CRII: CSR: Skeletor: Building a Platform for Quantitative Workload Characterization

CRII：CSR：Skeletor：构建定量工作负载表征平台

• 批准号: 1755958
• 负责人: Avani Wildani
• 金额: $ 17.49万
• 依托单位: Emory University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-09-01 至 2021-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1755958&HistoricalAwards=false
• 关键词: CRII; CSR; Skeletor; Building; Platform

Combined compute, memory, storage and networking systems are too complex to be modeled correctly. Thus, only way to determine performance of an entire system or a sub-system in the context of a whole system is to test it by applying a set of workloads under controlled conditions. The workloads should be representative of the real applications. In characterizing a system, the workloads that expose diverse system behaviors are more valuable than workloads that behave similarly. Thus, workload characterization is important. This proposal focuses on workloads for storage systems to enable storage systems performance optimization. From a systems perspective, if the behavior of the executing workload can be mapped to one or more of the known behaviors, then the system can be adapted to a configuration(s) that is best suited for that behavior. The goal of this project is to recognize the behavior of the workload to find workload archetype. The project will research workload metrics important to measure and define rigorous archetypes parameterized by these metrics to characterize new workloads without complex, slow predictive analytics or onerous domain specification. The investigator plans to produce a workload classification schema in preparation for developing adaptive, workload-aware automated storage tuning and provisioning framework. The project has two main goals, (1) learning what to measure by creating a parameterized taxonomy of the model workloads; and (2) learning how to measure by creating a framework to infer and categorize functionally distinct workloads.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.

组合计算、内存、存储和网络系统太复杂，无法正确建模。因此，在整个系统的上下文中确定整个系统或子系统的性能的唯一方法是通过在受控条件下应用一组工作负载来测试它。工作负载应该代表真实的应用程序。在表征系统时，暴露不同系统行为的工作负载比行为相似的工作负载更有价值。因此，工作负载特性很重要。本建议书侧重于存储系统的工作负载，以实现存储系统性能优化。从系统的角度来看，如果执行工作负载的行为可以被映射到一个或多个已知行为，则系统可以被适配为最适合于该行为的配置。该项目的目标是识别工作负载的行为，以找到工作负载原型。该项目将研究重要的工作负载指标，以测量和定义由这些指标参数化的严格原型，以表征新的工作负载，而无需复杂，缓慢的预测分析或繁重的域规范。调查人员计划制定一个工作负载分类方案，为开发自适应、工作负载感知的自动化存储调优和配置框架做准备。该项目有两个主要目标，（1）通过创建模型工作负载的参数化分类来学习测量什么;（2）通过创建一个框架来推断和分类功能不同的工作负载，学习如何测量。该奖项反映了NSF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

C ENSUS : Counting Interleaved Workloads on Shared Storage

• 发表时间: 2020
• 作者: Si Chen;Jianqiao Liu;Avani Wildani

Desperately Seeking ... Optimal Multi-Tier Cache Configurations

• 发表时间: 2020
• 作者: Tyler Estro;Pranav Bhandari;Avani Wildani;E. Zadok

Chasing the Signal: Statistically Separating Multi-Tenant I/O Workloads

追逐信号：统计分离多租户 I/O 工作负载

• 发表时间: 2018
• 期刊: Machine Learning in Systems
• 作者: Chen, S.;Wildani, A.
• 通讯作者: Wildani, A.