创建连贯、可扩展且可互操作的遥测解决方案.pdf

1、Jim Harford,System Architect,BroadcomCreating Coherent,Scalable,and Interoperable Telemetry SolutionsCreating Coherent,Scalable,and Interoperable Telemetry SolutionsJim Harford,System Architect,BroadcomHARDWARE MANAGEMENT“SCIM”=Scalable Cloud Infrastructure Management (part of Hardware Management)Ev

2、olution of previous CSM(Cloud Service Model)workstreamSCIMs charter is to address hardware management of large data centers(100,000 nodes or more),creating documents in the following areas:Architectures,protocols,and data models that communicate hardware management information across data center net

3、works;Analysis of existing communication mechanisms and interacting with standards bodies to improve the effectiveness of those mechanisms for hardware management at data center scale;Management challenges associated with new technologies(e.g.AI/ML,RDMA,PCIe/CXL,DDR,quantum computing);andPredictive

4、models that forecast how anticipated aspects of hardware systems affect the financial performance of the data center.OCP SCIM WorkstreamMany protocols to choose from!(Redfish,YANG,Open Telemetry,Prometheus,)Many data model schemas to choose from!Many naming choices for individual metrics!In our desi

5、re to bring order to the chaos,SCIM intends to follow these steps:1.Create a high-level,protocol-agnostic architecture for telemetry2.Map(as best we can)our architecture into the various protocols(see 1stbullet)3.Work with the responsible standards groups and open-source communities to improve their

6、 respective protocols4.Encourage the standardization of names for individual metrics of interestOur Approach to Telemetry StandardsBulk messages increase wire efficiencyStreaming measurements steady flow that doesnt require continuous queries from the consumer of the metricsFor DMTF-based standards,