CNS Core: Small: Embracing cross stack heterogeneity in next-generation cloud platforms

CNS 核心：小型：在下一代云平台中拥抱跨堆栈异构性

• 批准号: 2116962
• 负责人: Chitaranjan Das
• 金额: $ 50万
• 依托单位: Pennsylvania State Univ University Park
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2116962&HistoricalAwards=false
• 关键词: CNS; Core; Small; Embracing; cross

Cloud computing has emerged as an attractive computing paradigm to achieve better economy of scale and flexibility for many application domains. For catering to the increasing service demands of these applications, cloud platforms have steadily evolved to offer a myriad of resource and service types. This has complicated the cloud adoption process due to challenges with respect to resource provisioning, billing complexity, and inaccurate resource demand estimation. This leaves cloud-bound tenants with an uphill task of achieving an optimal cost-to-performance ratio when hosting their applications on public cloud resources. On the other hand, the ever-growing cloud market has pushed providers to embrace heterogeneity, both at the hardware and software levels, through the rapid adoption of state-of-the-art hardware (e.g. Graphics Processing Units (GPUs), Field-Programmable Gate Arrays (FPGAs), etc.,) and software (e.g. Virtual Machines (VMs), containers, and serverless functions) developments. However, the main mediator between the applications and the underlying cloud computing infrastructure, which is the resource manager and scheduler, is not fully equipped to make the best use of the ever-evolving system heterogeneity and software complexity. Thus, the cloud resource management and scheduling framework needs a fresh look in order to adapt to the emerging hardware and software heterogeneity to maximize the deliverable performance and energy efficiency of cloud infrastructures. This project adresses these challenges by investigating a two-fold approach for developing a tenant- and provider-cognizant scalable cloud scheduling framework. First, from a tenant's perspective, we are developing an efficient cloud configuration framework that caters to the application requirements in terms of cost and performance. Second, from a provider's perspective, we are developing an efficient and adaptive resource management framework that can exploit the available heterogeneity across the cloud stack. The resource management framework will be integrated with commercial resource orchestrators like Kubernetes and will be evaluated using public cloud platforms with real-world workloads. The outcomes of this project should enable the effective utilization of heterogeneity offered by cloud providers for enhanced performance and energy efficiency across many application domains. The results of this research will foster new research directions in several areas of cloud computing, and the scheduling framework will be made publicly available. Furthermore, student training and research opportunities for undergraduate students will be facilitated through this research. The outreach activities include mentoring of women and minority students, participation in K-12 activities, and strengthening the Broadening Participation in Computing (BPC) program in the department.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

云计算已经成为一种有吸引力的计算范式，可以为许多应用领域实现更好的规模经济和灵活性。为了满足这些应用程序不断增长的服务需求，云平台已经稳步发展，提供了无数的资源和服务类型。由于资源供应、计费复杂性和不准确的资源需求估计方面的挑战，这使得云采用过程变得复杂。这使得云计算租户在公共云资源上托管其应用程序时，要实现最佳的成本与性能比，这是一项艰巨的任务。另一方面，不断增长的云市场推动供应商通过快速采用最先进的硬件（如图形处理单元（gpu）、现场可编程门阵列（fpga）等）和软件（如虚拟机（vm）、容器和无服务器功能）开发，在硬件和软件层面上拥抱异构性。然而，应用程序和底层云计算基础设施之间的主要中介（即资源管理器和调度器）并不能充分利用不断变化的系统异构性和软件复杂性。因此，云资源管理和调度框架需要重新审视，以适应新兴的硬件和软件异构性，从而最大限度地提高云基础设施的可交付性能和能源效率。该项目通过研究一种双重方法来解决这些挑战，该方法用于开发一个租户和提供商认知的可扩展云调度框架。首先，从租户的角度来看，我们正在开发一个高效的云配置框架，以满足应用程序在成本和性能方面的需求。其次，从提供商的角度来看，我们正在开发一种高效且自适应的资源管理框架，该框架可以利用云堆栈中的可用异构性。资源管理框架将与商业资源编排器（如Kubernetes）集成，并将使用具有实际工作负载的公共云平台进行评估。该项目的成果应该能够有效地利用云提供商提供的异构性，从而在许多应用程序领域中增强性能和能源效率。这项研究的结果将在云计算的几个领域培育新的研究方向，并且调度框架将公开可用。此外，通过本研究，将促进本科生的学生培训和研究机会。外联活动包括为妇女和少数族裔学生提供指导，参与K-12活动，以及加强本系的“扩大计算机参与”（BPC）计划。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

CASH: A Credit Aware Scheduling for Public Cloud Platforms

• DOI: 10.1109/ccgrid51090.2021.00032
• 发表时间: 2020-09
• 期刊: 2021 IEEE/ACM 21st International Symposium on Cluster, Cloud and Internet Computing (CCGrid)
• 作者: Aakash Sharma;Saravanan Dhakshinamurthy;G. Kesidis;C. Das

Cocktail: A Multidimensional Optimization for Model Serving in Cloud

Cocktail：云中模型服务的多维优化

• 发表时间: 2023
• 期刊: 19th USENIX Symposium on Networked Systems Design and Implementation.
• 作者: Jashwant Raj Gunasekaran, Cyan Subhra

Cypress: input size-sensitive container provisioning and request scheduling for serverless platforms

• DOI: 10.1145/3542929.3563464
• 发表时间: 2022-11
• 期刊: Proceedings of the 13th Symposium on Cloud Computing
• 作者: Vivek M. Bhasi;Jashwant Raj Gunasekaran;Aakash Sharma;M. Kandemir;C. Das

Optimizing CPU Performance for Recommendation Systems At-Scale

大规模优化推荐系统的 CPU 性能

• DOI: 10.1145/3579371.3589112
• 发表时间: 2023
• 期刊: International Symposium on Computer Architecture 2023
• 作者: Jain, Rishabh;Cheng, Scott;Kalagi, Vishwas;Sanghavi, Vrushabh;Kaul, Samvit;Arunachalam, Meena;Maeng, Kiwan;Jog, Adwait;Sivasubramaniam, Anand;Kandemir, Mahmut Taylan
• 通讯作者: Kandemir, Mahmut Taylan

Kraken: Adaptive Container Provisioning for Deploying Dynamic DAGs in Serverless Platforms

• DOI: 10.1145/3472883.3486992
• 发表时间: 2021-11
• 期刊: Proceedings of the ACM Symposium on Cloud Computing
• 作者: Vivek M. Bhasi;Jashwant Raj Gunasekaran;P. Thinakaran;Cyan Subhra Mishra;M. Kandemir;C. Das