{"id":21056,"date":"2026-06-10T12:19:19","date_gmt":"2026-06-10T04:19:19","guid":{"rendered":"https:\/\/ab123.xyz\/?p=21056"},"modified":"2026-06-10T12:19:19","modified_gmt":"2026-06-10T04:19:19","slug":"%e8%85%be%e8%ae%af%e4%ba%91%e5%90%91%e9%87%8f%e6%95%b0%e6%8d%ae%e5%ba%93milvus-2-4%e9%9b%86%e7%be%a4%e8%b0%83%e4%bc%98%e5%ae%9e%e6%88%98%e6%8c%87%e5%8d%97","status":"publish","type":"post","link":"https:\/\/ab123.xyz\/?p=21056","title":{"rendered":"\u817e\u8baf\u4e91\u5411\u91cf\u6570\u636e\u5e93Milvus 2.4\u96c6\u7fa4\u8c03\u4f18\u5b9e\u6218\u6307\u5357"},"content":{"rendered":"<p>\u5728 AI \u5927\u6a21\u578b\u4e0e\u5411\u91cf\u68c0\u7d22\u9700\u6c42\u7206\u53d1\u7684\u5f53\u4e0b\uff0c<strong>\u817e\u8baf\u4e91\u5411\u91cf\u6570\u636e\u5e93 Milvus 2.4<\/strong> \u51ed\u501f\u9ad8\u6027\u80fd\u3001\u9ad8\u53ef\u7528\u4e0e\u4e91\u539f\u751f\u67b6\u6784\uff0c\u6210\u4e3a\u4f01\u4e1a\u6784\u5efa\u77e5\u8bc6\u5e93\u3001\u63a8\u8350\u7cfb\u7edf\u3001\u591a\u6a21\u6001\u641c\u7d22\u7684\u6838\u5fc3\u5f15\u64ce\u3002\u7136\u800c\uff0c\u96c6\u7fa4\u8c03\u4f18\u76f4\u63a5\u51b3\u5b9a\u4e86\u68c0\u7d22\u5ef6\u8fdf\u3001\u541e\u5410\u91cf\u4e0e\u6210\u672c\u3002\u672c\u6587\u5c06\u4ece\u7d22\u5f15\u7b56\u7565\u3001\u8d44\u6e90\u8c03\u5ea6\u3001\u53c2\u6570\u914d\u7f6e\u4e09\u5927\u7ef4\u5ea6\uff0c\u6df1\u5ea6\u89e3\u6790 Milvus 2.4 \u96c6\u7fa4\u8c03\u4f18\u7684\u5173\u952e\u6280\u5de7\uff0c\u5e2e\u52a9\u60a8\u5145\u5206\u53d1\u6325\u6027\u80fd\u6f5c\u529b\u3002<\/p>\n<p>\u5b98\u65b9\u6587\u6863\u4e0e\u6700\u65b0\u5de5\u5177\u4e0b\u8f7d\u8bf7\u8bbf\u95ee\uff1a<a href=\"https:\/\/cloud.tencent.com\/product\/milvus\" target=\"_blank\">\u817e\u8baf\u4e91\u5411\u91cf\u6570\u636e\u5e93 Milvus \u5b98\u65b9\u7f51\u7ad9<\/a><\/p>\n<h2>\u4e00\u3001Milvus 2.4 \u6838\u5fc3\u529f\u80fd\u4e0e\u67b6\u6784\u4f18\u52bf<\/h2>\n<p>Milvus 2.4 \u5728\u4e91\u539f\u751f\u5c42\u9762\u8fdb\u884c\u4e86\u91cd\u5927\u5347\u7ea7\uff1a\u652f\u6301\u6df7\u5408\u67e5\u8be2\uff08\u5411\u91cf + \u6807\u91cf\u8fc7\u6ee4\uff09\u3001GPU \u52a0\u901f\u7d22\u5f15\uff08\u5982 RAFT\u3001IVF_PQ\uff09\u3001\u4ee5\u53ca\u81ea\u52a8\u5f39\u6027\u4f38\u7f29\u3002\u5176\u96c6\u7fa4\u7531 DataNode\u3001QueryNode\u3001IndexNode \u7b49\u7ec4\u4ef6\u6784\u6210\uff0c\u5404\u7ec4\u4ef6\u53ef\u72ec\u7acb\u6269\u7f29\u5bb9\u3002\u8c03\u4f18\u7684\u9996\u8981\u4efb\u52a1\u662f\u7406\u89e3\u5404\u7ec4\u4ef6\u8d1f\u8f7d\u7279\u5f81\uff0c\u907f\u514d\u8d44\u6e90\u4e89\u62a2\u3002<\/p>\n<h3>1.1 \u5411\u91cf\u7d22\u5f15\u7c7b\u578b\u9009\u62e9<\/h3>\n<p>\u6839\u636e\u6570\u636e\u89c4\u6a21\u4e0e\u53ec\u56de\u7387\u8981\u6c42\uff0c\u9009\u62e9\u5408\u9002\u7684\u7d22\u5f15\uff1a<\/p>\n<ul>\n<li>IVF_FLAT\uff1a\u9002\u5408\u767e\u4e07\u7ea7\u6570\u636e\uff0c\u7cbe\u5ea6\u9ad8\u4f46\u67e5\u8be2\u8f83\u6162<\/li>\n<li>HNSW\uff1a\u9002\u5408\u4ebf\u7ea7\u6570\u636e\uff0c\u901f\u5ea6\u4e0e\u7cbe\u5ea6\u5e73\u8861\uff0c\u5185\u5b58\u6d88\u8017\u8f83\u5927<\/li>\n<li>DISKANN\uff1a\u9002\u5408\u8d85\u5927\u89c4\u6a21\uff08\u5341\u4ebf+\uff09\uff0c\u5229\u7528 SSD \u964d\u4f4e\u6210\u672c<\/li>\n<\/ul>\n<p>\u5efa\u8bae\u5728 Milvus 2.4 \u4e2d\u5f00\u542f GPU \u7d22\u5f15\uff08\u5982 RAFT\uff09\uff0c\u53ef\u5c06 HNSW \u6784\u5efa\u901f\u5ea6\u63d0\u5347 5-10 \u500d\u3002<\/p>\n<h2>\u4e8c\u3001\u96c6\u7fa4\u53c2\u6570\u8c03\u4f18\u6700\u4f73\u5b9e\u8df5<\/h2>\n<p>\u8c03\u4f18\u9700\u5173\u6ce8\u96c6\u7fa4\u7ea7\u4e0e Collection \u7ea7\u53c2\u6570\u3002\u4ee5\u4e0b\u4e3a\u5173\u952e\u4f18\u5316\u70b9\uff1a<\/p>\n<h3>2.1 \u8d44\u6e90\u5206\u914d\u4e0e\u5206\u7247\u7b56\u7565<\/h3>\n<p>\u6bcf\u4e2a shard \u5bf9\u5e94\u4e00\u4e2a DataNode\uff0c\u5efa\u8bae shard \u6570\u91cf = \u6570\u636e\u91cf\uff08\u5355\u4f4d GB\uff09\/ 10\u3002\u4f8b\u5982 100GB \u6570\u636e\u4f7f\u7528 10 \u4e2a shard\u3002QueryNode \u5185\u5b58\u5efa\u8bae\u4e3a\u6570\u636e\u91cf\u7684 1.2 \u500d\uff0c\u5e76\u914d\u7f6e <code>queryNode.gracefulTime<\/code> \u4e3a 2000ms \u4ee5\u51cf\u5c11\u5b9e\u65f6\u5199\u5165\u5f15\u8d77\u7684\u6296\u52a8\u3002<\/p>\n<h3>2.2 \u7d22\u5f15\u6784\u5efa\u4e0e\u67e5\u8be2\u5e76\u53d1\u8c03\u4f18<\/h3>\n<p>\u5728 IndexNode \u4e0a\u5f00\u542f GPU \u52a0\u901f\u65f6\uff0c\u9700\u8bbe\u7f6e <code>indexNode.gpu.deployMode<\/code> \u4e3a &#8216;gpu&#8217;\uff0c\u5e76\u6307\u5b9a GPU \u663e\u5b58\u4e0a\u9650\u3002\u5bf9\u4e8e\u9ad8 QPS \u573a\u666f\uff0c\u9002\u5f53\u589e\u5927 <code>queryNode.search.poolSize<\/code>\uff08\u9ed8\u8ba4 8\uff09\u81f3 16~32\uff0c\u5e76\u542f\u7528\u7ed3\u679c\u7f13\u5b58\uff08<code>common.retentionDuration<\/code> \u8bbe\u4e3a 3600 \u79d2\uff09\u3002<\/p>\n<h2>\u4e09\u3001\u751f\u4ea7\u73af\u5883\u8c03\u4f18\u4e0e\u76d1\u63a7<\/h2>\n<p>\u4e0a\u7ebf\u524d\u52a1\u5fc5\u901a\u8fc7\u538b\u6d4b\u5de5\u5177\uff08\u5982 Milvus Benchmark\uff09\u6a21\u62df\u771f\u5b9e\u6d41\u91cf\u3002\u91cd\u70b9\u89c2\u5bdf\uff1a<\/p>\n<ul>\n<li><strong>CPU \u4f7f\u7528\u7387<\/strong>\uff1aQueryNode \u5e94\u4f4e\u4e8e 70%\uff0c\u5426\u5219\u8003\u8651\u6269\u5bb9\u526f\u672c<\/li>\n<li><strong>\u5185\u5b58\u4ea4\u6362<\/strong>\uff1a\u5173\u95ed\u64cd\u4f5c\u7cfb\u7edf swap\uff0c\u907f\u514d\u7d22\u5f15\u88ab\u6362\u51fa<\/li>\n<li><strong>\u7f51\u7edc\u5ef6\u8fdf<\/strong>\uff1a\u540c\u673a\u623f\u90e8\u7f72\uff0c\u51cf\u5c11\u8de8\u53ef\u7528\u533a\u901a\u4fe1<\/li>\n<\/ul>\n<p>\u53e6\u5916\uff0cMilvus 2.4 \u65b0\u589e\u4e86 <strong>\u52a8\u6001\u53c2\u6570\u70ed\u52a0\u8f7d<\/strong> \u529f\u80fd\uff0c\u53ef\u5728\u4e0d\u91cd\u542f\u96c6\u7fa4\u7684\u60c5\u51b5\u4e0b\u8c03\u6574 <code>queryNode.search.bufferSize<\/code> \u7b49\u53c2\u6570\uff0c\u6781\u5927\u964d\u4f4e\u4e86\u8c03\u4f18\u8fed\u4ee3\u6210\u672c\u3002<\/p>\n<h2>\u56db\u3001\u5178\u578b\u5e94\u7528\u573a\u666f<\/h2>\n<p>\u7ecf\u8fc7\u8c03\u4f18\u7684 Milvus 2.4 \u96c6\u7fa4\u53ef\u652f\u6491\uff1a<br \/>&#8211; \u7535\u5546\u4ee5\u56fe\u641c\u56fe\uff1a\u4ebf\u7ea7\u5546\u54c1\u56fe\u5411\u91cf\uff0c\u54cd\u5e94&lt; 50ms<br \/>&#8211; \u5927\u6a21\u578b RAG \u77e5\u8bc6\u5e93\uff1a\u7ed3\u5408\u817e\u8baf\u4e91\u5411\u91cf\u6570\u636e\u5e93\u63d2\u4ef6\uff0c\u5b9e\u73b0\u6beb\u79d2\u7ea7\u8bed\u4e49\u68c0\u7d22<br \/>&#8211; \u98ce\u9669\u63a7\u5236\uff1a\u5b9e\u65f6\u6bd4\u5bf9\u6307\u7eb9\u76f8\u4f3c\u5ea6\uff0c\u541e\u5410\u91cf\u8fbe 5000 QPS<\/p>\n<p>\u817e\u8baf\u4e91\u63d0\u4f9b\u5168\u6258\u7ba1\u7684 Milvus \u670d\u52a1\uff0c\u652f\u6301\u4e00\u952e\u96c6\u7fa4\u90e8\u7f72\u4e0e\u81ea\u52a8\u4f18\u5316\uff0c\u5927\u5e45\u964d\u4f4e\u8fd0\u7ef4\u95e8\u69db\u3002\u7acb\u5373\u4f53\u9a8c\uff1a<a href=\"https:\/\/cloud.tencent.com\/product\/milvus\" target=\"_blank\">\u817e\u8baf\u4e91\u5411\u91cf\u6570\u636e\u5e93 Milvus<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5728 AI \u5927\u6a21\u578b\u4e0e\u5411\u91cf\u68c0\u7d22\u9700\u6c42\u7206\u53d1\u7684\u5f53\u4e0b\uff0c\u817e\u8baf\u4e91\u5411\u91cf\u6570\u636e\u5e93 Milvus 2.4 \u51ed\u501f\u9ad8\u6027\u80fd\u3001\u9ad8\u53ef\u7528\u4e0e\u4e91\u539f\u751f\u67b6 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[24429,24431,24432,24430,24428],"class_list":["post-21056","post","type-post","status-publish","format-standard","hentry","category-4","tag-milvus-2-4","tag-milvus","tag-ai","tag-24430","tag-24428"],"_links":{"self":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/21056","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=21056"}],"version-history":[{"count":1,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/21056\/revisions"}],"predecessor-version":[{"id":21057,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/21056\/revisions\/21057"}],"wp:attachment":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=21056"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=21056"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=21056"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}