• /  18
  • 下载费用: 19.90积分  

USC_New Measurement Abstractions for Data Centers

'USC_New Measurement Abstractions for Data Centers'
New Measurement Abstractions for Data Centers Minlan Yu University of Southern California 1 Key Challenges in Data Centers ? Service level agreements – Good, predictable performance ? Multi-tenancy – Resource isolation across tenants ? Cost reduction – Drive the network to high utilization 2 All require fine-grained network management Management = Measurement + Control 3 Performance Diagnosis Measure delay, throughput of flows Isolate and fix performance problems Security Capture traffic anomalies Filter out or block malicious traffic/tenants Traffic Engineering Collect per-flow statistics and changes Install routing rules The state of today’s Datacenter measurement The need for new measurement abstractions 4 Many Measurement Programs 5 Query independence: Specify many measurement queries with simple, programmable abstractions Performance diagnosis Monitor flow volumes Anomaly detection Many Data Collection Primitives 6 ? Passively collect different sources of data àLittle control on what (not) to measure Target independence: Specify which data to collect independent of where and how it is collected Limited Measurement Resources 7 Switch Small, on chip memory 40Gbps or 100Gbps Today, to reduce resource usage ? Low sampling rates ? Miss many important information Measurement Reconfigurable: Dynamically change the way resources are used for different measurement queries and traffic Control functions A New Abstraction for Measurement Specify measurement queries Measurement Framework Dynamically configure devices Bitmap; Reversible sketch Traffic change detectionCount-min sketch; Reversible sketch Traffi c entropy on port fi eld Multi-resolution classifi er; Count-min sketch Flow size distribution multi-resolution classifi er; hash table Build on P4-compatible Switches 17 Simple Hash: Traffic diversity adds randomness TCAM: Only require 10- 100 entries SRAM: Logical tables with flexible sizes Parsing (on headers & their hashes) Matching (On headers and hashes) Reconfigurable Counting pkt A New Abstraction for Measurement Specify measurement queries Measurement Framework Dynamically configure devices & manage resources Automatically collect the right data 18 Query independence: Expressive Abstractions for many queries Reconfigurable: Efficient Runtime Target independence: Implementable for many devices
关 键 词:
USC_NewMeasurementAbstractionsforDataCenters
 天天文库所有资源均是用户自行上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作他用。
关于本文
本文标题:USC_New Measurement Abstractions for Data Centers
链接地址: https://www.wenku365.com/p-39775640.html
关于我们 - 网站声明 - 网站地图 - 资源地图 - 友情链接 - 网站客服点击这里,给天天文库发消息,QQ:1290478887 - 联系我们

本站为“文档C2C交易模式”,即用户上传的文档直接卖给(下载)用户,本站只是中间服务平台,本站所有文档下载所得的收益归上传人(含作者)所有【成交的100%(原创)】。本站是网络服务平台方,若您的权利被侵害,侵权客服QQ:1290478887 欢迎举报。

1290478887@qq.com 2017-2027 https://www.wenku365.com 网站版权所有

粤ICP备19057495号 

收起
展开