[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-opensearch-vector-search-benchmark-5-parts-zh":3,"article-related-opensearch-vector-search-benchmark-5-parts-zh":32,"series-industry-d1753385-8c03-4dec-b939-e5ca8bae9030":76},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":24,"views":28,"created_at":29,"published_at":30,"topic_cluster_id":31},"d1753385-8c03-4dec-b939-e5ca8bae9030","opensearch-vector-search-benchmark-5-parts-zh","OpenSearch 向量搜尋基準的 5 種跑法","\u003Cp>OpenSearch \u003Ca href=\"\u002Fnews\u002Fvector-databases-work-in-production-zh\">向量\u003C\u002Fa>搜尋要怎麼做基準測試，才不必每次都重建索引？\u003C\u002Fp>\u003Cp data-speakable=\"summary\">這篇整理 OpenSearch 向量搜尋的 5 種基準流程，幫你分清索引、訓練、合併和搜尋該測哪一段。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>測什麼\u003C\u002Fth>\u003Cth>關鍵限制\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>No-train\u003C\u002Ftd>\u003Ctd>索引與搜尋\u003C\u002Ftd>\u003Ctd>適用不需要訓練的方法\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>No-train index only\u003C\u002Ftd>\u003Ctd>只測寫入\u003C\u002Ftd>\u003Ctd>不包含搜尋\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Amazon OpenSearch Serverless\u003C\u002Ftd>\u003Ctd>Serverless 向量集合\u003C\u002Ftd>\u003Ctd>不做 refresh 與 warmup\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Force merge\u003C\u002Ftd>\u003Ctd>段合併成本\u003C\u002Ftd>\u003Ctd>大資料集可能很貴\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Search\u003C\u002Ftd>\u003Ctd>既有索引的查詢表現\u003C\u002Ftd>\u003Ctd>包含 warmup\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>1. No-train test：免訓練的標準起點\u003C\u002Fh2>\u003Cp>這一項適合不需要訓練階段的向量方法。你可以直接定義 engine、space type 和其他方法參數，再看索引與查詢在負載下的表現。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783850566022-b79s.png\" alt=\"OpenSearch 向量搜尋基準的 5 種跑法\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>如果你的目標是比較不同配置，而不是先花時間訓練模型，這就是最省事的\u003Ca href=\"\u002Fnews\u002Fgpt-56-full-suite-work-entry-openai-zh\">入口\u003C\u002Fa>。它也支援 HDF5 和 BIG-ANN 格式，方便重複使用同一批資料。\u003C\u002Fp>\u003Cul>\u003Cli>適合直接建立索引的方法\u003C\u002Fli>\u003Cli>可比較 engine 與 space type\u003C\u002Fli>\u003Cli>支援 HDF5、BIG-ANN 資料集\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>2. No-train index only：只看寫入速度\u003C\u002Fh2>\u003Cp>這個變體把搜尋拿掉，專心看 ingestion。若你的問題是「這個向量索引到底建得多快」，它比完整流程更容易讀出\u003Ca href=\"\u002Fnews\u002Fllm-benchmarks-not-enough-2026-zh\">答案\u003C\u002Fa>。\u003C\u002Fp>\u003Cp>它特別適合 bulk load 調校、shard 大小測試，或不同 client 數量的實驗。相關參數像 \u003Ccode>target_index_bulk_size\u003C\u002Fcode>、\u003Ccode>target_index_bulk_index_clients\u003C\u002Fcode>、\u003Ccode>target_index_dimension\u003C\u002Fcode>，都能直接控制寫入路徑。\u003C\u002Fp>\u003Ccode>target_index_bulk_size: 1000\ntarget_index_bulk_index_clients: 4\ntarget_index_dimension: 768\u003C\u002Fcode>\u003Ch2>3. Amazon OpenSearch Serverless：對準受限環境\u003C\u002Fh2>\u003Cp>這個版本是給 \u003Ca href=\"https:\u002F\u002Fdocs.aws.amazon.com\u002Fopensearch-service\u002Flatest\u002Fdeveloperguide\u002Fserverless.html\">Amazon OpenSearch Serverless\u003C\u002Fa> 的向量搜尋集合用的。概念上仍是 no-train，但流程要跟 Serverless 的能力範圍對齊。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783850566994-1pkw.png\" alt=\"OpenSearch 向量搜尋基準的 5 種跑法\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>最重要的差異是它不包含 refresh 和 warmup，因為這些操作不支援用在向量搜尋集合上。若你要驗證的是 Serverless 部署，這種測法才不會把平台做不到的步驟算進去。\u003C\u002Fp>\u003Cul>\u003Cli>只用於 Serverless 向量集合\u003C\u002Fli>\u003Cli>不做 refresh\u003C\u002Fli>\u003Cli>不做 warmup\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>4. Force merge：看段合併代價\u003C\u002Fh2>\u003Cp>force merge 的重點是把向量索引壓到目標 segment 數，觀察合併成本。對大資料集來說，這一步可能很重，所以不適合硬塞進每一次基準流程。\u003C\u002Fp>\u003Cp>如果你想知道段數縮減後，對後續搜尋或儲存佈局有什麼影響，這一項就很有用。文件把 \u003Ccode>target_index_max_num_segments\u003C\u002Fcode> 和 \u003Ccode>target_index_force_merge_timeout\u003C\u002Fcode> 都列得很清楚，方便你控制目標與等待時間。\u003C\u002Fp>\u003Cul>\u003Cli>先壓 segment，再看後續效果\u003C\u002Fli>\u003Cli>適合維運或容量評估\u003C\u002Fli>\u003Cli>大資料集成本可能很高\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>5. Search：直接量查詢表現\u003C\u002Fh2>\u003Cp>search 程序是拿已建立好的向量索引來測查詢。這樣可以避開重建索引的時間，特別適合資料量大、載入成本高的情境。\u003C\u002Fp>\u003Cp>它還會包含 warmup，讓結果更接近穩態查詢表現。文件示例顯示 prod-queries 的 median latency 為 3.43393 ms，mean throughput 為 213.85 ops\u002Fs，已經能看出它能量化到什麼程度。\u003C\u002Fp>\u003Ccode>query_k: 10\nsearch_clients: 8\ntarget_throughput: 10\ntime_period: 900\u003C\u002Fcode>\u003Ch2>哪種適合你\u003C\u002Fh2>\u003Cp>如果你的方法不需要訓練，而且想先做一個通用比較，選 no-train。若你只想知道寫入有多快，就用 no-train index only；若你在驗證 Serverless 部署，則選 Amazon OpenSearch Serverless。\u003C\u002Fp>\u003Cp>要看段合併成本就跑 force merge；如果索引已經存在，重點是查詢延遲或吞吐，search 最省時間，也最接近真實使用情境。\u003C\u002Fp>","5 種 OpenSearch 向量搜尋基準跑法，從免訓練索引、只測寫入，到 Serverless、force merge 與搜尋壓測。","docs.opensearch.org","https:\u002F\u002Fdocs.opensearch.org\u002Flatest\u002Fbenchmark\u002Fworkloads\u002Fvectorsearch\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783850566022-b79s.png","industry","zh","dea6021b-f740-4ccb-b3b6-6f2c09c64414",[17,18,19,20,21,22,23],"OpenSearch","vector search","benchmark","serverless","force merge","search latency","indexing throughput",[25,26,27],"免訓練方法先用 no-train，能同時看索引與搜尋。","只想量寫入速度時，用 no-train index only。","Serverless、force merge、search 各自對應不同驗證目的。",0,"2026-07-12T10:02:22.269045+00:00","2026-07-12T10:02:22.26+00:00","de71f225-219b-4aa2-a3ed-96d3f7eae15e",{"tags":33,"relatedLang":35,"relatedPosts":39},[34],{"name":19,"slug":19},{"id":15,"slug":36,"title":37,"language":38},"opensearch-vector-search-benchmark-5-parts-en","OpenSearch’s vector search benchmark in 5 parts","en",[40,46,52,58,64,70],{"id":41,"slug":42,"title":43,"cover_image":44,"image_url":44,"created_at":45,"category":13},"6e790897-c9af-402c-a928-f2b0cc02f4e6","vector-databases-work-in-production-zh","4 種能上線的向量資料庫選擇","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783846963245-35py.png","2026-07-12T09:02:23.058273+00:00",{"id":47,"slug":48,"title":49,"cover_image":50,"image_url":50,"created_at":51,"category":13},"e5ae86b4-4434-48d4-86b4-146f609ce0a2","eu-ai-act-hits-business-systems-aug-2-2026-zh","歐盟 AI 法案上路前，企業先看這 5 件事","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783845168794-qyhi.png","2026-07-12T08:32:24.43396+00:00",{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"bc30f927-a6c9-4cdd-b734-6e8cd0b8265a","us-ai-law-2026-compliance-overview-zh","2026 美國 AI 法規控管地圖","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783843397232-2oow.png","2026-07-12T08:02:50.480302+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"4d5e43ec-56bf-4ddf-aca0-e3b31065f132","webx-2026-agenda-stablecoins-ai-zh","WebX 2026 將穩定幣與 AI 推上主舞台","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783841563555-spp1.png","2026-07-12T07:32:24.035669+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"4647fcd1-fee7-4819-958a-73a92587227a","gpt-56-full-suite-work-entry-openai-zh","GPT-5.6 全家桶不是炫技，是 OpenAI 的工作入口","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783837980636-jmwn.png","2026-07-12T06:32:32.520158+00:00",{"id":71,"slug":72,"title":73,"cover_image":74,"image_url":74,"created_at":75,"category":13},"8ffc6905-3e5a-4236-a031-bda41472e78d","half-price-ai-real-frontier-smarter-models-zh","半價 AI 才是主戰場，不是更聰明的模型","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783801974628-606c.png","2026-07-11T20:32:23.661553+00:00",[77,82,87,92,97,102,107,112,117,122],{"id":78,"slug":79,"title":80,"created_at":81},"ee073da7-28b3-4752-a319-5a501459fb87","ai-in-2026-what-actually-matters-now-zh","2026 AI 真正重要的事","2026-03-26T07:09:12.008134+00:00",{"id":83,"slug":84,"title":85,"created_at":86},"83bd1795-8548-44c9-9a7e-de50a0923f71","trump-ai-framework-power-speech-state-preemption-zh","川普 AI 框架瞄準電力、言論與州權","2026-03-26T07:12:18.695466+00:00",{"id":88,"slug":89,"title":90,"created_at":91},"ea6be18b-c903-4e54-97b7-5f7447a612e0","nvidia-gtc-2026-big-ai-announcements-zh","NVIDIA GTC 2026 重點拆解","2026-03-26T07:14:26.62638+00:00",{"id":93,"slug":94,"title":95,"created_at":96},"4bcec76f-4c36-4daa-909f-54cd702f7c93","claude-users-spreading-out-and-getting-better-zh","Claude 用戶更分散，也更會用","2026-03-26T07:22:52.325888+00:00",{"id":98,"slug":99,"title":100,"created_at":101},"bd903b15-2473-4178-9789-b7557816e535","openclaw-raises-hard-question-for-ai-models-zh","OpenClaw 逼問 AI 模型價值","2026-03-26T07:24:54.707486+00:00",{"id":103,"slug":104,"title":105,"created_at":106},"eeac6b9e-ad9d-4831-8eec-8bba3f9bca6a","gap-google-gemini-checkout-fashion-search-zh","Gap 把結帳搬進 Gemini","2026-03-26T07:28:23.937768+00:00",{"id":108,"slug":109,"title":110,"created_at":111},"0740e53f-605d-4d57-8601-c10beb126f3c","google-pushes-gemini-transition-to-march-2026-zh","Google 把 Gemini 轉換延到 2026 年 3…","2026-03-26T07:30:12.825269+00:00",{"id":113,"slug":114,"title":115,"created_at":116},"e660d801-2421-4529-8fa9-86b82b066990","metas-llama-4-benchmark-scandal-gets-worse-zh","Meta Llama 4 分數風波又擴大","2026-03-26T07:34:21.156421+00:00",{"id":118,"slug":119,"title":120,"created_at":121},"183f9e7c-e143-40bb-a6d5-67ba84a3a8bc","accenture-mistral-ai-sovereign-enterprise-deal-zh","Accenture 攜手 Mistral AI 賣主權 AI","2026-03-26T07:38:14.818906+00:00",{"id":123,"slug":124,"title":125,"created_at":126},"191d9b1b-768a-478c-978c-dd7431a38149","mistral-ai-faces-its-hardest-year-yet-zh","Mistral AI 迎來最硬的一年","2026-03-26T07:40:23.716374+00:00"]