Speculative Container Scheduling for Deep Learning Applications in a Kubernetes Cluster

Mao, Ying; Fu, Yuqi; Zheng, Wenjia; Cheng, Long; Liu, Qingzhi; Tao, Dingwen

doi:10.1109/jsyst.2021.3129974

Cited by 17 publications

(8 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It gains popularity due to its enormous capacity of handling big data processing and providing the best developing support platform and storage. However, scheduling these big data workloads into a cloudhosted cluster can be challenging because the workload can be CPU-intensive and networkintensive [3].…”

Section: Introductionmentioning

confidence: 99%

Scheduling of Jobs Allocation for Apache Spark Using Kubernetes for Efficient Execution of Big Data Application

2023

IJIES

View full text Add to dashboard Cite

The use of cloud services is in high demand due to their high storage and computing capacity. Apache spark provides an open deployment framework for data storage and computation using cluster computing. The specified spark core scheduler uses FIFO to manage job execution in batches. However, it may not be suitable for large-scale clusters due to the unevenness in managing resource allocation between different types of applications. Because of this, most of the work executors are still underutilized and resources are wasted in all the pods, leading to cost inefficiencies. Using apache spark on kubernetes to run cloud applications will ensure rapid resource management for workload execution. Incoming workloads vary widely across applications, so it is critical to manage workload allocation to ensure QoS and cost efficiency. This paper proposes a job scheduling mechanism (JSM) for apache spark on kubernetes to dynamically schedule job allocation for the efficient execution of various big data applications. The JSM process predicts the cluster load and suggests relocation of workloads to efficiently distribute the workload to the lower loaded pods in a standard cluster to optimize cost performance. It identifies the upcoming workload of a job and determines the best-fit pod and aims to reduce the usage of CPU and memory which result in enhancing cost efficiency. The JSM's effective management of job allocation and migration among the underload pods preserves the resource and enhances cost efficiency. Experimental settings are configured to evaluate cluster resources with benchmark statistics output for application job execution. The outcome results of cost, job performance, and scheduling overhead show improved cost efficiency for job execution. The comparison with the existing scheduler with varying request load shows an improvisation of 2% in cost efficiency and 3% lower scheduling overhead.

show abstract

Section: Introductionmentioning

confidence: 99%

Scheduling of Jobs Allocation for Apache Spark Using Kubernetes for Efficient Execution of Big Data Application

2023

IJIES

View full text Add to dashboard Cite

show abstract

“…In the past decade, remarkable progress has been achieved on top of advanced computing systems with various applications. At the backend side, these applications are powered by big data processing frameworks and cloud-optimized systems [1]- [5]. While these modern computing systems, they still requires significant computational power and network bandwidth to process a large amount of data.…”

Section: Introductionmentioning

confidence: 99%

“…We consider two types of search scenarios, (1) The values in the data set is unique; (2) There are duplicates in the data set.…”

mentioning

confidence: 99%

Iterative Qubits Management for Quantum Search

Mu¹

2022

Preprint

View full text Add to dashboard Cite

Recent advances in quantum computing systems attract tremendous attention. Commercial companies, such as IBM, Amazon, and IonQ, have started to provide access to noisy intermediate-scale quantum computers. Researchers and entrepreneurs attempt to deploy their applications that aim to achieve a quantum speedup. Grover’s algorithm and quantum phase estimation are the foundations of many applications with the potential for such a speedup. While these algorithms, in theory, obtain marvelous performance, deploying them on existing quantum devices is a challenging task. For example, quantum phase estimation requires extra qubits and a large number of controlled operations, which are impractical due to low-qubit and noisy hardware. To fully utilize the limited onboard qubits, we develop a distributed application with a key-value data structure based on Grover’s algorithm called IQuCS . Consider a database with duplicates. By encoding each element to a binary type with a unique key and forming a key-value pair, we can count the number of occurrences of each element in the database based on quantum computing. We have optimized the operation process by filtering data points to make it more efficient. To determine the effect of this optimization, we evaluate it with datasets of different sizes and with different numbers of duplicates. With the assistance of classical computers, IQuCS can reduce the problem set for each query. Due to this reduction, IQuCS requires fewer qubits. Through the iterative management, IQuCS achieves a reduction of qubit virtualized consumption, up to 66.2%, with reasonable accuracy.

show abstract

“…In the past decade, remarkable progress has been achieved on top of advanced computing systems with various applications. At the backend side, these applications are powered by big data processing frameworks and cloud-optimized systems [1]- [3]. While these modern computing systems, they still requires significant computational power and network bandwidth to process a large amount of data.…”

Section: Introductionmentioning

confidence: 99%

Iterative Qubits Management for Quantum Index Searching in a Hybrid System

Mao

Long

et al. 2022

2022 IEEE International Performance, Computing, and Communications Conference (IPCCC)

Self Cite

View full text Add to dashboard Cite

Recent advances in quantum computing systems attract tremendous attention. Commercial companies, such as IBM, Amazon, and IonQ, have started to provide access to noisy intermediate-scale quantum computers. Researchers and entrepreneurs attempt to deploy their applications that aim to achieve a quantum speedup. Grover's algorithm and quantum phase estimation are the foundations of many applications with the potential for such a speedup. While these algorithms, in theory, obtain marvelous performance, deploying them on existing quantum devices is a challenging task. For example, quantum phase estimation requires extra qubits and a large number of controlled operations, which are impractical due to low-qubit and noisy hardware. To fully utilize the limited onboard qubits, we propose IQuCS, which aims at index searching and counting in a quantum-classical hybrid system. IQuCS is based on Grover's algorithm. From the problem size perspective, it analyzes results and tries to filter out unlikely data points iteratively. A reduced data set is fed to the quantum computer in the next iteration. With a reduction in the problem size, IQuCS requires fewer qubits iteratively, which provides the potential for a shared computing environment. We implement IQuCS with Qiskit and conduct intensive experiments. The results demonstrate that it reduces qubits consumption by up to 66.2%.

show abstract

Speculative Container Scheduling for Deep Learning Applications in a Kubernetes Cluster

Cited by 17 publications

References 16 publications

Scheduling of Jobs Allocation for Apache Spark Using Kubernetes for Efficient Execution of Big Data Application

Scheduling of Jobs Allocation for Apache Spark Using Kubernetes for Efficient Execution of Big Data Application

Iterative Qubits Management for Quantum Search

Iterative Qubits Management for Quantum Index Searching in a Hybrid System

Contact Info

Product

Resources

About