Switches for HIRE: resource scheduling for data center in-network computing

Blöcher, Marcel; Wang, Lin; Eugster, Patrick; Schmidt, Max

doi:10.1145/3445814.3446760

Cited by 20 publications

(20 citation statements)

References 102 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our survey proposes a different taxonomy which is categorized based on in-network computing application as well as specialized involved topics. Furthermore, there are considerable amount of in-network computing papers that have not been reviewed by [22]: For example, papers in the scopes of recent technologies e.g., 5G/6G, NFV and edge computing (Please see Section VII of our survey for the wide range of papers we covered), also papers in other scopes such as [23], [24], [25], [26], [27], [28], [29] etc. In addition, we have evaluated and compared the studies from the aspects of novel methodology/performance/application-related criteria which have not been considered in [22].…”

Section: B Existing Relevant Surveys and Tutorialsmentioning

confidence: 99%

A Survey on In-Network Computing: Programmable Data Plane and Technology Specific Applications

Kianpisheh

Taleb

2023

IEEE Commun. Surv. Tutorials

View full text Add to dashboard Cite

In comparison with cloud computing, edge computing offers processing at locations closer to end devices and reduces the user experienced latency. The new recent paradigm of innetwork computing employs programmable network elements to compute on the path and prior to traffic reaching the edge or cloud servers. It advances common edge/cloud server based computing through proposing line rate processing capabilities at closer locations to the end devices. This paper discusses use cases, enabler technologies and protocols for in-network computing. According to our study, considering programmable data plane as an enabler technology, potential in-network computing applications are in-network analytics, in-network caching, innetwork security, and in-network coordination. There are also technology specific applications of in-network computing in the scopes of cloud computing, edge computing, 5G/6G, and NFV. In this survey, the state of the art, in the framework of the proposed categorization, is reviewed. Furthermore, comparisons are provided in terms of a set of proposed criteria which assess the methods from the aspects of methodology, main results, as well as application-specific criteria. Finally, we discuss lessons learned and highlight some potential research directions.

show abstract

Section: B Existing Relevant Surveys and Tutorialsmentioning

confidence: 99%

A Survey on In-Network Computing: Programmable Data Plane and Technology Specific Applications

Kianpisheh

Taleb

2023

IEEE Commun. Surv. Tutorials

View full text Add to dashboard Cite

show abstract

“…As case studies, we design two novel services: Quasi-Shortest-Service-First Scheduling to minimize the cluster-wide average JCT ( §4.2), and Cluster Energy Saving to improve the cluster energy efficiency ( §4.3). Other services based on machine learning prediction can also be integrated into our framework, e.g., burstiness-aware resource manager [80,83], network-aware job scheduler [12,38], etc. (2) High usability: our framework can be deployed into arbitrary GPU clusters.…”

Section: Qssfmentioning

confidence: 99%

Characterization and prediction of deep learning workloads in large-scale GPU datacenters

Sun

Yan

et al. 2021

Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis

View full text Add to dashboard Cite

Modern GPU datacenters are critical for delivering Deep Learning (DL) models and services in both the research community and industry. When operating a datacenter, optimization of resource scheduling and management can bring significant financial benefits. Achieving this goal requires a deep understanding of the job features and user behaviors. We present a comprehensive study about the characteristics of DL jobs and resource management. First, we perform a large-scale analysis of real-world job traces from SenseTime. We uncover some interesting conclusions from the perspectives of clusters, jobs and users, which can facilitate the cluster system designs. Second, we introduce a general-purpose framework, which manages resources based on historical data. As case studies, we design (1) a Quasi-Shortest-Service-First scheduling service, which can minimize the cluster-wide average job completion time by up to 6.5×; (2) a Cluster Energy Saving service, which improves overall cluster utilization by up to 13%.

show abstract

“…Second, we consider the scheduling at the job level, and do not cover the scheduling approaches at the hardware resource level (e.g., network I/O, power). For instance, HIRE [13] proposed a novel innetwork computing scheduling algorithm for datacenter switches. A number of works [49,99,145] utilized the DVFS mechanism on CPU and GPU chips to achieve cluster energy conservation.…”

Section: Relevant Studies Not Included In This Surveymentioning

confidence: 99%

Deep Learning Workload Scheduling in GPU Datacenters: Taxonomy, Challenges and Vision

Gao¹,

Hu²,

Ye³

et al. 2022

Preprint

View full text Add to dashboard Cite

Deep learning (DL) shows its prosperity in a wide variety of fields. The development of a DL model is a time-consuming and resource-intensive procedure. Hence, dedicated GPU accelerators have been collectively constructed into a GPU datacenter. An efficient scheduler design for such GPU datacenter is crucially important to reduce the operational cost and improve resource utilization. However, traditional approaches designed for big data or high performance computing workloads can not support DL workloads to fully utilize the GPU resources. Recently, substantial schedulers are proposed to tailor for DL workloads in GPU datacenters. This paper surveys existing research efforts for both training and inference workloads. We primarily present how existing schedulers facilitate the respective workloads from the scheduling objectives and resource consumption features. Finally, we prospect several promising future research directions. More detailed summary with the surveyed paper and code links can be found at our project website: https://github.com/S-Lab-System-Group/Awesome-DL-Scheduling-Papers. CCS Concepts: • General and reference → Surveys and overviews; • Computing methodologies → Machine learning; • Computer systems organization → Cloud computing.

show abstract

Switches for HIRE: resource scheduling for data center in-network computing

Cited by 20 publications

References 102 publications

A Survey on In-Network Computing: Programmable Data Plane and Technology Specific Applications

A Survey on In-Network Computing: Programmable Data Plane and Technology Specific Applications

Characterization and prediction of deep learning workloads in large-scale GPU datacenters

Deep Learning Workload Scheduling in GPU Datacenters: Taxonomy, Challenges and Vision

Contact Info

Product

Resources

About