Batch Jobs Research Articles

Efficient algorithms to minimize makespan of the unrelated parallel batch-processing machines scheduling problem with unequal job ready times

This paper considers the minimization of makespan in the unrelated parallel batch processing machines scheduling problem with considering non-identical job size and dynamic job ready time. The considered unrelated machines have different capacity and different processing speed. Each machine processes a number of the jobs as a batch at the same time so that the machine’s capacity is not exceeded. The batch processing time and the batch ready time are equal to the largest processing time and the largest ready time of jobs in the same batch, respectively. In this paper, a Mixed Integer Linear Programming (MILP) model, two categories of the heuristic procedures (six heuristics) and a meta-heuristic algorithm are proposed to solve the problem. A lower bound is also presented by relaxing of the original problem to evaluate the quality of the proposed algorithms. The computational experiments show the performance of the proposed algorithms under the considered measures.

GCsnap: Interactive Snapshots for the Comparison of Protein-Coding Genomic Contexts

The biological function and evolutionary history of protein-coding genes are not only written in their nucleotide sequences. The comparison of genomic contexts throughout different lineages may highlight genomic mechanisms in the generation of new protein families, while the conservation of gene clusters may unravel, for instance, metabolic pathways. Various tools and databases exist that allow for the analysis and comparison of genomic contexts, but each has its own limitations. Online databases allow for quick comparisons, but only for those genomes for which data were pre-calculated. More advanced tools may allow for the comparison of any genome, but are often limited to a given phylogenetic kingdom or provide only a snapshot of the genomic contexts without further information about the genes involved. Here, we introduce GCsnap, a flexible Python-based tool that allows for the interactive comparison of the genomic contexts of protein-coding genes from any genome at any taxonomic level, integrating them with functional and structural information. By connecting the output to different protein databases, users can navigate through the different genomic contexts from a simple interactive platform, facilitating the further analysis of the contexts found. GCsnap is not limited to a single input format, can perform batch jobs and accepts protein classification maps. Results are stored in detailed, human and machine-readable files, and customizable, publication-ready figures. GCsnap is freely available from https://github.com/JoanaMPereira/GCsnap and can be set up easily on any computer.

Improving the cost efficiency of large-scale cloud systems running hybrid workloads - A case study of Alibaba cluster traces

The pandemic of coronavirus has dramatically disrupted the retail industry, as many stores are forced to close and people across the world are shelter-in-place with online shopping as the inevitable choice. To meet the rapidly increasing demand for e-commerce, more data centers are expected to provide new or significantly improve existing cloud services that can better support hybrid workloads (e.g. online purchase jobs and batch jobs that support ranking or recommendation systems). Successful cloud systems need to efficiently handle and quickly respond to huge volume of traffic with such hybrid workloads. Meanwhile, it is critical to reduce the total cost of ownership (TCO) for profitability. Improving system utilization is one of the effective techniques to achieve the twin goals of high performance and low TCO. This paper conducts a comprehensive analysis on the 2017 and 2018 cluster traces released by Alibaba, which provides a case study about Alibaba's best practices in improving the performance and cost efficiency of its large-scale cloud systems by consolidating time-sensitive online service jobs with time-insensitive batch jobs. Our investigation indicates that the over-subscription (causing resource waste and low utilization) and under-subscription (causing performance degradation) problems co-exist in the current Alibaba system. We develop a simulator that allows us to evaluate possible solutions to address this problem and their impact on the performance, energy consumption, and TCO. Our experiments show that the estimated TCO can be reduced by $600,000 for the 2018 trace running on over 4,000 machines without compromising performance. The TCO can decrease by nearly $68 million if similar strategy is extrapolated to Alibaba's 432,000 web facing servers.

배치 작업의 자원 가용성 최대화를 위한 스케줄링 기법

수많은 계산 노드로 구성된 클러스터 환경에서 배치 작업 스케줄러는 계산 자원의 상태를 인지하여 순서에 맞게 효율적으로 작업을 할당하는 역할을 수행한다. 클러스터 내의 계산 자원에 작업을 할당하는 방법은 제출된 시간의 순서를 따르는 것이 공평한 방법이다. 그러나 한정된 가용 자원을 효율적으로 사용하기 위해서는 작업의 규모와 수행 시간을 고려한 스케줄링 알고리즘이 필요하며 다양한 연구들이 진행되어 왔다.BR 본 연구는 클러스터 환경에서 배치 작업 스케줄러의 수행 환경을 분석하고 작업 회수시간(Turn-around Time) 최소화하기 위해 고려할 요인들을 검토하였다. 나아가 자원의 가용성을 높이기 위한 스케줄링 알고리즘 기법을 연구하고 각 기법에 대한 시뮬레이션을 수행을 통해 성능을 평가했다. 이로써 클러스터의 유휴 계산자원의 활용성을 높임으로써 사용자는 작업 수행에 필요한 자원을 빠르게 확보할 수 있다. 향후 실제 클러스터 환경에 접목함으로써 자원의 활용률과 최적화에 기여할 수 있다.

Batch loading and scheduling problem with processing time deterioration and rate-modifying activities

ABSTRACT This research addresses a single machine batch loading and scheduling problem. Jobs in the same family are processed as a batch in the machine with a known family-specific processing time. Each job in a batch requires a known volume or space, and the total batch volume cannot exceed the available volume/capacity of the machine. Batch processing times increase proportionately with the time since the most recent rate-modifying activity and the starting time of a batch. A rate-modifying activity can be executed which restores original batch processing times. In this research, a solution procedure is proposed that simultaneously determines the appropriate batching of jobs and the number of rate-modifying activities. Job batches and the rate-modifying activities are then sequenced to minimise the makespan. To develop a solution procedure, a mixed integer linear programming model is formulated and a tight lower bound is proposed. Three genetic algorithms (GAs), including batch loading and sequencing heuristics, are proposed. The performance of the three GAs is compared, and the best GA is compared to other meta-heuristic algorithms.

Unbounded Serial-Batching Scheduling on Hierarchical Optimization

The paper considers a serial-batching scheduling problem on hierarchical optimization with two regular maximum costs, where hierarchical optimization means the primary objective function is minimized, and keeping the minimum value of the primary objective function, the secondary objective function is also minimized. In serial-batching machine environment, the machine processes the jobs in batch, and the jobs in the identical batch are processed by entering into the machine together and leaving the machine together. The time taken to process a batch amounts to the total processing time of the jobs in the batch. Moreover, a fixed switching time s is inserted when a machine begins to process a new batch. We only study the unbounded model, i.e., the batch capacity is unbounded. We give an algorithm that can solve the hierarchical optimization problem in $$O(n^4)$$ time, where n denotes the number of jobs.

Optimizing Internal Overlaps by Self-Adjusting Resource Allocation in Multi-Stage Computing Systems

With the rise of big data, more and more users will launch computing systems to process a large volume of data in various applications. A Scheduling algorithm is crucial to the performance of the processing platforms, especially when they are concurrently executing a batch of jobs. Such jobs usually represent multiple stages. Each stage produces the intermediate data which will be piped to the next stage for further processing. However, the scheduling problem in a big data computing system is different from the traditional multi-stage job scheduling problem as for any two consecutive stages, the later stage usually starts before the former stage is finished to “shuffle” the intermediate data. In this paper, we consider MapReduce/Hadoop as a representative computing system and develop a new strategy named OMO, Optimize MapReduce Overlap with a Good Start (Reduce) and a Good Finish (Map). A MapReduce job contains two consecutive phases: map and reduce. Our general target is to optimize the internal overlap between these two phases. There are two new techniques included in our solution, Lazy start of reduce tasks and Batch finish of map tasks, which aim to approach an effective alignment of the two phases based on the characteristics of the MapReduce process. OMO has been implemented on the Hadoop system with extensive experiments for performance evaluation. The results show that OMO's performance is superior in terms of total completion time (i.e., makespan) of a batch of jobs.

Distributed statistical inference with pyhf enabled through funcX

In High Energy Physics facilities that provide High Performance Computing environments provide an opportunity to efficiently perform the statistical inference required for analysis of data from the Large Hadron Collider, but can pose problems with orchestration and efficient scheduling. The compute architectures at these facilities do not easily support the Python compute model, and the configuration scheduling of batch jobs for physics often requires expertise in multiple job scheduling services. The combination of the pure-Python libraries pyhf and funcX reduces the common problem in HEP analyses of performing statistical inference with binned models, that would traditionally take multiple hours and bespoke scheduling, to an on-demand (fitting) “function as a service” that can scalably execute across workers in just a few minutes, offering reduced time to insight and inference. We demonstrate execution of a scalable workflow using funcX to simultaneously fit 125 signal hypotheses from a published ATLAS search for new physics using pyhf with a wall time of under 3 minutes. We additionally show performance comparisons for other physics analyses with openly published probability models and argue for a blueprint of fitting as a service systems at HPC centers.

A discrete artificial bee colony algorithm for the distributed heterogeneous no-wait flowshop scheduling problem

With the development of globalization, distributed manufacturing has become one of the main modes of manufacturing. The situation in which a number of heterogeneous factories coproduce a batch of jobs is ubiquitous in the field of distributed manufacturing. Compared with isomorphic factories, heterogeneous factories bring further difficulty in assigning jobs to factories. This paper considers the heterogeneity between factories in distributed flow-shop scheduling for the first time. This paper addresses a distributed heterogeneous no-wait flowshop scheduling problem (DHNWFSP) to minimize the makespan, where the factories have differences among them, including the number of machines, machine technology, raw material supply, and transportation conditions. In this problem, the numbers and types of machines in each factory are different, and this means that the jobs have to be processed through different processing paths. To effectively solve this DHNWFSP, a discrete artificial bee colony algorithm (DABC) is proposed. Firstly, to obtain a feasible neighborhood solution, four neighborhood search operators based on the characteristics of this problem are presented to search neighborhoods during the employed bee phase and onlooker bee phase. Then, a new method to accelerate the evaluation of the obtained neighborhood is proposed to reduce the computation time. Moreover, an efficient population update method is designed in the onlooker bee phase. Finally, a variable neighborhood descent (VND) algorithm based on four local-search methods is embedded into the scout bee phase to strengthen the local search ability of the overall algorithm. To validate the performance of the proposed algorithm, a series of numerical experiments are executed for small- and large-scale problems to compare the DABC with some state-of-art algorithms in terms of solving the DHNWFSP. The results show that the proposed DABC obtains the highest-quality solutions.

Characterizing Co-Located Workloads in Alibaba Cloud Datacenters

Workload characteristics are vital for both data center operation and job scheduling in co-located data centers, where online services and batch jobs are deployed on the same production cluster. In this article, a comprehensive analysis is conducted on Alibaba's cluster-trace-v2018 of a production cluster of 4034 machines. The findings and insights are the following: (1) The workload on the production cluster poses a daily cyclical fluctuation, in terms of CPU and disk I/O utilization, and the memory system has become the performance bottleneck of a co-located cluster. (2) Batch jobs including their tasks and derived instances can be approximated as Zipf distribution. However, for all batch jobs with directed acyclic graph dependency, they suffer from co-location with online services since the online services are highly prioritized. (3) The resource usages of containers have similar cyclical fluctuation consistent with the whole cluster, while their memory usages remain approximately constant. (4) The number of batch jobs co-located with online services is dependent on the mispredictions per kilo instructions of online services. In order to guarantee the QoS of online services, when the MPKI of online services rises, the number of batch jobs to be co-located on the same machine should decrease.

Performance Analysis of Batch Jobs in Cloud Environment

The complexity of running batch jobs in cloud environment is increasing day by day due to huge volume of back end data increasing dynamically in different formats, growing number of online or offline requests from various environments or availability of system resources. This paper explores the various behaviors of batch jobs from production cluster and the following observations are made: (i) The number of user jobs submitted at offline are high thereby increasing the wait time and processing time. (ii) Each job submitted with huge volume of data causing increase in queue size, job failure and errors on completion. (iii) Server utilization is unbalanced during the peak time thereby causing the server crash.

Performance-Aware Speculative Resource Oversubscription for Large-Scale Clusters

It is a long-standing challenge to achieve a high degree of resource utilization in cluster scheduling. Resource oversubscription has become a common practice in improving resource utilization and cost reduction. However, current centralized approaches to oversubscription suffer from the issue with resource mismatch and fail to take into account other performance requirements, e.g., tail latency. In this article we present ROSE, a new resource management platform capable of conducting performance-aware resource oversubscription. ROSE allows latency-sensitive long-running applications (LRAs) to co-exist with computation-intensive batch jobs. Instead of waiting for resource allocation to be confirmed by the centralized scheduler, job managers in ROSE can independently request to launch speculative tasks within specific machines according to their suitability for oversubscription. Node agents of those machines can however, avoid any excessive resource oversubscription by means of a mechanism for admission control using multi-resource threshold control and performance-aware resource throttle. Experiments show that in case of mixed co-location of batch jobs and latency-sensitive LRAs, the CPU utilization and the disk utilization can reach 56.34 and 43.49 percent, respectively, but the 95th percentile of read latency in YCSB workloads only increases by 5.4 percent against the case of executing the LRAs alone.

Performance-aware energy-efficient parallel job scheduling in HPC grid using nature-inspired hybrid meta-heuristics

High-Performance Computing (HPC) systems offer massive computation strength to execute large-scale applications. However, the availability of thousands of CPU cores in the HPC Systems has also triggered a significant increase in the associated energy consumption translating to higher energy expenses of system providers and carbon emissions in the environment. Therefore efficient job schedulers, which can trade-off between user-desired performance and conflicting energy-efficiency objectives simultaneously, are the need of the hour and must nowadays. Job scheduling in HPC systems is a known NP-Hard problem for which meta-heuristics may provide a near-to-optimal solution. Cuckoo search (CS) is a well-known robust swarm-intelligence based meta-heuristic, which has been applied extensively in many optimization problems due to the strong searching efficiency and requirement of very few tuning parameters. However, it suffers from the likelihood of trapping in the local minima and lack of solution diversity towards the end of the algorithm. These drawbacks could result in unacceptable results when the CS algorithm applies to the parallel job scheduling problem. To overcome these limitations and improve the searching efficiency of the traditional CS, we have proposed a multi-objective hybrid scheduling algorithm called MOHCSFA to optimally schedule the batch of parallel jobs in HPC Grid. The proposed MOHCSFA policy combines the solution search mechanisms of both Cuckoo Search (CS) and Firefly algorithm (FA) during each generation. Our proposed policy is further integrated with efficient resource allocation (ERA) heuristic to improve job scheduler performance by effectively using multi-site resource allocation. The experiments are conducted on the GridSim simulator and the benchmarking of the proposed algorithm is done using real data-sets extracted from two supercomputing workload logs. The simulation results showed that the proposed MOHCSFA policy outperforms many heuristics and meta-heuristic scheduling policies for different test cases for both performance and energy-efficiency objectives. Specifically, in the case of Unilu-Gaia workloads, the MOHCSFA obtained 5.87–24.05%, 3.46–28.50%, and 7.06–26.76% performance improvement for the makespan, energy consumption and avg. flowtime, respectively over other tested scheduling policies. The statistical tests validated the stability and robustness of the proposed policy over other scheduling policies.

Near-linear-time approximation algorithms for scheduling a batch-processing machine with setups and job rejection

In this paper we study a single batch-processing machine scheduling model. In our model, a set of jobs having different release dates needs to be scheduled onto a single machine that can process a batch of jobs simultaneously at a time. Each batch incurs a fixed setup time and a fixed setup cost. The decision maker may reject some of the jobs by paying penalty cost so as to achieve a tight makespan, but the total rejection penalty cost is required to be no greater than a given value. Our model extends several existing batch-processing machine scheduling models in the literature. We present efficient approximation algorithms with near-linear-time complexities.

Static job scheduling for environments with vertical elasticity

SummaryIn virtualized environments, such as Clouds, allocating a fixed amount of resources to a job a priori, may result in underutilization of the shared host. Meanwhile, vertical elasticity can be adopted to reduce the impact by resizing virtual machines (VMs) dynamically, in conjunction with suspension and/or migration before the host had been overloaded. In order to reduce the number of such events, but still avoid overloading the host, it is important to find an effective initial VM allocation. To achieve this, the scheduler must work in unison with the local host elasticity controllers to reduce interference. This work proposes and evaluates a framework for job scheduling in elastic memory managed virtualized environments. The memory elasticity management in clouds (MEMiC) framework is a two‐tier VM scheduler for batch jobs which attempts to predict the impact caused by competition for the memory of a host in shared Cloud‐like environments for harnessing VM allocation, suspension, and migration. Evaluations show that the MEMiC achieves a reduction of approximately 19% in the total time for jobs execution in comparison to other approaches, by reducing the degree of interference.

Revenue-optimal task scheduling and resource management for IoT batch jobs in mobile edge computing

With the growing prevalence of Internet of Things (IoT) devices and technology, a burgeoning computing paradigm namely mobile edge computing (MEC) is delicately proposed and designed to accommodate the application requirements of IoT scenario. In this paper, we focus on the problems of dynamic task scheduling and resource management in MEC environment, with the specific objective of achieving the optimal revenue earned by edge service providers. While the majority of task scheduling and resource management algorithms are formulated by an integer programming (IP) problem and solved in a dispreferred NP-hard manner, we innovatively investigate the problem structure and identify a favorable property namely totally unimodular constraints. The totally unimodular property further helps to design an equivalent linear programming (LP) problem which can be efficiently and elegantly solved at polynomial computational complexity. In order to evaluate our proposed approach, we conduct simulations based on real-life IoT dataset to verify the effectiveness and efficiency of our approach.

Exact algorithms to minimize makespan on single and parallel batch processing machines

Batch-processing machine scheduling problem is one of the challenging problems in the machine scheduling literature where machines are capable of processing a batch of jobs simultaneously. In this paper, we tackle single and parallel batch processing machine scheduling problems with the objective of minimizing makespan. We propose a reformulation for parallel batch processing machine scheduling, which is based on decomposition in two levels, and an exact algorithm for its solution. To the best of our knowledge, there is no exact algorithm to solve this problem in the literature, except for its formulations solved by off-the-shelf solvers. In the first part of the proposed algorithm, we solve the single batch processing machine problem by a column-and-cut generation algorithm that provides a lower bound for the parallel machine problem. The second part of our proposed algorithm employs a search mechanism to find the minimum makespan for the parallel machine problem, which entails the solution of the reformulation of this problem by column generation at every iteration. The novel aspect of this column generation algorithm is the integration of batch generation and machine schedule generation in a single pricing subproblem. We test the performance of the proposed algorithms on randomly generated instances and show that, on average, they outperform the off-the-shelf solver. The major findings are that the single machine problem provides tight lower and upper bounds for the parallel machine problem, and the proposed algorithm for the parallel machine problem solves more instances to optimality than that for the single machine problem.

The BNLBox Cloud Storage Service

Large scientific data centers have recently begun providing a number of different types of data storage in order to satisfy the various needs of their users. Users with interactive accounts, for example, might want a POSIX interface for easy access to the data from their interactive machines. Grid computing sites, on the other hand, likely need to provide an X509-based storage protocol, like SRM and GridFTP, since the data management system is built upon them. Meanwhile, an experiment producing large amounts of data typically demands a service that provides archival storage for the safe keeping of their unique data. To access these various types of data, users must use specific sets of commands tailored to their respective storage, making access to their data complex and difficult. BNLBox is an attempt to provide a unified and easy to use storage service for all BNL users, to store their important documents, code and data. It is a cloud storage system with an intuitive web interface for novice users. It provides an automated synchronization feature that enables users to upload data to their cloud storage without manual intervention, freeing them to focus on analysis rather than data management software. It provides a POSIX interface for local interactive users, which simplifies data access from batch jobs as well. At the same time, it also provides users with a straightforward mechanism for archiving large data sets for later processing. The storage space can be used for both code and data within the compute job environment. This paper will describe various aspects of the BNLBox storage service.

Optimizing the Cost of Executing Mixed Interactive and Batch Workloads on Transient VMs

Container Orchestration Platforms (COPs), such as Kubernetes, are increasingly used to manage large-scale clusters by automating resource allocation between applications encapsulated in containers. Increasingly, the resources underlying COPs are virtual machines (VMs) dynamically acquired from cloud platforms. COPs may choose from many different types of VMs offered by cloud platforms, which differ in their cost, performance, and availability. While transient VMs cost significantly less than on-demand VMs, platforms may revoke them at any time, causing them to become unavailable. While transient VMs' price is attractive, their unreliability is a problem for COPs designed to support mixed workloads composed of, not only delay-tolerant batch jobs, but also long-lived interactive services with high availability requirements. To address the problem,we design TR-Kubernetes, a COP that optimizes the cost of executing mixed interactive and batch workloads on cloud platforms using transient VMs. To do so, TR-Kubernetes enforces arbitrary availability requirements specified by interactive services despite transient VM unavailability by acquiring many more transient VMs than necessary most of the time, which it then leverages to opportunistically execute batch jobs when excess resources are available. When cloud platforms revoke transient VMs, TR-Kubernetes relies on existing Kubernetes functions to internally revoke resources from batch jobs to maintain interactive services' availability requirements. We show that TR-Kubernetes requires minimal extensions to Kubernetes, and is capable of lowering the cost (by 53%) and improving the availability (99.999%) of a representative interactive/batch workload on Amazon EC2 when using transient compared to on-demand VMs.

Hybrid Metaheuristic Based Offline Parallel Job Scheduling in Heterogeneous Computing Systems

Over two decades, Heterogeneous Computing Systems (HCS) are offering large amount of federated computing resources, spanning across different administrative domains, to compute-intensive user applications. Efficient job schedulers are required to allocate HCS resources to user applications to satisfy system provider and user requirements. Offline scheduling is most popular kind of job scheduling in heterogeneous system, in which jobs are collected in batch and scheduled together. Job scheduling in HCS has become NP-hard problem due to system scale, federated structure and high resource as well as job heterogeneity. Simple queuing and deterministic heuristics have failed to provide optimal solution to NP-hard job scheduling problem. Due to NP-hard nature of job scheduling problem, there is always a scope to propose new scheduling solutions using meta-heuristics. Offline scheduling in HCS has been focused more on scheduling independent sequential tasks viz. Bag-of-tasks or Many-tasks. Offline scheduling of parallel jobs (composed of collaborating tasks with no precedence) in HCS has not gained much attention. In this paper, a novel hybrid multi-objective meta-heuristic known as HCSPSO, which combines the qualities of Cuckoo search (CS) and Particle Swarm Optimization (PSO), has been proposed to schedule batch of parallel jobs in multi-cluster HCS platform. Proposed HCSPSO policy is extensively compared with different heuristics and metaheuristics using different resource configurations and real supercomputing workload logs. Comparative results have showed the dominance of the proposed hybrid scheduling algorithm over other algorithms.