{"id":1185,"date":"2026-03-30T09:40:43","date_gmt":"2026-03-30T01:40:43","guid":{"rendered":"https:\/\/www.ndnlab.com\/?p=1185"},"modified":"2026-03-30T09:40:44","modified_gmt":"2026-03-30T01:40:44","slug":"hydro-adaptive-query-processing-of-ml-queries","status":"publish","type":"post","link":"https:\/\/www.ndnlab.com\/?p=1185","title":{"rendered":"Hydro: Adaptive Query Processing of ML Queries"},"content":{"rendered":"\n<p><strong>1. \u6458\u8981\uff08Abstract\uff09<\/strong><\/p>\n\n\n\n<p>Hydro\u63d0\u51fa\u4e86\u4e00\u79cd\u9762\u5411\u673a\u5668\u5b66\u4e60\u67e5\u8be2\u7684\u6570\u636e\u5e93\u7ba1\u7406\u7cfb\u7edf\uff08ML-centric DBMS\uff09\uff0c\u6838\u5fc3\u521b\u65b0\u5728\u4e8e\u5c06\u81ea\u9002\u5e94\u67e5\u8be2\u5904\u7406\uff08Adaptive Query Processing\uff0cAQP\uff09\u6280\u672f\u5168\u9762\u5f15\u5165UDF\u5bc6\u96c6\u578b\u67e5\u8be2\u573a\u666f\uff0c\u4ece\u800c\u5f7b\u5e95\u89e3\u51b3\u4f20\u7edf\u9759\u6001\u67e5\u8be2\u4f18\u5316\u5728ML\u67e5\u8be2\u4e2d\u7684\u4e24\u5927\u6838\u5fc3\u74f6\u9888\u3002\u9996\u5148\uff0cUDF\u7684\u6267\u884c\u4ee3\u4ef7\u548c\u9009\u62e9\u6027\u9ad8\u5ea6\u4f9d\u8d56\u8f93\u5165\u6570\u636e\uff0c\u4e14\u53d7\u7cfb\u7edf\u7ea7\u7f13\u5b58\u7b49\u52a8\u6001\u56e0\u7d20\u5f71\u54cd\uff0c\u5bfc\u81f4\u67e5\u8be2\u4f18\u5316\u9636\u6bb5\u7684\u9884\u4f30\u7edf\u8ba1\u4fe1\u606f\u6781\u4e0d\u53ef\u9760\uff1b\u5176\u6b21\uff0c\u6700\u4f18\u67e5\u8be2\u8ba1\u5212\u662f\u6570\u636e\u4f9d\u8d56\u7684\uff0c\u9759\u6001\u8ba1\u5212\u65e0\u6cd5\u5728\u6267\u884c\u8fc7\u7a0b\u4e2d\u52a8\u6001\u8c03\u6574\u3002Hydro\u57fa\u4e8eEvaDB\u6784\u5efa\uff0c\u5728\u67e5\u8be2\u6267\u884c\u5168\u7a0b\u5b9e\u65f6\u76d1\u63a7UDF\u7edf\u8ba1\u4fe1\u606f\uff0c\u901a\u8fc7\u52a8\u6001\u8c03\u6574\u8c13\u8bcd\u6c42\u503c\u987a\u5e8f\u548c\u786c\u4ef6\u8d44\u6e90\u5206\u914d\uff0c\u5b9e\u73b0\u6700\u4f18\u6267\u884c\u8def\u5f84\u3002<\/p>\n\n\n\n<p>\u7cfb\u7edf\u5bf9\u7ecf\u5178Eddy AQP\u6846\u67b6\u8fdb\u884c\u4e86\u9488\u5bf9ML\u67e5\u8be2\u7684\u6df1\u5ea6\u5b9a\u5236\uff0c\u5f15\u5165Laminar\u8def\u7531\u673a\u5236\u8d1f\u8d23\u81ea\u52a8\u6269\u5c55\u5de5\u4f5c\u7ebf\u7a0b\u5e76\u5b9e\u73b0\u6570\u636e\u611f\u77e5\u7684\u8d1f\u8f7d\u5747\u8861\u3002\u8bba\u6587\u901a\u8fc7\u56db\u4e2a\u591a\u6837\u5316\u7528\u4f8b\uff08\u4e09\u4e2a\u89c6\u9891\u5206\u6790\u573a\u666f\u548c\u4e00\u4e2a\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5206\u6790\u573a\u666f\uff09\u8fdb\u884c\u4e86\u5168\u9762\u9a8c\u8bc1\uff0c\u7ed3\u679c\u663e\u793aHydro\u76f8\u6bd4\u57fa\u7ebf\u9759\u6001\u4f18\u5316\u7cfb\u7edf\u6700\u9ad8\u53ef\u5e26\u676511.52\u500d\u7684\u7aef\u5230\u7aef\u52a0\u901f\uff0c\u540c\u65f6\u4fdd\u6301\u67e5\u8be2\u7ed3\u679c\u7cbe\u5ea6\u4e0d\u53d8\u3002\u8be5\u5de5\u4f5c\u4e0d\u4ec5\u586b\u8865\u4e86AQP\u5728UDF\u5bc6\u96c6\u573a\u666f\u4e0b\u7684\u7814\u7a76\u7a7a\u767d\uff0c\u8fd8\u4e3a\u4e0b\u4e00\u4ee3\u667a\u80fd\u6570\u636e\u5e93\u7684\u67e5\u8be2\u5904\u7406\u63d0\u4f9b\u4e86\u53ef\u6269\u5c55\u3001\u53ef\u901a\u7528\u7684\u7cfb\u7edf\u6846\u67b6\uff0c\u5177\u6709\u91cd\u8981\u7684\u7406\u8bba\u610f\u4e49\u548c\u5de5\u7a0b\u5b9e\u7528\u4ef7\u503c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"663\" height=\"507\" src=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/1-4.png\"  class=\"wp-image-1186\" srcset=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/1-4.png 663w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/1-4-300x229.png 300w\" sizes=\"auto, (max-width: 663px) 100vw, 663px\" title=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe\" alt=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe\" \/><\/figure>\n\n\n\n<p><strong>2. \u5f15\u8a00\uff08Introduction\uff09<\/strong><\/p>\n\n\n\n<p>\u73b0\u4ee3\u6570\u636e\u5e93\u7cfb\u7edf\u5df2\u5e7f\u6cdb\u652f\u6301\u5c06\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u76f4\u63a5\u5c01\u88c5\u4e3a\u7528\u6237\u81ea\u5b9a\u4e49\u51fd\u6570\uff08UDF\uff09\u5d4c\u5165SQL\u67e5\u8be2\uff0c\u8fd9\u4f7f\u5f97\u7528\u6237\u80fd\u591f\u5728\u6570\u636e\u5e93\u5185\u90e8\u9ad8\u6548\u5b8c\u6210\u590d\u6742\u7684ML\u4efb\u52a1\u3002\u7136\u800c\uff0cUDF\u7684\u8ba1\u7b97\u5bc6\u96c6\u7279\u6027\u5bfc\u81f4\u5176\u6210\u4e3a\u67e5\u8be2\u6267\u884c\u7684\u9996\u8981\u74f6\u9888\u3002\u4ee5\u89c6\u9891\u6570\u636e\u5e93\u7ba1\u7406\u7cfb\u7edf\uff08VDBMS\uff09\u4e3a\u4f8b\uff0c\u5927\u91cf\u8ba1\u7b97\u673a\u89c6\u89c9UDF\uff08\u5982\u76ee\u6807\u68c0\u6d4b\u3001\u5206\u7c7b\u6a21\u578b\uff09\u4f7f\u5f97\u67e5\u8be2\u6027\u80fd\u4e25\u91cd\u53d7\u9650\u3002\u4f20\u7edfML-centric DBMS\u867d\u7136\u5c1d\u8bd5\u901a\u8fc7\u91c7\u6837\u201c\u91d1\u4e1d\u96c0\u201d\u6570\u636e\u9884\u4f30UDF\u7edf\u8ba1\u4fe1\u606f\u6765\u4f18\u5316\u67e5\u8be2\u8ba1\u5212\uff0c\u4f46\u8fd9\u79cd\u9759\u6001\u4f18\u5316\u65b9\u5f0f\u5728\u5b9e\u9645\u6267\u884c\u4e2d\u66b4\u9732\u51fa\u660e\u663e\u7f3a\u9677\uff1aUDF\u4ee3\u4ef7\u968f\u8f93\u5165\u6570\u636e\uff08\u5982\u8fb9\u754c\u6846\u5c3a\u5bf8\uff09\u52a8\u6001\u53d8\u5316\uff0c\u7cfb\u7edf\u7ea7\u7f13\u5b58\u590d\u7528\u8fdb\u4e00\u6b65\u626d\u66f2\u9884\u4f30\u503c\uff0c\u6700\u7ec8\u751f\u6210\u6b21\u4f18\u7684\u56fa\u5b9a\u67e5\u8be2\u8ba1\u5212\u3002<\/p>\n\n\n\n<p>Hydro\u7684\u6838\u5fc3\u8d21\u732e\u5728\u4e8e\u8ba4\u8bc6\u5230ML\u67e5\u8be2\u7684\u6700\u4f18\u8ba1\u5212\u9ad8\u5ea6\u4f9d\u8d56\u5b9e\u65f6\u6570\u636e\u7279\u6027\uff0c\u5fc5\u987b\u5728\u6267\u884c\u8fc7\u7a0b\u4e2d\u6301\u7eed\u81ea\u9002\u5e94\u8c03\u6574\u3002\u8bba\u6587\u4ee5\u201c\u5bfb\u627e\u4e22\u5931\u7684\u9ed1\u5927\u4e39\u72ac\u201d\u8fd9\u4e00\u5178\u578b\u89c6\u9891\u67e5\u8be2\u4e3a\u4f8b\uff0c\u8be6\u7ec6\u8bf4\u660e\u4e86\u8c13\u8bcd\u6c42\u503c\u987a\u5e8f\u5bf9\u6574\u4f53\u6267\u884c\u65f6\u95f4\u7684\u51b3\u5b9a\u6027\u5f71\u54cd\uff0c\u5e76\u6307\u51fa\u9759\u6001\u4f18\u5316\u5728UDF\u573a\u666f\u4e0b\u7684\u4e24\u4e2a\u81f4\u547d\u5c40\u9650\uff1a\u4e00\u662f\u7edf\u8ba1\u4fe1\u606f\u4e0d\u53ef\u9760\uff0c\u4e8c\u662fUDF\u6267\u884c\u96be\u4ee5\u6269\u5c55\u3002Hydro\u901a\u8fc7\u5f15\u5165AQP\u6846\u67b6\uff0c\u5b8c\u5168\u6d88\u9664\u67e5\u8be2\u4f18\u5316\u9636\u6bb5\u7684\u9884\u4f30\u5f00\u9500\uff0c\u5b9e\u73b0\u201c\u8fb9\u6267\u884c\u8fb9\u4f18\u5316\u201d\u7684\u5168\u65b0\u8303\u5f0f\u3002\u8be5\u65b9\u6cd5\u4e0d\u4ec5\u663e\u8457\u63d0\u5347\u4e86\u67e5\u8be2\u6027\u80fd\uff0c\u8fd8\u4e3a\u63a2\u7d22\u5f0f\u5206\u6790\u3001\u5b9e\u65f6\u89c6\u9891\u5904\u7406\u7b49\u573a\u666f\u63d0\u4f9b\u4e86\u66f4\u5177\u666e\u9002\u6027\u7684\u89e3\u51b3\u65b9\u6848\uff0c\u5177\u6709\u91cd\u8981\u7684\u5b9e\u8df5\u6307\u5bfc\u610f\u4e49\u3002<\/p>\n\n\n\n<p><strong>3. \u80cc\u666f\u4e0e\u52a8\u673a\uff08Background and Motivation\uff09<\/strong><\/p>\n\n\n\n<p>\u89c6\u9891\u6570\u636e\u5e93\u7ba1\u7406\u7cfb\u7edf\uff08VDBMS\uff09\u662fML\u67e5\u8be2\u7684\u5178\u578b\u4ee3\u8868\uff0c\u5176\u6838\u5fc3\u6311\u6218\u5728\u4e8e\u8ba1\u7b97\u5bc6\u96c6\u7684UDF\u5f80\u5f80\u5360\u636e\u67e5\u8be2\u6267\u884c\u65f6\u95f4\u7684\u7edd\u5927\u90e8\u5206\u3002\u73b0\u6709VDBMS\u5de5\u4f5c\u5982VIVA\u901a\u8fc7\u7528\u6237\u63d0\u4f9b\u7684\u5173\u7cfb\u63d0\u793a\u8fdb\u884c\u4f18\u5316\uff0cExSample\u3001FiGO\u3001Chameleon\u5219\u5c1d\u8bd5\u5728\u6267\u884c\u4e2d\u9009\u62e9\u6700\u4f18\u7b97\u6cd5\u6216\u91c7\u6837\u7b56\u7565\uff0c\u4f46\u8fd9\u4e9b\u65b9\u6848\u5927\u591a\u91c7\u7528\u5b9a\u5236\u5316\u6267\u884c\u6846\u67b6\uff0c\u901a\u7528\u6027\u548c\u53ef\u6269\u5c55\u6027\u8f83\u5dee\uff0c\u4e14\u90e8\u5206\u4ee5\u727a\u7272\u67e5\u8be2\u7cbe\u5ea6\u4e3a\u4ee3\u4ef7\u6362\u53d6\u901f\u5ea6\u3002<\/p>\n\n\n\n<p>\u5728\u4f20\u7edf\u5173\u7cfb\u578bDBMS\u4e2d\uff0cEddy\u6846\u67b6\u662f\u6700\u5177\u4ee3\u8868\u6027\u7684AQP\u65b9\u6848\uff0c\u5b83\u5728\u67e5\u8be2\u6267\u884c\u65f6\u6309\u5143\u7ec4\u7c92\u5ea6\u52a8\u6001\u91cd\u6392\u5e8f\u7ba1\u9053\u7b97\u5b50\uff0c\u901a\u8fc7\u8fd0\u884c\u65f6\u7edf\u8ba1\uff08\u5982\u961f\u5217\u957f\u5ea6\u63a8\u65ad\u6267\u884c\u4ee3\u4ef7\u3001\u5f69\u7968\u673a\u5236\u63a8\u65ad\u9009\u62e9\u6027\uff09\u5b9e\u73b0\u81ea\u9002\u5e94\u8def\u7531\u3002\u540e\u7eed\u7684Content-based Routing\u8fdb\u4e00\u6b65\u5f15\u5165\u6570\u636e\u5185\u5bb9\u611f\u77e5\uff0c\u4f46\u8def\u7531\u5f00\u9500\u8f83\u9ad8\u3002Hydro\u5728\u7ee7\u627fEddy\u6846\u67b6\u4f18\u52bf\u7684\u540c\u65f6\uff0c\u9488\u5bf9ML UDF\u7684\u72ec\u7279\u7279\u6027\u8fdb\u884c\u4e86\u6df1\u5ea6\u5b9a\u5236\uff1a\u5b83\u4e0d\u4ec5\u4fdd\u7559\u4e86\u52a8\u6001\u8def\u7531\u80fd\u529b\uff0c\u8fd8\u65b0\u589e\u4e86\u786c\u4ef6\u8d44\u6e90\u611f\u77e5\u548c\u6570\u636e\u611f\u77e5\u8d1f\u8f7d\u5747\u8861\u673a\u5236\uff0c\u5f7b\u5e95\u89e3\u51b3\u4e86UDF\u573a\u666f\u4e0b\u9759\u6001\u4f18\u5316\u96be\u4ee5\u5e94\u5bf9\u7684\u52a8\u6001\u6027\u95ee\u9898\u3002\u8bba\u6587\u5f3a\u8c03\uff0c\u4f20\u7edf\u9759\u6001\u67e5\u8be2\u4f18\u5316\u5df2\u65e0\u6cd5\u6ee1\u8db3UDF\u5bc6\u96c6\u578bML\u67e5\u8be2\u7684\u9700\u6c42\uff0c\u800cHydro\u901a\u8fc7\u8fd0\u884c\u65f6\u76d1\u63a7\u4e0e\u5b9e\u65f6\u8c03\u6574\uff0c\u4e3a\u667a\u80fd\u6570\u636e\u5e93\u7cfb\u7edf\u63d0\u4f9b\u4e86\u5168\u65b0\u7684\u67e5\u8be2\u5904\u7406\u8303\u5f0f\uff0c\u5177\u6709\u91cd\u8981\u7684\u7406\u8bba\u4ef7\u503c\u548c\u5de5\u7a0b\u610f\u4e49\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"269\" src=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/2-5-1024x269.png\"  class=\"wp-image-1188\" srcset=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/2-5-1024x269.png 1024w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/2-5-300x79.png 300w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/2-5-768x202.png 768w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/2-5.png 1323w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" title=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe1\" alt=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe1\" \/><\/figure>\n\n\n\n<p><strong>4. Hydro\u7cfb\u7edf\u8bbe\u8ba1\uff08AQP in Hydro\uff09<\/strong><\/p>\n\n\n\n<p>Hydro\u65e0\u7f1d\u96c6\u6210\u5230EvaDB\u8fd9\u4e00ML-centric\u6570\u636e\u5e93\u7cfb\u7edf\u4e2d\uff0c\u67e5\u8be2\u4f18\u5316\u5668\u5728\u9047\u5230UDF\u8c13\u8bcd\u65f6\u81ea\u52a8\u5207\u6362\u4e3aAQP\u8ba1\u5212\uff0c\u4ec5\u4fdd\u7559\u89c4\u5219-based\u7684\u9759\u6001\u4f18\u5316\uff08\u5982\u8c13\u8bcd\u4e0b\u63a8\u3001\u7ed3\u679c\u7f13\u5b58\u590d\u7528\uff09\u3002AQP\u6267\u884c\u5668\u662f\u6574\u4e2a\u7cfb\u7edf\u7684\u6838\u5fc3\uff0c\u7531Eddy\u8def\u7531\u548cLaminar\u8def\u7531\u4e24\u5927\u7ec4\u4ef6\u6784\u6210\u3002Eddy\u8def\u7531\u8d1f\u8d23\u52a8\u6001\u51b3\u5b9a\u8c13\u8bcd\u6c42\u503c\u987a\u5e8f\uff0cLaminar\u8def\u7531\u5219\u8d1f\u8d23\u76d1\u63a7\u786c\u4ef6\u5229\u7528\u7387\uff08\u5982GPU\u5360\u7528\u7387\uff09\u3001\u81ea\u52a8\u6269\u5c55\u5de5\u4f5c\u7ebf\u7a0b\u6570\u91cf\uff0c\u5e76\u5b9e\u73b0\u6570\u636e\u611f\u77e5\u7684\u8d1f\u8f7d\u5747\u8861\u3002<\/p>\n\n\n\n<p>\u6570\u636e\u4ee5\u8def\u7531\u6279\u6b21\uff08routing batch\uff0c\u9ed8\u8ba410\u884c\uff09\u4e3a\u5355\u4f4d\u5728\u4e2d\u592e\u961f\u5217\u4e2d\u6d41\u52a8\uff1aEddy Pull\u4ece\u5b50\u6267\u884c\u5668\u62c9\u53d6\u6279\u6b21\u5e76\u63d2\u5165\u4e2d\u592e\u961f\u5217\uff1bEddy Router\u6839\u636e\u5f53\u524d\u8def\u7531\u7b56\u7565\u5c06\u6279\u6b21\u5206\u53d1\u81f3\u5404Laminar Router\u7684\u8f93\u5165\u961f\u5217\uff1bLaminar Router\u518d\u6839\u636e\u5b9e\u65f6\u786c\u4ef6\u7edf\u8ba1\u548c\u6570\u636e\u7279\u6027\u9009\u62e9\u6700\u4f18\u5de5\u4f5c\u7ebf\u7a0b\u5e76\u5b8c\u6210\u8d1f\u8f7d\u5206\u914d\u3002\u8c13\u8bcd\u6c42\u503c\u5b8c\u6210\u540e\uff0c\u6279\u6b21\u56de\u5199\u4e2d\u592e\u961f\u5217\uff0c\u5df2\u5b8c\u6210\u6240\u6709\u8c13\u8bcd\u7684\u6279\u6b21\u6700\u7ec8\u8fdb\u5165\u8f93\u51fa\u961f\u5217\u3002\u8be5\u8bbe\u8ba1\u5b9e\u73b0\u4e86\u771f\u6b63\u7684\u201c\u96f6\u9884\u4f30\u3001\u5b9e\u65f6\u81ea\u9002\u5e94\u201d\u67e5\u8be2\u6267\u884c\u6d41\u7a0b\uff0c\u540c\u65f6\u901a\u8fc7\u5143\u6570\u636e\u8ddf\u8e2a\u3001\u9884\u7f6e\u7269\u5316\u3001\u6b7b\u9501\u9884\u9632\u7b49\u5de5\u7a0b\u4f18\u5316\uff0c\u786e\u4fdd\u7cfb\u7edf\u5728\u9ad8\u5e76\u53d1UDF\u573a\u666f\u4e0b\u7684\u7a33\u5b9a\u6027\u548c\u9ad8\u6548\u6027\uff0c\u4e3aML\u67e5\u8be2\u5904\u7406\u63d0\u4f9b\u4e86\u575a\u5b9e\u7684\u7cfb\u7edf\u57fa\u7840\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"334\" src=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/3-3-1024x334.png\"  class=\"wp-image-1189\" srcset=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/3-3-1024x334.png 1024w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/3-3-300x98.png 300w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/3-3-768x251.png 768w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/3-3.png 1314w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" title=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe2\" alt=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe2\" \/><\/figure>\n\n\n\n<p><strong>5. Eddy\u4f7f\u7528\u6848\u4f8b\u4e0e\u8def\u7531\u7b56\u7565\uff08Use Cases of Eddy\uff09<\/strong><\/p>\n\n\n\n<p>\u8bba\u6587\u901a\u8fc7\u56db\u4e2a\u5b9e\u9645\u7528\u4f8b\u7cfb\u7edf\u5c55\u793a\u4e86Hydro\u7684\u5b9e\u7528\u4ef7\u503c\uff0c\u5176\u4e2dUC1\uff08\u89c6\u9891\u4e2d\u5bfb\u627e\u9ed1\u5927\u4e39\u72ac\uff09\u548cUC2\uff08\u5e26\u7f13\u5b58\u590d\u7528\u7684\u63a2\u7d22\u5f0f\u5206\u6790\uff09\u6700\u5177\u4ee3\u8868\u6027\u3002\u5728UC1\u4e2d\uff0c\u4f5c\u8005\u8be6\u7ec6\u5bf9\u6bd4\u4e86\u4e09\u79cd\u8def\u7531\u7b56\u7565\uff1a\u9009\u62e9\u6027\u9a71\u52a8\u3001\u5f97\u5206\u9a71\u52a8\uff08\u57fa\u4e8e1-\u9009\u62e9\u6027\uff09\u548c\u4ee3\u4ef7\u9a71\u52a8\u3002\u5b9e\u9a8c\u8868\u660e\uff0c\u5f53\u4e24\u4e2a\u8c13\u8bcd\u53ef\u5e76\u884c\u6267\u884c\uff08\u4e00\u4e2a\u8fd0\u884c\u4e8eCPU\u3001\u4e00\u4e2a\u8fd0\u884c\u4e8eGPU\uff09\u65f6\uff0c\u5355\u7eaf\u7684\u9009\u62e9\u6027\u6216\u5f97\u5206\u9a71\u52a8\u7b56\u7565\u4e0d\u518d\u6700\u4f18\uff0c\u4ee3\u4ef7\u9a71\u52a8\u7b56\u7565\u80fd\u66f4\u597d\u5730\u5229\u7528\u8ba1\u7b97\u91cd\u53e0\uff0c\u663e\u8457\u63d0\u5347\u6574\u4f53\u6027\u80fd\u3002<\/p>\n\n\n\n<p>UC2\u8fdb\u4e00\u6b65\u5f15\u5165\u4e86\u7f13\u5b58\u611f\u77e5\uff08Reuse-Aware\uff09\u8def\u7531\u673a\u5236\uff1a\u5728\u63a2\u7d22\u5f0f\u67e5\u8be2\u573a\u666f\u4e2d\uff0c\u4e0d\u540c\u8c13\u8bcd\u7684\u7f13\u5b58\u547d\u4e2d\u7387\u968f\u89c6\u9891\u6bb5ID\u52a8\u6001\u53d8\u5316\u3002Hydro\u901a\u8fc7\u5b9e\u65f6\u76d1\u63a7\u7f13\u5b58\u547d\u4e2d\u7387\uff0c\u52a8\u6001\u8c03\u6574\u6bcf\u4e2a\u8c13\u8bcd\u7684\u5b9e\u9645\u6267\u884c\u4ee3\u4ef7\uff0c\u4ece\u800c\u5728\u590d\u7528Q1\u3001Q2\u7f13\u5b58\u7ed3\u679c\u7684Q3\u67e5\u8be2\u4e2d\u81ea\u52a8\u9009\u62e9\u6700\u4f18\u8c13\u8bcd\u987a\u5e8f\uff0c\u5b9e\u73b0\u4e861.25\u00d7\u7684\u989d\u5916\u52a0\u901f\u3002\u8fd9\u4e9b\u7528\u4f8b\u5145\u5206\u9a8c\u8bc1\u4e86Hydro\u5728\u5b9e\u9645ML\u67e5\u8be2\u573a\u666f\u4e2d\u7684\u9002\u5e94\u6027\u548c\u9ad8\u6548\u6027\uff0c\u540c\u65f6\u4e5f\u4e3a\u540e\u7eed\u7814\u7a76\u63d0\u4f9b\u4e86\u5b9d\u8d35\u7684\u7cfb\u7edf\u6d1e\u89c1\u548c\u5b9e\u73b0\u53c2\u8003\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"686\" height=\"819\" src=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/4-2.png\"  class=\"wp-image-1190\" srcset=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/4-2.png 686w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/4-2-251x300.png 251w\" sizes=\"auto, (max-width: 686px) 100vw, 686px\" title=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe3\" alt=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe3\" \/><\/figure>\n\n\n\n<p><strong>6. \u5b9e\u9a8c\u8bc4\u4f30\u4e0e\u6027\u80fd\u5206\u6790\uff08Evaluation\uff09<\/strong><\/p>\n\n\n\n<p>\u5b9e\u9a8c\u5728\u914d\u5907AMD EPYC 32\u6838CPU + NVIDIA A40 GPU\u7684\u670d\u52a1\u5668\u4e0a\u8fdb\u884c\uff0c\u57fa\u7ebf\u7cfb\u7edf\u4e3aEvaDB\u7684\u9ed8\u8ba4\u9759\u6001\u4f18\u5316\u6a21\u5f0f\u3002\u9488\u5bf9UC1\uff0c\u4ee3\u4ef7\u9a71\u52a8\u8def\u7531\u8f83\u65e0\u91cd\u6392\u5e8f\u83b7\u5f971.70\u00d7\u52a0\u901f\uff0c\u4f18\u4e8e\u5f97\u5206\u9a71\u52a8\uff081.68\u00d7\uff09\u548c\u9009\u62e9\u6027\u9a71\u52a8\uff081.52\u00d7\uff09\u3002\u654f\u611f\u6027\u5206\u6790\u8fdb\u4e00\u6b65\u9a8c\u8bc1\u4e86\u5728\u4e0d\u540c\u9009\u62e9\u6027\u4e0e\u4ee3\u4ef7\u7ec4\u5408\u4e0b\uff0c\u4ee3\u4ef7\u9a71\u52a8\u7b56\u7565\u59cb\u7ec8\u4fdd\u6301\u6700\u4f73\u9c81\u68d2\u6027\u3002<\/p>\n\n\n\n<p>UC2\u4e2d\uff0cReuse-Aware\u8def\u7531\u5145\u5206\u5229\u7528\u90e8\u5206\u7f13\u5b58\uff0c\u8fdb\u4e00\u6b65\u5c06\u6027\u80fd\u63d0\u5347\u81f31.25\u00d7\u3002\u8bba\u6587\u8fd8\u5bf9\u7cfb\u7edf\u5404\u7ec4\u4ef6\u7684\u5f00\u9500\u8fdb\u884c\u4e86\u7ec6\u81f4\u5256\u6790\uff0c\u5f3a\u8c03Hydro\u7684\u6240\u6709\u4f18\u5316\u5747\u5728\u7cfb\u7edf\u5c42\u9762\u5b9e\u73b0\uff0c\u4e0d\u5f71\u54cd\u67e5\u8be2\u7ed3\u679c\u7cbe\u5ea6\uff0c\u4e14\u80fd\u4e0e\u73b0\u6709ML\u67e5\u8be2\u4f18\u5316\u6280\u672f\uff08\u5982\u7ed3\u679c\u7f13\u5b58\uff09\u65e0\u7f1d\u534f\u540c\u3002\u6574\u4f53\u800c\u8a00\uff0cHydro\u5728\u56db\u4e2a\u591a\u6837\u5316\u7528\u4f8b\u4e2d\u6700\u9ad8\u5b9e\u73b0\u4e8611.52\u00d7\u7684\u52a0\u901f\u6548\u679c\uff0c\u5145\u5206\u8bc1\u660e\u4e86AQP\u6280\u672f\u5728UDF\u5bc6\u96c6\u578bML\u67e5\u8be2\u573a\u666f\u4e0b\u7684\u5de8\u5927\u6f5c\u529b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"672\" height=\"597\" src=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/5-2.png\"  class=\"wp-image-1191\" srcset=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/5-2.png 672w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/5-2-300x267.png 300w\" sizes=\"auto, (max-width: 672px) 100vw, 672px\" title=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe4\" alt=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe4\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"511\" src=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/6-3-1024x511.png\"  class=\"wp-image-1192\" srcset=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/6-3-1024x511.png 1024w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/6-3-300x150.png 300w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/6-3-768x383.png 768w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/6-3.png 1352w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" title=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe5\" alt=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe5\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"681\" height=\"309\" src=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/7-4.png\"  class=\"wp-image-1194\" srcset=\"https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/7-4.png 681w, https:\/\/www.ndnlab.com\/wp-content\/uploads\/2026\/03\/7-4-300x136.png 300w\" sizes=\"auto, (max-width: 681px) 100vw, 681px\" title=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe6\" alt=\"Hydro: Adaptive Query Processing of ML Queries\u63d2\u56fe6\" \/><\/figure>\n\n\n\n<p><strong>7. \u8d21\u732e\u4e0e\u7ed3\u8bba\uff08Contributions and Conclusion\uff09<\/strong><\/p>\n\n\n\n<p>Hydro\u7684\u4e3b\u8981\u8d21\u732e\u5305\u62ec\u4e09\u65b9\u9762\uff1a\u4e00\u662f\u63d0\u51fa\u4e86\u4e00\u79cdML-centric DBMS\u6846\u67b6\uff0c\u5c06AQP\u6280\u672f\u9002\u914d\u5230UDF\u5bc6\u96c6\u578b\u67e5\u8be2\u573a\u666f\uff0c\u8bbe\u8ba1\u4e86ML\u67e5\u8be2\u4e13\u5c5e\u7684\u8def\u7531\u7b56\u7565\u548cUDF\u7edf\u8ba1\u52a8\u6001\u91c7\u96c6\u673a\u5236\uff1b\u4e8c\u662f\u5bf9Eddy\u6846\u67b6\u8fdb\u884c\u6269\u5c55\uff0c\u65b0\u589eLaminar\u81ea\u52a8\u6269\u5c55\u7ec4\u4ef6\u53ca\u6570\u636e\u611f\u77e5\u8d1f\u8f7d\u5747\u8861\u7b56\u7565\uff0c\u5b9e\u73b0\u786c\u4ef6\u8d44\u6e90\u7684\u9ad8\u6548\u5229\u7528\uff1b\u4e09\u662f\u901a\u8fc7\u56db\u4e2a\u771f\u5b9e\u7528\u4f8b\u5168\u9762\u9a8c\u8bc1\u4e86\u7cfb\u7edf\u6027\u80fd\uff0c\u6700\u9ad8\u83b7\u5f9711.52\u00d7\u52a0\u901f\u6548\u679c\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u7ed3\u8bba\u660e\u786e\u6307\u51fa\uff0c\u4f20\u7edf\u9759\u6001\u67e5\u8be2\u4f18\u5316\u5df2\u65e0\u6cd5\u6ee1\u8db3UDF\u5bc6\u96c6\u7684ML\u67e5\u8be2\u9700\u6c42\uff0c\u800cHydro\u901a\u8fc7\u8fd0\u884c\u65f6\u76d1\u63a7\u4e0e\u52a8\u6001\u8c03\u6574\uff0c\u8bc1\u660e\u4e86AQP\u5728\u8fd9\u4e00\u573a\u666f\u4e0b\u7684\u5de8\u5927\u6f5c\u529b\u3002\u8be5\u5de5\u4f5c\u4e0d\u4ec5\u4e3a\u89c6\u9891\u6570\u636e\u5e93\u3001\u63a2\u7d22\u5f0f\u5206\u6790\u7b49\u5e94\u7528\u63d0\u4f9b\u4e86\u53ef\u6269\u5c55\u7684\u67e5\u8be2\u5904\u7406\u6846\u67b6\uff0c\u4e5f\u4e3a\u672a\u6765\u667a\u80fd\u6570\u636e\u5e93\u7cfb\u7edf\u7684\u8bbe\u8ba1\u63d0\u4f9b\u4e86\u91cd\u8981\u53c2\u8003\uff0c\u5177\u6709\u5e7f\u6cdb\u7684\u5b66\u672f\u4ef7\u503c\u548c\u5de5\u7a0b\u610f\u4e49\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>1. \u6458\u8981\uff08Abstract\uff09 Hydro\u63d0\u51fa\u4e86\u4e00\u79cd\u9762\u5411\u673a\u5668\u5b66\u4e60\u67e5\u8be2\u7684\u6570\u636e\u5e93\u7ba1\u7406\u7cfb\u7edf\uff08ML-centric DBMS\uff09\uff0c\u6838\u5fc3\u521b\u65b0\u5728\u4e8e\u5c06\u81ea\u9002\u5e94\u67e5\u8be2\u5904\u7406\uff08Adaptive Query Processing\uff0cAQP\uff09\u6280\u672f\u5168\u9762\u5f15\u5165UDF\u5bc6\u96c6\u578b\u67e5\u8be2\u573a\u666f\uff0c\u4ece\u800c\u5f7b\u5e95\u89e3\u51b3\u4f20\u7edf\u9759\u6001\u67e5\u8be2\u4f18\u5316\u5728ML\u67e5\u8be2\u4e2d\u7684\u4e24\u5927\u6838\u5fc3\u74f6\u9888\u3002\u9996\u5148\uff0cUDF\u7684\u6267\u884c\u4ee3\u4ef7\u548c\u9009\u62e9\u6027\u9ad8\u5ea6\u4f9d\u8d56\u8f93\u5165\u6570\u636e\uff0c\u4e14\u53d7\u7cfb\u7edf\u7ea7\u7f13\u5b58\u7b49\u52a8\u6001\u56e0\u7d20\u5f71\u54cd\uff0c\u5bfc\u81f4\u67e5\u8be2\u4f18\u5316\u9636\u6bb5\u7684\u9884\u4f30\u7edf\u8ba1\u4fe1\u606f\u6781\u4e0d\u53ef\u9760\uff1b\u5176\u6b21\uff0c\u6700\u4f18\u67e5\u8be2\u8ba1\u5212\u662f\u6570\u636e\u4f9d\u8d56\u7684\uff0c\u9759\u6001\u8ba1\u5212\u65e0\u6cd5\u5728\u6267\u884c\u8fc7\u7a0b\u4e2d\u52a8\u6001\u8c03 &hellip; <a href=\"https:\/\/www.ndnlab.com\/?p=1185\">\u7ee7\u7eed\u9605\u8bfb <span class=\"meta-nav\">&rarr;<\/span><\/a><\/p>\n","protected":false},"author":5,"featured_media":1187,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5,24,1],"tags":[],"class_list":["post-1185","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-rengongzhineng","category-24","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/posts\/1185","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\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1185"}],"version-history":[{"count":1,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/posts\/1185\/revisions"}],"predecessor-version":[{"id":1195,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/posts\/1185\/revisions\/1195"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=\/wp\/v2\/media\/1187"}],"wp:attachment":[{"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1185"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1185"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ndnlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1185"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}