Mashup

Roy, Rohan Basu; Patel, Tirthak; Gadepally, Vijay; Tiwari, Devesh

doi:10.1145/3503221.3508407

Cited by 27 publications

(3 citation statements)

References 83 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead, serverless computing executes tasks with lightweight containers, thus allowing fine-grained resource provisioning with instant function launch/release, which charges users by the amount of resources (e.g., CPU/GPU and memory) only in actual execution (e.g., second). Due to the unique features, serverless computing is particularly appealing for tasks that require elasticity and high concurrency, such as scientific computing (Chard et al 2020;Roy et al 2022) and distributed training (Wang, Niu, and Li 2019; Guo et al 2022;Thorpe et al 2021;Yu et al 2021Yu et al , 2022.…”

Section: Serverless Drl Trainingmentioning

confidence: 99%

“…Unlike physical clusters and traditional cloud computing that require tedious configuration, serverless computing packages and executes tasks (e.g., DRL actors and learner) as functions with instant toggling (i.e., sub-second level) and auto-scaling. Thus, serverless computing has been widely deployed to serve computation-intensive applications, such as deep learning (Ali et al 2020;Carreira et al 2019;Wang, Niu, and Li 2019;Yu et al 2021Yu et al , 2022 and scientific computing (Chard et al 2020;Roy et al 2022). Fig.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Cheaper and Faster: Distributed Deep Reinforcement Learning with Serverless Computing

Yu,

Li,

Hua

et al. 2024

AAAI

View full text Add to dashboard Cite

Deep reinforcement learning (DRL) has gained immense success in many applications, including gaming AI, robotics, and system scheduling. Distributed algorithms and architectures have been vastly proposed (e.g., actor-learner architecture) to accelerate DRL training with large-scale server-based clusters. However, training on-policy algorithms with the actor-learner architecture unavoidably induces resource wasting due to synchronization between learners and actors, thus resulting in significantly extra billing. As a promising alternative, serverless computing naturally fits on-policy synchronization and alleviates resource wasting in distributed DRL training with pay-as-you-go pricing. Yet, none has leveraged serverless computing to facilitate DRL training. This paper proposes MinionsRL, the first serverless distributed DRL training framework that aims to accelerate DRL training- and cost-efficiency with dynamic actor scaling. We prototype MinionsRL on top of Microsoft Azure Container Instances and evaluate it with popular DRL tasks from OpenAI Gym. Extensive experiments show that MinionsRL reduces total training time by up to 52% and training cost by 86% compared to latest solutions.

show abstract

Section: Serverless Drl Trainingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Cheaper and Faster: Distributed Deep Reinforcement Learning with Serverless Computing

Yu,

Li,

Hua

et al. 2024

AAAI

View full text Add to dashboard Cite

show abstract

“…No entanto, parte dessa complexidade não é comunicada do WMS para o gerenciador de recursos, levando a decisões de escalonamento não-ótimas [Lehmann et al 2023]. Além disso, ao escalonar tais aplicações em clusters de HPC com alocação estática de recursos, é comum observar tanto subutilização quanto superutilização desses recursos [Roy et al 2022b, Roy et al 2022a].…”

Section: Motivaçãounclassified

Escalonamento com Consciência Energética para Fluxos de Trabalho Científicos sem Servidor: Uma Abordagem de Aprendizado de Máquina

Rosa,

Goldman

2024

Anais Da XV Escola Regional De Alto Desempenho De São Paulo (ERAD-SP 2024)

View full text Add to dashboard Cite

Este artigo propõe abordar os desafios de eficiência energética e escalonamento de fluxos de trabalho científicos em ambientes de computação sem servidor. Integrando técnicas de aprendizado de máquina e simulação, a pesquisa visa preencher lacunas entre eficiência energética e escalonamento sem servidor. A metodologia inclui coleta de dados históricos, previsão de consumo energético por meio de aprendizado de máquina e desenvolvimento de políticas de escalonamento com redes neurais profundas. O projeto também envolve adaptação de sistemas de gerenciamento de fluxo de trabalho e validação em ambientes reais, visando oferecer soluções viáveis para os desafios atuais em HPC.

show abstract

Scientific workflow execution in the cloud using a dynamic runtime model

Erbel

Grabowski

2023

Softw Syst Model

View full text Add to dashboard Cite

To explain specific phenomena, scientists perform a sequence of tasks, e.g., to gather, analyze and interpret data, forming a scientific workflow. Depending on the complexity of the workflow, scientists require access to various kinds of tools, applications and infrastructures for individual tasks. Current approaches are often limited to managing these resources at design time, requiring the scientist to preemptively set up applications essential for their workflow. Therefore, a dynamic provisioning and configuration of computing resources are required that fulfills these needs at runtime. In this paper, we present a dynamic runtime model that couples workflow tasks with their individual applications and infrastructure requirements. This runtime model is used as a knowledge base by a model-driven workflow execution engine orchestrating the sequence of tasks and their infrastructure. We exhibit that the simplicity of the runtime model supports the creation of highly tailored infrastructures, the integration of self-developed applications, as well as a human-in-the-loop allowing scientists to monitor and interact with the workflow at runtime. To tackle the heterogeneity of cloud provider interfaces, we implement the workflow runtime model by extending the Open Cloud Computing Interface cloud standard, which provides an extensible data model as well as a uniform interface to manage cloud resources. We demonstrate the applicability of our approach using three case studies and discuss the benefits of the runtime model from a user and system perspective.

show abstract

Mashup

Cited by 27 publications

References 83 publications

Cheaper and Faster: Distributed Deep Reinforcement Learning with Serverless Computing

Cheaper and Faster: Distributed Deep Reinforcement Learning with Serverless Computing

Escalonamento com Consciência Energética para Fluxos de Trabalho Científicos sem Servidor: Uma Abordagem de Aprendizado de Máquina

Scientific workflow execution in the cloud using a dynamic runtime model

Contact Info

Product

Resources

About