平台效率挑战与OCP机遇：面向DC-SCM的上下文感知AI_ML代理实现实时优化和安全.pdf

1、Raghu Kondapalli,Chief Technology Officer,AxiadoContributors:Sundaram Arumugasundaram,Principal Security Research ArchitectZhichao Zhang,Principal Machine Learning ArchitectPlatform Efficiency Challenges and the OCP Opportunity:Content-Aware AI/ML Agents on DC-SCM for Real-Time Optimization and Secu

2、rityRaghu Kondapalli,Chief Technology Officer,AxiadoContributors:Sundaram Arumugasundaram,Principal Security Research ArchitectZhichao Zhang,Principal Machine Learning ArchitectPlatform Efficiency Challenges and the OCP Opportunity:Content-Aware AI/ML Agents on DC-SCM for Real-Time Optimization and

3、SecurityOCP SPECIAL FOCUS:ARTIFICIAL INTELLIGENCE(AI)“AI and HPC workloads are also distributed and diverse,much more so than other virtualized and cloud-native applications.In the pursuit of greater performance,infrastructure specialization for some workload categories is burgeoning,but risks fract

4、uring the supply chain into silos,raising costs and reducing efficiencies”-https:/www.opencompute.org/projects/open-systems-for-aiServer/platform efficiency encompasses energy use,resource utilization,security,thermal management,GPU/CPU usage optimization,effective workload distribution and the over

5、all system health.Recent platform studies highlight low infrastructure utilization across servers,racks,switches,and pods.GPU utilization is around 60%,with CPUs and networks similarly affected Proposed Solution:Real time monitoring,processing and platform actions using AI-ML models/agents running o

6、n OCP standard DC-SCM hardware to achieve maximum automation with precisionCall for action:OCP is well-positioned to drive the standards and solutions to improve platform efficiencyOverview High levelArchitecture Real-time data/metrics sensing layer1Continuous data processing layer 2Just-in-time act