Parallel Jobs Research Articles

Adaptive scheduling of multiprogrammed dynamic-multithreading applications

Modern parallel platforms, such as clouds or servers, are often shared among many different jobs. However, existing parallel programming runtime systems are designed and optimized for running a single parallel job, so it is generally hard to directly use them to schedule multiple parallel jobs without incurring high overhead and inefficiency. In this work, we develop AMCilk (Adaptive Multiprogrammed Cilk), a novel runtime system framework, designed to support multiprogrammed parallel workloads. AMCilk has client-server architecture where users can dynamically submit parallel jobs to the system. AMCilk has a single runtime system that runs these jobs while dynamically reallocating cores, last-level cache, and memory bandwidth among these jobs according to the scheduling policy. AMCilk exposes the interface to the system designer, which allows the designer to easily build different scheduling policies meeting the requirements of various application scenarios and performance metrics, while AMCilk transparently (to designers) enforces the scheduling policy. AMCilk also enables its use in cloud environment where other processes may be sharing the system with AMCilk. In this scenario, an external scheduler can change the resource availability for AMCilk and AMCilk seamlessly adapts to these changes. The primary feature of AMCilk is the low-overhead and responsive preemption mechanism that allows fast reallocation of cores between jobs. Our empirical evaluation indicates that AMCilk incurs small overheads and provides significant benefits on application-specific criteria for a set of 4 practical applications due to its fast and low-overhead core reallocation mechanism.

LINGUISTIC AND LITERARY ANALYSIS OF BUNIN’S SHORT STORY «DARK ALLEYS»

Problem statement. The article is devoted to the linguistic and linguistic-and-literary analysis of the short story «Dark alleys» by I.A. Bunin aiming at mastering the methods of such type of analysis of an art work. The article aims to consider the linguistic structure of the story in terms of vocabulary, translation, word formation, morphology and onomastics. Special attention is paid to the linguistic basis and lexical base of this story. Results. This paper highlights and describes the features of outdated, non-equivalent, borrowed and calque words, which are used to express the historical space of the story «Dark alleys» by I.A. Bunin. The author's means of information coding concerning important historical events are analyzed. It is found out by what language means the author expresses the ideological and emotional sides of his story and linguistically significant information about the language units used to reflect the real character in the story is revealed. Conclusion. Analyzing the short story of I.A. Bunin, we came to the conclusion that the onomastic material of the story plays a specific role: the name of the main character was chosen by the author in honor of the Russian officer — Nikolai Alekseevich Kireev, and the name of Job in parallel to the "unhappiness" of the author for his whim. As a result, we prove that the text of the story cannot be covered with any quick reading. This story was written using the tools of different language sections: vocabulary, word formation, morphology and onomastics. The story should be read carefully, because placing it in the collection «Dark alleys», the author thus emphasizes that he describes not only love, but also history, and not one, but two at once.

ROA-CONS: Raccoon Optimization for Job Scheduling

High-performance computing comprises thousands of processing powers in order to deliver higher performance computation than a typical desktop computer or workstation in order to solve large problems in science, engineering, or business. The scheduling of these machines has an important impact on their performance. HPC’s job scheduling is intended to develop an operational strategy which utilises resources efficiently and avoids delays. An optimised schedule results in greater efficiency of the parallel machine. In addition, processes and network heterogeneity is another difficulty for the scheduling algorithm. Another problem for parallel job scheduling is user fairness. One of the issues in this field of study is providing a balanced schedule that enhances efficiency and user fairness. ROA-CONS is a new job scheduling method proposed in this paper. It describes a new scheduling approach, which is a combination of an updated conservative backfilling approach further optimised by the raccoon optimisation algorithm. This algorithm also proposes a technique of selection that combines job waiting and response time optimisation with user fairness. It contributes to the development of a symmetrical schedule that increases user satisfaction and performance. In comparison with other well-known job scheduling algorithms, the simulation assesses the effectiveness of the proposed method. The results demonstrate that the proposed strategy offers improved schedules that reduce the overall system’s job waiting and response times.

The case for phase-aware scheduling of parallelizable jobs

Parallelizable jobs typically consist of multiple phases of computation, where the job is more parallelizable in some phases and less parallelizable in others. For example, in a database, a query may consist of a highly parallelizable table scan, followed by a less parallelizable table join. In the past, this phase-varying parallelizability was summarized by a single sub-linear speedup curve that measures a job’s average parallelizability over its entire lifetime. Today, however, modern systems have fine-grained knowledge of the exact phase each job is in at every moment in time. Unfortunately, these systems do not fully leverage this real-time feedback when scheduling parallelizable jobs. Theory has failed to produce practical phase-aware scheduling policies, and thus scheduling in current systems is largely heuristic.A phase-aware scheduling policy must decide, given its knowledge of each job’s current phase, how many servers or cores to allocate to each job in the system at every moment in time. This paper provides the first stochastic model of a system processing parallelizable jobs composed of phases. Using our model, we derive an optimal phase-aware scheduling policy that minimizes the mean response time across jobs. Our provably optimal policy, Inelastic-First (IF), gives strict priority to jobs that are currently in less parallelizable phases. We validate our theoretical results using a simulation of a database running queries from the Star Schema Benchmark. We compare IF to a range of policies from both systems and theory, and show that IF reduces mean response time by up to a factor of 3.

Optimizing resources to mitigate stragglers through virtualization in run time

Modern computing systems are generally enormous in scale, consisting of hundreds to thousands of heterogeneous machine nodes, to meet rising demands for Cloud services. MapReduce and other parallel computing frameworks are frequently used on such cluster architecture to offer consumers dependable and timely services. However, Cloud workloads’ complex features, such as multi-dimensional resource requirements and dynamically changing system settings, such as dynamic node performance, are posing new difficulties for providers in terms of both customer experience and system efficiency. The straggler problem occurs when a small subset of parallelized jobs takes an excessively long time to execute in contrast to their siblings, resulting in a delayed job response and the possibility of late-timing failure. Speculative execution is the state-of-the-art method to straggler mitigation. Speculative execution has been used in numerous real-world systems with a variety of implementation improvements, but the results of this thesis’ research demonstrate that it is typically wasteful. The failure rate of speculative execution might be as high as 71 percent, according to different data center production trace logs. Straggler mitigation is a difficult task in and of itself: 1) stragglers may have varying degrees of severity in parallel job execution; 2) whether a task should be considered a straggler is highly subjective, depending on various application and system conditions; 3) the efficiency of speculative execution would be improved if dynamic node quality could be adequately modeled and predicted; 4) Other sorts of stragglers, such as those generated by data skews, are beyond speculative execution’s capabilities.

Flexible job-shop scheduling problem with unrelated parallel machines and resources-dependent processing times: a tabu search algorithm

ABSTRACT Flexible job shop scheduling problem is one of the most important topics in production management and is one of the most complex topics in combinatorial optimization. This problem is a generalization of job shop and parallel machines scheduling problem. Since the efficient allocation of resources can improve the performance of manufacturing, here, to reduce the processing time of jobs, additional resources are assigned to machines. In fact, in this paper, the effect of flexible resources in the flexible job shop scheduling problem with unrelated parallel machines and sequence-dependent setup time is investigated. Also, by presenting a mixed-integer linear programming model, an attempt has been made to minimize the costs of makespan, total weighted tardiness, delivery time and inventory. After solving this model by the GAMS, due to the NP-hardness of the problem, a tabu search (TS) algorithm is utilized for large-size instances. Finally, the obtained results are compared with the genetic algorithm (GA). To verify the statistical validity of the computational experiments and confirm which the best algorithm between the TS algorithm and GA is, a Kruskal–Wallis test is used. The results show that the TS algorithm is better than the GA.

Achieving Zero Asymptotic Queueing Delay for Parallel Jobs

Zero queueing delay is highly desirable in large-scale computing systems. Existing work has shown that it can be asymptotically achieved by using the celebrated Power-of-d-choices (Pod) policy with a probe overhead d = Ω(log N/1-λ), and it is impossible when d = O(1/1-λ), where N is the number of servers and λ is the load of the system. However, these results are based on the model where each job is an indivisible unit, which does not capture the parallel structure of jobs in today's predominant parallel computing paradigm. This paper considers a model where each job consists of a batch of parallel tasks. Under this model, we propose a new notion of zero (asymptotic) queueing delay that requires the job delay under a policy to approach the job delay given by the max of its tasks' service times, i.e., the job delay assuming its tasks entered service right upon arrival. This notion quantifies the effect of queueing on a job level for jobs consisting of multiple tasks, and thus deviates from the conventional zero queueing delay for single-task jobs in the literature. We show that zero queueing delay for parallel jobs can be achieved using the batch-filling policy (a variant of the celebrated Pod policy) with a probe overhead d = Ω(1/(1-λ)log k) in the sub-Halfin-Whitt heavy-traffic regime, where k is the number of tasks in each job and k properly scales with N (the number of servers). This result demonstrates that for parallel jobs, zero queueing delay can be achieved with a smaller probe overhead. We also establish an impossibility result: we show that zero queueing delay cannot be achieved if d = (o(log N/log k)). Simulation results are provided to demonstrate the consistency between numerical results and theoretical results under reasonable settings, and to investigate gaps in the theoretical analysis.

Triggerflow: Trigger-based orchestration of serverless workflows

As more applications are being moved to the Cloud thanks to serverless computing, it is increasingly necessary to support the native life cycle execution of those applications in the data center.But existing cloud orchestration systems either focus on short-running workflows (like IBM Composer or Amazon Step Functions Express Workflows) or impose considerable overheads for synchronizing massively parallel jobs (Azure Durable Functions, Amazon Step Functions). None of them are open systems enabling extensible interception and optimization of custom workflows.We present Triggerflow: an extensible Trigger-based Orchestration architecture for serverless workflows. We demonstrate that Triggerflow is a novel serverless building block capable of constructing different reactive orchestrators (State Machines, Directed Acyclic Graphs, Workflow as code, Federated Learning orchestrator). We also validate that it can support high-volume event processing workloads, auto-scale on demand with scale down to zero when not used, and transparently guarantee fault tolerance and efficient resource usage when orchestrating long running scientific workflows.

Analysis of the Binding Sites on BAX and the Mechanism of BAX Activators through Extensive Molecular Dynamics Simulations.

The BAX protein is a pro-apoptotic member of the Bcl-2 family, which triggers apoptosis by causing permeabilization of the mitochondrial outer membrane. However, the activation mechanism of BAX is far from being understood. Although a few small-molecule BAX activators have been reported in the literature, their crystal structures in complex with BAX have not been resolved. So far, their binding modes were modeled at most by simple molecular docking efforts. Lack of an in-depth understanding of the activation mechanism of BAX hinders the development of more effective BAX activators. In this work, we employed cosolvent molecular dynamics simulation to detect the potential binding sites on the surface of BAX and performed a long-time molecular dynamics simulation (50 μs in total) to derive the possible binding modes of three BAX activators (i.e., BAM7, BTC-8, and BTSA1) reported in the literature. Our results indicate that the trigger, S184, and vMIA sites are the three major binding sites on the full-length BAX structure. Moreover, the canonical hydrophobic groove is clearly detected on the α9-truncated BAX structure, which is consistent with the outcomes of relevant experimental studies. Interestingly, it is observed that solvent probes bind to the trigger bottom pocket more stably than the PPI trigger site. Each activator was subjected to unbiased molecular dynamics simulations started at the three major binding sites in five parallel jobs. Our MD results indicate that all three activators tend to stay at the trigger site with favorable MM-GB/SA binding energies. BAM7 and BTSA1 can enter the trigger bottom pocket and thereby enhance the movement of the α1-α2 loop, which may be a key factor at the early stage of BAX activation. Our molecular modeling results may provide useful guidance for designing smart biological experiments to further explore BAX activation and directing structure-based efforts toward discovering more effective BAX activators.

Optimal preventive replacement policies for a system with discrete scheduled times

In reality, most operating systems are repaired or replaced when they have failed. For such systems, it would be reliable to maintain them in a control-limit consideration of repairs or operations. This article addresses a system scheduling policy for preventive replacement with discrete control limits of minimal repair and random work. An operating system which works at random times for multiple jobs (N tandem jobs or N parallel jobs) is considered. The system subjects to two failure mechanisms and maintained imperfectly: type-I (minor) failure rectified by a minimal repair and type-II (catastrophic) failure removed by a corrective replacement. To control the deterioration process, the preventive replacement is planned to undergo before catastrophic failure at number n of minimal repairs or the completion of N multiple jobs. The objective is to determine the optimal discrete scheduling replacement parameters (n* or N*) that minimizes the mean cost rate function in a finite time horizon. Model analysis relating to the existence and uniqueness of optimal solutions is provided. A Numerical example is presented for illustrative purposes.

Approximate and robust bounded job start scheduling for Royal Mail delivery offices

Motivated by mail delivery scheduling problems arising in Royal Mail, we study a generalization of the fundamental makespan scheduling P||C_{max } problem which we call the bounded job start scheduling problem. Given a set of jobs, each specified by an integer processing time p_j, that have to be executed non-preemptively by a set of m parallel identical machines, the objective is to compute a minimum makespan schedule subject to an upper bound gle m on the number of jobs that may simultaneously begin per unit of time. With perfect input knowledge, we show that Longest Processing Time First (LPT) algorithm is tightly 2-approximate. After proving that the problem is strongly {mathcal {N}}{mathcal {P}}-hard even when g=1, we elaborate on improving the 2-approximation ratio for this case. We distinguish the classes of long and short instances satisfying p_jge m and p_j<m, respectively, for each job j. We show that LPT is 5/3-approximate for the former and optimal for the latter. Then, we explore the idea of scheduling long jobs in parallel with short jobs to obtain tightly satisfied packing and bounded job start constraints. For a broad family of instances excluding degenerate instances with many very long jobs, we derive a 1.985-approximation ratio. For general instances, we require machine augmentation to obtain better than 2-approximate schedules. In the presence of uncertain job processing times, we exploit machine augmentation and lexicographic optimization, which is useful for P||C_{max } under uncertainty, to propose a two-stage robust optimization approach for bounded job start scheduling under uncertainty aiming in a low number of used machines. Given a collection of schedules of makespan le D, this approach allows distinguishing which are the more robust. We substantiate both the heuristics and our recovery approach numerically using Royal Mail data. We show that for the Royal Mail application, machine augmentation, i.e., short-term van rental, is especially relevant.

HeSRPT

Modern data centers serve workloads which can exploit parallelism. When a job parallelizes across multiple servers it completes more quickly. However, it is unclear how to share a limited number of servers between many parallelizable jobs. In this paper we consider a typical scenario where a data center composed of N servers will be tasked with completing a set of M parallelizable jobs. Typically, M is much smaller than N. In our scenario, each job consists of some amount of inherent work which we refer to as a job's size. We assume that job sizes are known up front to the system, and each job can utilize any number of servers at any moment in time. These assumptions are reasonable for many parallelizable workloads such as training neural networks using TensorFlow [2]. Our goal in this paper is to allocate servers to jobs so as to minimize the mean slowdown across all jobs, where the slowdown of a job is the job's completion time divided by its running time if given exclusive access to all N servers. Slowdown measures how a job was interfered with by other jobs in the system, and is often the metric of interest in the theoretical parallel scheduling literature (where it is also called stretch), as well as the HPC community (where it is called expansion factor).

Market-oriented online bi-objective service scheduling for pleasingly parallel jobs with variable resources in cloud environments

In this paper, we study the market-oriented online bi-objective service scheduling problem for pleasingly parallel jobs with variable resources in cloud environments, from the perspective of SaaS (Software-as-as-Service) providers who provide job-execution services. The main process of scheduling SaaS services in clouds is: a SaaS provider purchases cloud instances from IaaS providers to schedule end users’ jobs and charges users accordingly. This problem has several particular features, such as the job-oriented end users, the pleasingly parallel jobs with soft deadline constraints, the online settings, and the variable numbers of resources. For maximizing both the revenue and the user satisfaction rate, we design an online algorithm for SaaS providers to optimally purchase IaaS instances and schedule pleasingly parallel jobs. The proposed algorithm can achieve competitive objectives in polynomial run-time. The theoretical analysis and simulations based on real-world Google job traces as well as synthetic datasets validate the effectiveness and efficiency of our algorithm.

Deployment of Elastic Virtual Hybrid Clusters Across Cloud Sites

Virtual clusters are widely used computing platforms than can be deployed in multiple cloud platforms. The ability to dynamically grow and shrink the number of nodes has paved the way for customised elastic computing both for High Performance Computing and High Throughput Computing workloads. However, elasticity is typically restricted to a single cloud site, thus hindering the ability to provision computational resources from multiple geographically distributed cloud sites. To this aim, this paper introduces an architecture of open-source components that coherently deploy a virtual elastic cluster across multiple cloud sites to perform large-scale computing. These hybrid virtual elastic clusters are automatically deployed and configured using an Infrastructure as Code (IaC) approach on a distributed hybrid testbed that spans different organizations, including on-premises and public clouds, supporting automated tunneling of communications across the cluster nodes with advanced VPN topologies. The results indicate that cluster-based computing of embarrassingly parallel jobs can benefit from hybrid virtual clusters that aggregate computing resources from multiple cloud back-ends and bring them together into a dedicated, albeit virtual network. The work presented in this article has been partially funded by the European Union's (EU) Horizon 2020 research project DEEP Hybrid-DataCloud (grant agreement No 777435).

Implementation of human whole genome sequencing data analysis: A containerized framework for sustained and enhanced throughput

Abstract Whole Genome Sequencing (WGS) provides information for each base of the entire 3.2 billion base pairs of the diploid human genome. Therefore, WGS plays an important role in identifying genetic variations for populations and understanding disease signatures in cohort studies or cases with rare genetic disorders. Nonetheless, discoveries from high throughput WGS are dependent on efficient processing, analyzing, and storing this enormous amount of genomic sequencing data, often in the scale of petabytes. Although there has been a significant reduction in genome sequencing costs in recent years, high-performance computation costs have not decreased in a directly proportional fashion. The objective of the present work is to develop a Docker-based container method for human whole genome sequencing data processing and analysis for detecting genetic variations from paired end WGS short reads. Our method provides an approach to simultaneously process multiple genomes within a single compute system while guaranteeing sustained and stable handling of the memory requirements for the genomic data processing and ensuring no unwanted termination of the currently running parallel jobs. This method also achieves a 40 % reduction in execution time. To encourage widespread adoption and ease of WGS analysis, our containerized pipeline will be made publicly available. We have tested this approach for human genome data from Illumina WGS platforms and report the benchmark metrics in two different workstation environments in this communication. Compared to truth sets, our approach calls variants with 99 % precision and recall.

Preemptive Parallel Job Scheduling for Heterogeneous Systems Supporting Urgent Computing

Dedicated infrastructures are commonly used for urgent computations. However, using dedicated resources is not always affordable due to budget constraints. As a result, utilizing shared infrastructures becomes an alternative solution for urgent computations. Since the infrastructures are meant to serve many users, the urgent jobs may arrive when regular jobs are using the necessary resources. In such a case, it is necessary to preempt the regular jobs so that urgent jobs can be executed immediately. Most conventional methods for job scheduling have focused on reducing the response times and waiting times of all jobs. However, these methods can delay urgent jobs and hinder them from being completed within a stipulated deadline. Furthermore, in heterogeneous systems with coprocessors, preemption becomes more difficult because coprocessors rely on several system software functionalities provided by the host processor. In this paper, we propose a parallel job scheduling method to effectively use shared heterogeneous systems for urgent computations. Our method employs an in-memory process swapping mechanism to preempt jobs running on the coprocessor devices. The results of our simulations show that our method can achieve a significant reduction in the response time and slowdown of regular jobs without substantial delays of urgent jobs.

Resilient Scheduling Heuristics for Rigid Parallel Jobs

This paper focuses on the resilient scheduling of parallel jobs on high-performance computing (HPC) platforms to minimize the overall completion time, or the makespan. We revisit the classical problem while assuming that jobs are subject to failures caused by transient or silent errors, and hence may need to be re-executed each time they fail to complete successfully. This work generalizes the classical framework where jobs are known offline and do not fail: in this framework, list scheduling that gives priority to the longest jobs is known to be a 3-approximation when imposing to use shelves, and a 2-approximation without this restriction. We show that when jobs can fail, using shelves can be arbitrarily bad, but unrestricted list scheduling remains a 2-approximation. The paper focuses on the design of several heuristics, some list-based and some shelf-based, along with different priority rules and backfilling strategies. We assess and compare their performance through an extensive set of simulations using both synthetic jobs and log traces from the Mira supercomputer.

Research on optimization of setup time for product in the multi-product production system

Consumers' consumption habits are more and more personalized and diversified, which makes the multi-product production system has been applied extensively in the factory worldwide. This brings a difficult problem to a large number of manufacturing enterprises: how to optimize the setup time of the product to achieve the purpose of improving the time efficiency. Based on this problem, this paper proposes the TCP technology for the optimization of setup time, that is, using the Times Series model, the Clustering Algorithm, and the Parallel Job technology in the Single Minute Exchange of Die (SMED), to form an application framework focusing on optimizing the product setup time. The validity of the technology is verified by a case study. This paper enriches the research field of setup time optimization, production planning, and the application of the Clustering Algorithm in the multi-product production system. It provides a new way for manufacturing enterprises to pursue an excellent efficiency of product setup time.

Achieving Zero Asymptotic Queueing Delay for Parallel Jobs

Zero queueing delay is highly desirable in large-scale computing systems. Existing work has shown that it can be asymptotically achieved by using the celebrated Power-of-d-choices (pod) policy with a probe overhead $d = ømegałeft(\fracłog N 1-łambda \right)$, and it is impossible when $d = Ołeft(\frac1 1-łambda \right)$, where N is the number of servers and $łambda$ is the load of the system. However, these results are based on the model where each job is an indivisible unit, which does not capture the parallel structure of jobs in today's predominant parallel computing paradigm. This paper thus considers a model where each job consists of a batch of parallel tasks. Under this model, we propose a new notion of zero (asymptotic) queueing delay that requires the job delay under a policy to approach the job delay given by the max of its tasks' service times, i.e., the job delay assuming its tasks entered service right upon arrival. This notion quantifies the effect of queueing on a job level for jobs consisting of multiple tasks, and thus deviates from the conventional zero queueing delay for single-task jobs in the literature. We show that zero queueing delay for parallel jobs can be achieved using the batch-filling policy (a variant of the celebrated \pod\ policy) with a probe overhead $d = ømegałeft(\frac1 (1-łambda)łog k \right)$ in the sub-Halfin-Whitt heavy-traffic regime, where k is the number of tasks in each job and k properly scales with N (the number of servers). This result demonstrates that for parallel jobs, zero queueing delay can be achieved with a smaller probe overhead. We also establish an impossibility result: we show that zero queueing delay cannot be achieved if $d = \expłeft(ołeft(\fracłog N łog k \right) \right)$. Simulation results are provided to demonstrate the consistency between numerical results and theoretical results under reasonable settings, and to investigate gaps in the theoretical analysis.

An architecture for scheduling with the capability of minimum share to heterogeneous Hadoop systems

Job scheduling in Hadoop has been thus far investigated in several studies. However, some challenges including minimum share (min-share), heterogeneous cluster, execution time estimation, and scheduling program size facing Hadoop clusters have received less attention. Accordingly, one of the most important algorithms with regard to min-share is that presented by Facebook Inc., i.e., FAIR scheduler, based on its own needs, in which an equal min-share has been considered for users. In this article, an attempt has been made to make the proposed method superior to existing methods through automation and configuration, performance optimization, fairness and data locality. A high-level architectural model is designed. Then a scheduler is defined on this architectural model. The provided scheduler contains four components. Three components schedule jobs and one component distributes the data for each job among the nodes. The given scheduler will be capable of being executed on heterogeneous Hadoop clusters and running jobs in parallel, in which disparate min-shares can be assigned to each job or user. Moreover, an approach is presented for each problem associated with min-share, cluster heterogeneity, execution time estimation, and scheduler program size. These approaches can be also utilized on its own to improve the performance of other scheduling algorithms. The scheduler presented in this paper showed acceptable performance compared with First-In, First-Out (FIFO), and FAIR schedulers.