{"id":696,"date":"2025-12-04T21:24:14","date_gmt":"2025-12-04T13:24:14","guid":{"rendered":"https:\/\/www.ndnlab.com\/?p=696"},"modified":"2025-12-04T21:24:16","modified_gmt":"2025-12-04T13:24:16","slug":"m3-accurate-flow-level-performance-estimationusing-machine-learning","status":"publish","type":"post","link":"https:\/\/www.ndnlab.com\/?p=696","title":{"rendered":"m3: Accurate Flow-Level Performance Estimationusing Machine Learning"},"content":{"rendered":"\n

\u5f53\u7f51\u7edc\u4eff\u771f\u4e0d\u518d\u201c\u6162 + \u91cd\u201d \u2014\u2014 m3 \u5e26\u6765\u7a81\u7834<\/h2>\n\n\n\n

\u5728\u73b0\u4ee3\u6570\u636e\u4e2d\u5fc3\u6216\u4e91\u670d\u52a1\u67b6\u6784\u4e2d\uff0c\u7f51\u7edc\u6027\u80fd \u2014 \u5c24\u5176\u662f\u6d41 (flow) \u5b8c\u6210\u65f6\u95f4 (Flow Completion Time, FCT) \u548c\u5c3e\u90e8\u65f6\u5ef6 (tail latency) \u2014 \u662f\u51b3\u5b9a\u7528\u6237\u4f53\u9a8c\u548c\u8d44\u6e90\u6548\u7387\u7684\u5173\u952e\u6307\u6807\u3002\u8fc7\u53bb\uff0c\u6211\u4eec\u8981\u4e48\u7528\u5305\u7ea7 (packet-level) \u4eff\u771f\u5de5\u5177\uff08\u5982 ns-3\u3001OMNeT++ \u7b49\uff09\uff0c\u5f97\u5230\u8f83\u9ad8\u7cbe\u5ea6\uff0c\u5374\u6162\u5230\u65e0\u6cd5\u5e94\u4ed8\u5927\u89c4\u6a21\u7f51\u7edc\uff1b\u8981\u4e48\u7528\u7b80\u5316\u8fd1\u4f3c\u65b9\u6848\uff0c\u5feb\u901f\u4f46\u4e0d\u53ef\u9760\u3002m3 \u7684\u8bde\u751f\uff0c\u4e3a\u8fd9\u4e2a\u7ecf\u5178\u96be\u9898\u5e26\u6765\u4e86\u65b0\u7684\u53ef\u80fd \u2014\u2014 \u5b83\u7528\u4e00\u79cd\u8f7b\u91cf\u3001\u5feb\u901f\u3001\u5374\u4f9d\u7136\u51c6\u786e\u7684\u65b9\u6cd5\uff0c\u4f7f\u5f97\u7f51\u7edc\u56e2\u961f\u80fd\u591f\u5728\u53ef\u63a5\u53d7\u65f6\u95f4\u5185\uff0c\u83b7\u5f97\u53ef\u4fe1\u8d56\u7684\u6027\u80fd\u4f30\u8ba1\u3002<\/p>\n\n\n\n

m3 \u7684\u6838\u5fc3\u521b\u65b0\uff1a\u62bd\u8c61 + \u673a\u5668\u5b66\u4e60 + \u8def\u5f84\u7ea7\u91c7\u6837<\/h2>\n\n\n\n

\u2022 \u4ece\u5305\u7ea7\u5230\u6d41\/\u8def\u5f84\u7ea7 \u2014 \u91cd\u6784\u4eff\u771f\u7c92\u5ea6<\/h3>\n\n\n\n

m3 \u653e\u5f03\u5bf9\u6bcf\u4e2a\u6570\u636e\u5305\u3001\u6bcf\u4e2a\u4e8b\u4ef6\u7684\u7ec6\u81f4\u4eff\u771f\uff0c\u800c\u662f\u91c7\u7528\u201c\u6d41 (flow) \/ \u8def\u5f84 (path) \u7ea7\u522b\u201d\u7684\u62bd\u8c61\u3002\u901a\u8fc7\u4e00\u79cd\u5feb\u901f\u7684\u6d41\u4f53 (fluid) \u6a21\u62df (flow-level fluid simulation)\uff0c\u6a21\u62df\u7f51\u7edc\u4e2d\u7ecf\u8fc7\u67d0\u4e00\u8def\u5f84 (path) \u7684\u6d41\u91cf\u60c5\u51b5 \u2014\u2014 \u5c24\u5176\u5173\u6ce8\u4ea4\u53c9\u6d41\u91cf (cross-traffic) \u5bf9\u5f53\u524d\u6d41 (foreground flow) \u7684\u5f71\u54cd\u3002\u8fd9\u6837\u65e2\u6781\u5927\u7b80\u5316\u4e86\u4eff\u771f\u590d\u6742\u5ea6\uff0c\u4e5f\u4fdd\u7559\u4e86\u4e0e\u7f51\u7edc\u6027\u80fd\u5bc6\u5207\u76f8\u5173\u7684\u5173\u952e\u7279\u6027\u3002 <\/p>\n\n\n\n

\u2022 \u8def\u5f84\u7ea7\u5206\u89e3 + \u62bd\u6837 + \u6c47\u603b \u2014 \u89c4\u6a21\u65e0\u60e7<\/h3>\n\n\n\n

m3 \u7684\u8bbe\u8ba1\u5047\u8bbe\uff1a\u6d41\u91cf\u5728\u4e0d\u76f8\u4ea4\u8def\u5f84 (non-intersecting paths) \u4e0a\u7684\u76f8\u4e92\u5f71\u54cd\u53ef\u4ee5\u5ffd\u7565\u4e0d\u8ba1\u3002\u4e5f\u5c31\u662f\u8bf4\uff0c\u53ea\u8981\u6211\u4eec\u5bf9\u4e00\u90e8\u5206\u4ee3\u8868\u6027\u7684\u8def\u5f84 (sample of paths) \u505a\u4eff\u771f \/ \u9884\u6d4b\uff0c\u5c31\u80fd\u8fd8\u539f\u6574\u4e2a\u7f51\u7edc\u7684 aggregate \u884c\u4e3a (\u4f8b\u5982\u6574\u4f53\u7684 tail latency \u5206\u5e03)\u3002\u901a\u8fc7\u8fd9\u79cd\u201c\u8def\u5f84\u5206\u89e3 + \u62bd\u6837 + \u7ed3\u679c\u6c47\u603b\u201d\u7684\u65b9\u5f0f\uff0cm3 \u7684\u6a21\u578b\u65e0\u8bba\u7f51\u7edc\u89c4\u6a21\u591a\u5927\uff0c\u90fd\u80fd\u4fdd\u6301\u9ad8\u6548\u7387\u4e0e\u8f83\u4f4e\u6210\u672c\u3002 <\/p>\n\n\n\n

\u2022 \u5c06\u7c97\u4eff\u771f + \u673a\u5668\u5b66\u4e60\u7ed3\u5408 \u2014 \u517c\u987e\u901f\u5ea6\u4e0e\u7cbe\u5ea6<\/h3>\n\n\n\n

m3 \u7528\u6d41\u7ea7\u4eff\u771f (flowSim) \u751f\u6210\u201c\u7279\u5f81\u56fe (feature map)\u201d\u2014\u2014\u5373\u7f51\u7edc\u5728\u7ed9\u5b9a\u5de5\u4f5c\u8d1f\u8f7d (workload)\u3001\u8def\u5f84 (path)\u3001\u914d\u7f6e (\u5982\u62e5\u585e\u63a7\u5236\u534f\u8bae\u3001\u8def\u7531\u7b56\u7565\u7b49) \u4e0b\u7684\u7c97\u7565\u884c\u4e3a\u7279\u5f81\u3002\u7136\u540e\uff0c\u4f7f\u7528\u673a\u5668\u5b66\u4e60 (ML) \u6a21\u578b (\u8bad\u7ec3\u65f6\u4ee5 packet-level \u4eff\u771f\u8f93\u51fa\u4f5c\u4e3a ground-truth)\uff0c\u6765\u9884\u6d4b\u66f4\u7cbe\u786e\u7684\u6307\u6807 (\u5982 FCT \u5206\u5e03)\u3002\u8fd9\u6837\uff0cm3 \u65e2\u5229\u7528\u6d41\u4f53\u6a21\u62df\u5feb\u901f\u751f\u6210\u7279\u5f81\uff0c\u4e5f\u501f\u52a9 ML \u201c\u6821\u6b63\u201d\u7c97\u4eff\u771f\u672c\u8eab\u7684\u4e0d\u7cbe\u7ec6\uff0c\u5f97\u5230\u63a5\u8fd1\u7cbe\u4eff\u771f\u7684\u51c6\u786e\u5ea6\u3002 <\/p>\n\n\n\n

\u2022 \u901a\u7528\u6027\u5f3a \u2014 \u652f\u6301\u591a\u79cd\u7f51\u7edc\u914d\u7f6e \/ \u62e5\u585e\u63a7\u5236 \/ \u62d3\u6251<\/h3>\n\n\n\n

m3 \u7684 ML \u6a21\u578b\u4e0d\u4ec5\u9002\u7528\u4e8e\u4e00\u79cd\u56fa\u5b9a\u7f51\u7edc\uff0c\u5b83\u88ab\u8bbe\u8ba1\u4e3a\u53ef\u4ee5\u6cdb\u5316\u5230\u4e0d\u540c\u62e5\u585e\u63a7\u5236 (congestion control, CC) \u534f\u8bae\u3001\u4e0d\u540c\u53c2\u6570\u8bbe\u7f6e\u3001\u4e0d\u540c\u884c\u8def (routing) \u7b56\u7565\u4ee5\u53ca\u4e0d\u540c\u7f51\u7edc\u62d3\u6251\u3002\u4e5f\u5c31\u662f\u8bf4\uff0c\u65e0\u8bba\u4f60\u7684\u6570\u636e\u4e2d\u5fc3\u7f51\u7edc\u5982\u4f55\u53d8\u5316\u3001\u6d41\u91cf\u5982\u4f55\u6ce2\u52a8\uff0cm3 \u90fd\u80fd\u4fdd\u6301\u53ef\u4fe1\u8d56\u3002 <\/p>\n\n\n\n

\u2022 \u663e\u8457\u63d0\u5347\u6027\u80fd \u2014 \u5feb\u5f88\u591a\uff0c\u8bef\u5dee\u5c11\u5f88\u591a<\/h3>\n\n\n\n

\u6839\u636e\u8bba\u6587\u8bc4\u4f30\uff0c\u76f8\u8f83\u4e8e\u4e4b\u524d\u4e00\u79cd\u9ad8\u6548\u4f30\u8ba1\u7cfb\u7edf Parsimon\uff0cm3 \u7684\u901f\u5ea6\u63d0\u5347\u5927\u7ea6 5.7 \u500d<\/strong>\uff0c\u4f30\u8ba1\u8bef\u5dee\u76f8\u6bd4 packet-level \u4eff\u771f\u51cf\u5c11\u7ea6 45.9%<\/strong>\u3002\u76f8\u6bd4\u4f20\u7edf\u7684 packet-level \u6a21\u62df (\u4f8b\u5982 ns-3) \u66f4\u662f\u5feb\u51e0\u4e2a\u6570\u91cf\u7ea7\u3002<\/p>\n\n\n\n

\u7528 m3 \u5feb\u901f\u9884\u89c1\u7f51\u7edc\u53d8\u5316\u5e26\u6765\u7684\u6027\u80fd\u5f71\u54cd<\/h2>\n\n\n\n

\u2022 \u5feb\u901f\u505a\u201c\u5047\u8bbe-\u9a8c\u8bc1\u201d<\/h3>\n\n\n\n

\u60f3\u77e5\u9053\u6269\u5bb9\u3001\u8c03\u6574\u62e5\u585e\u63a7\u5236\u7b97\u6cd5\u3001\u6539\u53d8\u8def\u7531\u3001\u91cd\u65b0\u5206\u914d\u6d41\u91cf\u6216\u53d8\u66f4\u5e03\u5c40\u540e\uff0c\u5bf9\u7f51\u7edc\u6027\u80fd\uff08\u5c24\u5176\u662f\u5c3e\u65f6\u5ef6 \/ \u6d41\u5b8c\u6210\u65f6\u95f4 FCT\uff09\u4f1a\u4ea7\u751f\u600e\u6837\u7684\u5f71\u54cd\uff1f\u4f7f\u7528 m3\uff0c\u4f60\u53ef\u4ee5\u5728\u51e0\u5206\u949f\u3001\u51e0\u5341\u79d2\u5185\u5f97\u5230\u53ef\u9760\u4f30\u8ba1\uff0c\u65e0\u9700\u8017\u8d39\u6570\u5c0f\u65f6\u751a\u81f3\u6570\u5929\u8fd0\u884c\u4f20\u7edf packet-level \u4eff\u771f\u3002<\/p>\n\n\n\n

\u5927\u5e45\u964d\u4f4e\u4eff\u771f\u6210\u672c<\/h3>\n\n\n\n

\u4f20\u7edf\u7684\u3001\u7ec6\u7c92\u5ea6\u7684\u4eff\u771f\u901a\u5e38\u8d44\u6e90\u5360\u7528\u5de8\u5927\u2014\u2014CPU\u3001\u5185\u5b58\u3001\u65f6\u95f4\u5f00\u9500\u6781\u9ad8\u3002m3 \u5219\u80fd\u8ba9\u4eff\u771f\u53d8\u5f97\u8f7b\u91cf\u3001\u7075\u6d3b\uff0c\u5c31\u50cf\u8fd0\u884c\u4e00\u6bb5\u811a\u672c\u4e00\u6837\u7b80\u5355\uff0c\u7279\u522b\u9002\u5408\u5927\u89c4\u6a21\u3001\u591a\u573a\u666f\u3001\u591a\u914d\u7f6e\u7ec4\u5408\u4e0b\u7684\u6027\u80fd\u8bc4\u4f30\u3002<\/p>\n\n\n\n

\u9002\u5e94\u771f\u5b9e\u3001\u5927\u89c4\u6a21\u3001\u5f02\u6784\u7f51\u7edc\u73af\u5883<\/h3>\n\n\n\n

\u73b0\u5b9e\u751f\u4ea7\u73af\u5883\u4e2d\u7684\u6570\u636e\u4e2d\u5fc3\u7f51\u7edc\u901a\u5e38\u590d\u6742\u4e14\u4e0d\u65ad\u53d8\u5316\uff1a\u62d3\u6251\u7ed3\u6784\u3001\u6d41\u91cf\u6a21\u5f0f\u3001\u8def\u7531\u7b56\u7565\u3001\u62e5\u585e\u63a7\u5236\u7b97\u6cd5\u2026\u2026\u53ef\u80fd\u5343\u5dee\u4e07\u522b\u3002m3 \u7684\u8bbe\u8ba1\u5177\u6709\u5f88\u5f3a\u7684\u901a\u7528\u6027\u548c\u6cdb\u5316\u80fd\u529b\uff0c\u5176\u4f30\u8ba1\u7ed3\u679c\u5bf9\u591a\u79cd\u7f51\u7edc\u914d\u7f6e\u3001\u4e0d\u540c\u6d41\u91cf\u4e0e\u62d3\u6251\u73af\u5883\u90fd\u80fd\u4fdd\u6301\u6709\u6548\uff0c\u56e0\u6b64\u6bd4\u5f88\u591a\u5047\u8bbe\u56fa\u5b9a\u3001\u7406\u60f3\u5316\u7684\u4eff\u771f\u5de5\u5177\u66f4\u9002\u5408\u771f\u5b9e\u751f\u4ea7\u573a\u666f\u3002<\/p>\n\n\n\n

\u52a0\u901f\u7f51\u7edc\u67b6\u6784\uff0f\u914d\u7f6e\uff0f\u4f18\u5316\u8fed\u4ee3<\/h3>\n\n\n\n

\u65e0\u8bba\u662f\u6d4b\u8bd5\u65b0\u7684\u62e5\u585e\u63a7\u5236\u7b97\u6cd5\u3001\u9a8c\u8bc1\u65b0\u7684\u8def\u7531\u7b56\u7565\u3001\u5bf9\u6bd4\u5bb9\u91cf\u89c4\u5212\u65b9\u6848\uff0c\u8fd8\u662f\u4e3a\u6d41\u91cf\u9ad8\u5cf0\u3001\u6d41\u91cf\u5206\u5e03\u53d8\u5316\u505a\u9884\u4f30\u2014\u2014m3 \u90fd\u53ef\u4ee5\u4f5c\u4e3a\u7b2c\u4e00\u8f6e\u5feb\u901f\u5224\u65ad\u5de5\u5177\u3002\u5728\u5f97\u5230\u6ee1\u610f\u7ed3\u679c\u540e\uff0c\u518d\u4f7f\u7528\u66f4\u91cd\u3001\u66f4\u771f\u5b9e\u7684\u65b9\u6cd5\u505a\u6df1\u5165\u9a8c\u8bc1\u3002\u8fd9\u6837\uff0c\u4f60\u548c\u56e2\u961f\u5728\u7f51\u7edc\u8bbe\u8ba1\u3001\u4f18\u5316\u3001\u51b3\u7b56\u5468\u671f\u4e0a\u5c06\u66f4\u9ad8\u6548\u3001\u66f4\u654f\u6377\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

\u5f53\u7f51\u7edc\u4eff\u771f\u4e0d\u518d\u201c\u6162 + \u91cd\u201d \u2014\u2014 m3 \u5e26\u6765\u7a81\u7834 \u5728\u73b0\u4ee3\u6570\u636e\u4e2d\u5fc3\u6216\u4e91\u670d\u52a1\u67b6\u6784\u4e2d\uff0c\u7f51\u7edc\u6027\u80fd \u2014 \u5c24\u5176\u662f\u6d41 (flow) \u5b8c\u6210\u65f6\u95f4 (Flow Completion Time, FCT) \u548c\u5c3e\u90e8\u65f6\u5ef6 (tail latency) \u2014 \u662f\u51b3\u5b9a\u7528\u6237\u4f53\u9a8c\u548c\u8d44\u6e90\u6548\u7387\u7684\u5173\u952e\u6307\u6807\u3002\u8fc7\u53bb\uff0c\u6211\u4eec\u8981\u4e48\u7528\u5305\u7ea7 (packet-level) \u4eff\u771f\u5de5\u5177\uff08\u5982 ns-3\u3001OMNeT++ \u7b49\uff09\uff0c\u5f97\u5230\u8f83\u9ad8\u7cbe\u5ea6\uff0c\u5374\u6162\u5230\u65e0\u6cd5\u5e94\u4ed8\u5927\u89c4\u6a21\u7f51\u7edc\uff1b\u8981\u4e48\u7528\u7b80\u5316\u8fd1\u4f3c\u65b9\u6848\uff0c\u5feb\u901f\u4f46\u4e0d\u53ef\u9760\u3002m3 \u7684\u8bde\u751f\uff0c\u4e3a\u8fd9\u4e2a\u7ecf\u5178\u96be\u9898\u5e26\u6765\u4e86\u65b0\u7684\u53ef … \u7ee7\u7eed\u9605\u8bfb →<\/span><\/a><\/p>\n","protected":false},"author":3,"featured_media":701,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[17,5,1],"tags":[],"class_list":["post-696","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-17","category-rengongzhineng","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/posts\/696","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=696"}],"version-history":[{"count":4,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/posts\/696\/revisions"}],"predecessor-version":[{"id":700,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/posts\/696\/revisions\/700"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/media\/701"}],"wp:attachment":[{"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=696"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=696"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=696"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}