for their data processing jobs, while the resource usage of these jobs also typically fluctuates considerably. Therefore, multiple jobs usually get scheduled onto the same shared resources to increase the resource utilization and throughput of clusters. However, job runtimes and the utilization of shared resources can vary significantly depending on the specific combinations of co-located jobs. This paper presents…
Despite the fact that size-based schedulers can give excellent results in terms of both average response times and fairness, data-intensive computing execution engines generally do not employ size-based schedulers, mainly because of the fact that job size is not known a priori. In this work, we perform a simulation-based analysis of the performance of size-based schedulers when they are employed with the workload of typical data-intensive schedules and with approximated size estimations. We show results that are very promising: even when size estimation is very imprecise, response times of siz…
In the rapidly expanding field of parallel processing, job schedulers are the "operating systems" of modern big data architectures and supercomputing systems. Job schedulers allocate computing resources and control the execution of processes on those resources. Historically, job schedulers were the domain of supercomputers, and job schedulers were designed to run massive, long-running computations over days and weeks. More recently, big data workloads have created a need for a new class of computations consisting of many short computations taking seconds or minutes that process enormous quanti…
Efficient scheduling of distributed deep learning (DL) jobs in large GPU clusters is crucial for resource efficiency and job performance. While server sharing among jobs improves resource utilization, interference among co-located DL jobs occurs due to resource contention. Interference-aware job placement has been studied, with white-box approaches based on explicit interference modeling and black-box schedulers with reinforcement learning. In today's clusters containing thousands of GPU servers, running a single scheduler to manage all arrival jobs in a timely and effective manner is challeng…
Hadoop has become the de facto standard for processing large data in today's cloud environment. The performance of Hadoop in the cloud has a direct impact on many important applications ranging from web analytic, web indexing, image and document processing to high-performance scientific computing. However, because of the scale, complexity and dynamic nature of the cloud, failures are common and these failures often impact the performance of jobs running in Hadoop. Although Hadoop possesses built-in failure detection and recovery mechanisms, several scheduled jobs still fail because of unforese…
To solve the limitation of Hadoop on scalability, resource sharing, and application support, the open-source community proposes the next generation of Hadoop's compute platform called Yet Another Resource Negotiator (YARN) by separating resource management functions from the programming model. This separation enables various application types to run on YARN in parallel. To achieve fair resource sharing and high resource utilization, YARN provides the capacity scheduler and the fair scheduler. However, the performance impacts of the two schedulers are not clear when mixed applications run on a …
Size-based schedulers have very desirable performance properties: optimal or near-optimal response time can be coupled with strong fairness guarantees. Despite this, such systems are very rarely implemented in practical settings, because they require knowing a priori the amount of work needed to complete jobs: this assumption is very difficult to satisfy in concrete systems. It is definitely more likely to inform the system with an estimate of the job sizes, but existing studies point to somewhat pessimistic results if existing scheduler policies are used based on imprecise job size estimation…
The ever-growing processing power of supercomputers in recent decades enables us to explore increasing complex scientific problems. Effective scheduling these jobs is crucial for individual job performance and system efficiency. The traditional job schedulers in high performance computing (HPC) are simple and concentrate on improving CPU utilization. The emergence of new hardware resources and novel hardware structure impose severe challenges on traditional schedulers. The increasing diverse workloads, including compute-intensive and data-intensive applications, require more efficient schedule…
In this paper, a method for efficient scheduling to obtain optimum job throughput in a distributed campus grid environment is presented; Traditional job schedulers determine job scheduling using user and job resource attributes. User attributes are related to current usage, historical usage, user priority and project access. Job resource attributes mainly comprise of soft requirements (compilers, libraries) and hard requirements like memory, storage and interconnect. A job scheduler dispatches jobs to a resource if a job's hard and soft requirements are met by a resource. In current scenario d…
Diverse workloads such as interactive supercomputing, big data analysis, and large-scale AI algorithm development, requires a high-performance scheduler. This paper presents a novel node-based scheduling approach for large scale simulations of short running jobs on MIT SuperCloud systems, that allows the resources to be fully utilized for both long running batch jobs while simultaneously providing fast launch and release of large-scale short running jobs. The node-based scheduling approach has demonstrated up to 100 times faster scheduler performance that other state-of-the-art systems.
Resource allocation in High Performance Computing (HPC) settings is still not easy for end-users due to the wide variety of application and environment configuration options. Users have difficulties to estimate the number of processors and amount of memory required by their jobs, select the queue and partition, and estimate when job output will be available to plan for next experiments. Apart from wasting infrastructure resources by making wrong allocation decisions, overall user response time can also be negatively impacted. Techniques that exploit batch scheduler systems to predict waiting t…
High-performance computing (HPC) is undergoing significant changes. Next generation HPC systems are equipped with diverse global and local resources, such as I/O burst buffer resources, memory resources (e.g., on-chip and off-chip RAM, external RAM/NVRA), network resources, and possibly other resources. Job schedulers play a crucial role in efficient use of resources. However, traditional job schedulers are single-objective and fail to efficient use of other resources. In this paper, we propose ROME, a novel multi-dimensional job scheduling framework to explore potential tradeoffs among multip…
The ability to accurately estimate job runtime properties allows a scheduler to effectively schedule jobs. State-of-the-art online cluster job schedulers use history-based learning, which uses past job execution information to estimate the runtime properties of newly arrived jobs. However, with fast-paced development in cluster technology (in both hardware and software) and changing user inputs, job runtime properties can change over time, which lead to inaccurate predictions. In this paper, we explore the potential and limitation of real-time learning of job runtime properties, by proactively…
Today's clusters often have to divide resources among a diverse set of jobs. These jobs are heterogeneous both in execution time and in their rate of arrival. Execution time heterogeneity has lead to the development of hybrid schedulers that can schedule both short and long jobs to ensure good task placement. However, arrival rate heterogeneity, or burstiness, remains a problem in existing schedulers. These hybrid schedulers manage resources on statically provisioned cluster, which can quickly be overwhelmed by bursts in the number of arriving jobs. In this paper we propose CloudCoaster, a hyb…
The University of Hong Kong（香港大学）；Alibaba Inc.（阿里巴巴公司）
More and more companies have deployed machine learning (ML) clusters, where deep learning (DL) models are trained for providing various AI-driven services. Efficient resource scheduling is essential for maximal utilization of expensive DL clusters. Existing cluster schedulers either are agnostic to ML workload characteristics, or use scheduling heuristics based on operators' understanding of particular ML framework and workload, which are less efficient or not general enough. In this paper, we show that DL techniques can be adopted to design a generic and efficient scheduler. DL2 is a DL-drive…
University of North Carolina at Charlotte（北卡罗来纳大学夏洛特分校）；Iowa State University（爱荷华州立大学）；Oak Ridge National Laboratory（橡树岭国家实验室）
Today high-performance computing (HPC) platforms are still dominated by batch jobs. Accordingly, effective batch job scheduling is crucial to obtain high system efficiency. Existing HPC batch job schedulers typically leverage heuristic priority functions to prioritize and schedule jobs. But, once configured and deployed by the experts, such priority functions can hardly adapt to the changes of job loads, optimization goals, or system settings, potentially leading to degraded system efficiency when changes occur. To address this fundamental issue, we present RLScheduler, an automated HPC batch …
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…
The proliferation of multi-core and multiprocessor-based computer systems has led to explosive development of parallel applications and hence the need for efficient schedulers. In this paper, we study hierarchical scheduling for malleable parallel jobs on multiprocessor-based systems, which appears in many distributed and multilayered computing environments. We propose a hierarchical scheduling algorithm, named AC-DS, that consists of a feedback-driven adaptive scheduler, a desire aggregation scheme and an efficient resource allocation policy. From theoretical perspective, we show that AC-DS h…
Advance reservation is important to guarantee the quality of services of jobs by allowing exclusive access to resources over a defined time interval on resources. It is a challenge for the scheduler to organize available resources efficiently and to allocate them for parallel AR jobs with deadline constraint appropriately. This paper provides a slot-based data structure to organize available resources of multiprocessor systems in a way that enables efficient search and update operations, and formulates a suite of scheduling policies to allocate resources for dynamically arriving AR requests. T…
In hadoop, the job scheduling is an independent module, users can design their own job scheduler based on their actual application requirements, thereby meet their specific business needs. Currently, hadoop has three schedulers: FIFO, computing capacity scheduling and fair scheduling policy, all of them are take task allocation strategy that considerate data locality simply. They neither support data locality well nor fully apply to all cases of jobs scheduling. In this paper, we took the concept of resources-prefetch into consideration, and proposed a job scheduling algorithm based on data lo…
Several companies and research institutes are moving their CPU-intensive applications to hybrid High Performance Computing (HPC) cloud environments. Such a shift depends on the creation of software systems that help users decide where a job should be placed considering execution time and queue wait time to access on-premise clusters. Relying blindly on turnaround prediction techniques will affect negatively response times inside HPC cloud environments. This paper introduces a tool to make job placement decisions in HPC hybrid cloud environments taking into account the inaccuracy of execution a…
MapReduce has become a popular programming model for running data intensive applications on the cloud. Completion time goals or deadlines of MapReduce jobs set by users are becoming crucial in existing cloud-based data processing environments like Hadoop. There is a conflict between the scheduling MR jobs to meet deadlines and "data locality" (assigning tasks to nodes that contain their input data). To meet the deadline a task may be scheduled on a node without local input data for that task causing expensive data transfer from a remote node. In this paper, a novel scheduler is proposed to add…
Clusters of computers have emerged as mainstream parallel and distributed platforms for high-performance, high-throughput and high-availability computing. To enable effective resource management on clusters, numerous cluster managements systems and schedulers have been designed. However, their focus has essentially been on maximizing CPU performance, but not on improving the value of utility delivered to the user and quality of services. This paper presents a new computational economy driven scheduling system called Libra, which has been designed to support allocation of resources based on the…
Applications in science and engineering often require huge computational resources for solving problems within a reasonable time frame. Parallel supercomputers provide the computational infrastructure for solving such problems. A traditional application scheduler running on a parallel cluster only supports static scheduling where the number of processors allocated to an application remains fixed throughout the lifetime of execution of the job. Due to the unpredictability in job arrival times and varying resource requirements, static scheduling can result in idle system resources thereby decrea…
Virtualization technology has enabled applications to be decoupled from the underlying hardware providing the benefits of portability, better control over execution environment and isolation. It has been widely adopted in scientific grids and commercial clouds. Since virtualization, despite its benefits incurs a performance penalty, which could be significant for systems dealing with uncertainty such as High Performance Computing (HPC) applications where jobs have tight deadlines and have dependencies on other jobs before they could run. The major obstacle lies in bridging the gap between perf…
This work seeks to design decisionmaking rules for autonomous agents to jointly influence and optimize the behavior of teamed human decisionmakers in the presence of an adversary. We study a situation in which computational jobs are scheduled on servers by a collection of autonomous machines in concert with self-interested human decisionmakers, and the human and machine schedulers must react to an adversary's attack on one of the servers. We show a simple machine scheduling policy such that if all schedulers have permission to schedule jobs on all servers, increasing the penetration of machine…
We propose an approach to utilize idle computational resources of supercomputers. The idea is to maintain an additional queue of low-priority non-parallel jobs and execute them in containers, using container migration tools to break the execution down into separate intervals. We propose a container management system that can maintain this queue and interact with the supercomputer scheduler. We conducted a series of experiments simulating supercomputer scheduler and the proposed system. The experiments demonstrate that the proposed system increases the effective utilization of supercomputer res…
Microsoft Research（微软研究院）；University of Texas at Austin（德克萨斯大学奥斯汀分校）；VMware Research（威睿研究院）
Training Deep Neural Networks (DNNs) is a widely popular workload in both enterprises and cloud data centers. Existing schedulers for DNN training consider GPU as the dominant resource, and allocate other resources such as CPU and memory proportional to the number of GPUs requested by the job. Unfortunately, these schedulers do not consider the impact of a job's sensitivity to allocation of CPU, memory, and storage resources. In this work, we propose Synergy, a resource-sensitive scheduler for shared GPU clusters. Synergy infers the sensitivity of DNNs to different resources using optimistic p…
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…
The under exploitation of the available resources risks to be one of the main problems for a computing center. The growing demand of computational power necessarily entails more complex approaches in the management of the computing resources, with particular attention to the batch queue system scheduler. In a heterogeneous batch queue system, available for both serial single core processes and parallel multi core jobs, it may happen that one or more computational nodes composing the cluster are not fully occupied, running a number of jobs lower than their actual capability. A typical case is r…
Rapid launch of thousands of jobs is essential for effective interactive supercomputing, big data analysis, and AI algorithm development. Achieving thousands of launches per second has required hardware to be available to receive these jobs. This paper presents a novel preemptive approach to implement spot jobs on MIT SuperCloud systems allowing the resources to be fully utilized for both long running batch jobs while still providing fast launch for interactive jobs. The new approach separates the job preemption and scheduling operations and can achieve 100 times faster performance in the sche…
Deep Reinforcement Learning (DRL) is being increasingly applied to the problem of resource allocation for emerging System-on-Chip (SoC) applications, and has shown remarkable promises. In this paper, we introduce SoCRATES (SoC Resource AdapTivE Scheduler), an extremely efficient DRL-based SoC scheduler which maps a wide range of hierarchical jobs to heterogeneous resources within SoC using the Eclectic Interaction Matching (EIM) technique. It is noted that the majority of SoC resource management approaches have been targeting makespan minimization with fixed number of jobs in the system. In co…
National University of Singapore（新加坡国立大学）；ByteDance（字节跳动）
Efficient GPU resource scheduling is essential to maximize resource utilization and save training costs for the increasing amount of deep learning workloads in shared GPU clusters. Existing GPU schedulers largely rely on static policies to leverage the performance characteristics of deep learning jobs. However, they can hardly reach optimal efficiency due to the lack of elasticity. To address the problem, we propose ONES, an ONline Evolutionary Scheduler for elastic batch size orchestration. ONES automatically manages the elasticity of each job based on the training batch size, so as to maximi…
We present Henge, a system to support intent-based multi-tenancy in modern stream processing applications. Henge supports multi-tenancy as a first-class citizen: everyone inside an organization can now submit their stream processing jobs to a single, shared, consolidated cluster. Additionally, Henge allows each tenant (job) to specify its own intents (i.e., requirements) as a Service Level Objective (SLO) that captures latency and/or throughput. In a multi-tenant cluster, the Henge scheduler adapts continually to meet jobs' SLOs in spite of limited cluster resources, and under dynamic input wo…
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 sys…
Computational Grid is enormous environments with heterogeneous resources and stable infrastructures among other Internet-based computing systems. However, the managing of resources in such systems has its special problems. Scheduler systems need to get last information about participant nodes from information centers for the purpose of firmly job scheduling. In this paper, we focus on online updating resource information centers with processed and provided data based on the assumed hierarchical model. A hybrid knowledge extraction method has been used to classifying grid nodes based on predict…
