当前位置:首页 > 报告详情

基准测试很难:CMS 工作负载基准测试的“密封”容器.pdf

上传人: 明**** 编号:1011793 2025-12-21 11页 1.07MB

1、Grant Mackey,Jackrabbit LabsVikrant Soman,UberBenchmarking is HARD:A Hermetically Sealed Container for CMS Workload BenchmarkingSERVER:COMPOSABLE MEMORY SYSTEMS(CMS)CMS project provides end users the ability to deploy a variety of native and pooled memory solutions using industry standard device cap

2、abilities and configurationsFor end-user adoption its important to be able to map the right CMS solution to the right workload configurations via benchmarks to realize TCO benefitsCMSBench announced on github to provide the community ability to access benchmark configs and results run against variou

3、s CMS solutions Challenges to adoption of this repo-How do I trust the results?How can I setup and configure the benchmarks easily?What benchmarks should I be running to evaluate a specific CMS solution?Executive SummaryNew folder structure aligned with CMS solutions and workload configurations that

4、 help TCO analysis(more in the next talk)Hermetically sealed container to make it easy to run the benchmarks Immutable container config during execution to build trust in the submitted resultsRebooting CMSBench to address adoption challenges Host w/Firmware,OS&VMMDDRCXL.memPCIeDRAMCXL Memory Buffer(

5、Direct Attached)CXL ControllerMedia CPUMemory controllerApplication server using CXL memory onlyApplication server using DRAM main memory onlyTopology specific foldersDocker compose files for specific benchmarks CMSBench github overviewHermetically sealed(adjective)/hmet.kl.i sild/:separated and pro

6、tected from very different conditions outside.A container or space tightly closed that no air can leave or enter it.Runtime containers are very good at capturing and replicating a software stack suitable for the task it is made to execute,and can drastically simplify execution of said taskDocker bui

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
根据报告的内容,全文主要内容概括如下: - **CMSBench项目**:旨在提供一种容器化的CMS(可组合内存系统)基准测试工具,帮助用户评估不同CMS解决方案的性能和成本效益。 - **挑战与解决方案**:解决用户对结果的可信度、配置复杂性和评估特定解决方案的基准问题。 - **容器化基准测试**:使用“密封容器”方法,确保基准测试的可重复性和结果的可信度。 - **基准测试内容**:包括硬件配置、软件栈和特定工作负载的配置。 - **目标**:集成3-5个用户应用,如内存缓存、向量数据库和数据分析工具。 - **现状**:Intel MLC和STREAM基准测试已完成,正在转向应用层面的基准测试。 - **重要性**:促进CXL等远内存技术的生态系统发展,降低组织采用新技术的风险。 - **参与方式**:通过GitHub贡献反馈、测试用例和功能请求。
如何轻松评估内存系统?" 如何信任CMSBench的测试结果?" 简化内存系统基准测试!"
客服
商务合作
小程序
服务号
折叠