多樣性算力需求下，業內大咖共話操作系統產業未來發展

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"隨着越來越多的新技術興起，算力的需求出現了爆發式增長且變得更加多樣化。在異構、算力多樣性逐漸成爲行業新需求的背景下，操作系統產業的發展也來到了新的路口。在 2021 年 11 月 9 日舉行的"},{"type":"link","attrs":{"href":"https:\/\/www.openeuler.org\/zh\/interaction\/summit-list\/summit2021\/","title":"xxx","type":null},"content":[{"type":"text","text":"操作系統產業峯會 2021 & openEuler Summit 2021 "}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"普華基礎軟件分論壇圓桌對話環節，來自操作系統、芯片、數據庫、中間件及應用的產業鏈上下游的多位大咖齊聚，共同探討操作系統產業的未來發展。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"圓桌對話嘉賓："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"普華基礎軟件事業部總經理 沈翔（主持人）"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"鯤鵬計算領域副總裁 張立鵬"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"人大金倉高級副總裁 冷建全"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"慧點科技副總裁 陳科"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"申泰公司市場總監 張雲橋"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"金蝶天燕客戶服務中心副總經理 姚玉鵬"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"飛騰公司操作系統部總監 陳鮑孜"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/fe\/ae\/fed676aa5363e4425580e45af3yy9bae.jpeg","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"多樣性算力給基礎軟件產業帶來的變化與挑戰"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"張立鵬認爲，隨着整體數據結構不斷變化，算力逐漸在不同領域有着越來越多的應用，同時在處理器技術不斷進步的當下，可以預見的是，"},{"type":"link","attrs":{"href":"https:\/\/s.geekbang.org\/search\/c=0\/k=%E5%A4%9A%E6%A0%B7%E6%80%A7%E7%AE%97%E5%8A%9B\/t=","title":"xxx","type":null},"content":[{"type":"text","text":"多樣性算力"}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"必然會成爲重要的趨勢。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"“站在芯片角度來講，以前更多的是面向文本或者結構化的數據，所有的計算都是圍繞 CPU 進行的。現在數據整體結構發生了變化，逐漸也出現了像 GPU、DPU 不同的計算形態。業界一些大的互聯網公司、算力公司，可能也逐漸地在向不同的算力、多樣化的算力方向嘗試。”張立鵬表示。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"陳鮑孜則站在計算機體系結構、處理器的角度進行分析。在他看來，當前在學術界以及產業界已經形成的共識是，計算機體系結構一定朝着通用配專用的方向發展。“在 2019 年的時候，圖靈獎獲得者、計算機體系結構經典教材的撰寫者曾發表一篇文章，大致意思就是計算機體系結構的黃金時代，未來發展一定是領域特定的體系結構和領域特定編程語言融合的體系發展，傳統的摩爾定律、功耗都會受到一定限制。要想提升芯片的能耗、性能、生態等，必須從這些方面入手進行擴展。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"而作爲數據存儲和計算的核心，數據庫本身面臨的計算越來越豐富，也有越來越多新的模型數據庫在庫內存儲。冷建全認爲，這些新型運算有一個非常大的特點：計算密集。傳統數據庫一般用做事務處理，現在多模計算做的是海量數據複雜計算，包括大數據分析、機器學習都是計算密集型的。可以看到的是，異構計算在數據庫領域的應用是非常重要的一個趨勢。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"“相對 GPU，傳統 CPU 的計算能力相對偏弱。從性能角度來講，傳統做優化基本都是在軟件的算法層面做優化，現在數據庫裏一些常見的分析操作都可以在 GPU 上做運算，包括 FPGA 這些新型硬件。這些硬件跟數據庫軟件算法結合，會釋放更多數據庫算力。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在應用軟件層面上，用戶在算力需求上也有自己特有的理解。陳科介紹道，首先，不同的應用軟件會有不同的功能組件，每個組件需要的算力模式不太一樣。在應用軟件逐步走入智能化時代，有越來越多的用戶對一些智能化的場景提出要求，軟件在做一些人工智能處理的時候，其實是需要更多的計算量的。其次，很多大型集團性企業對大數據量、大用戶量，以及分佈式微服務架構，也需要更大的計算需求，做分佈式部署時候也會存在混合應用架構。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在中間件層面上，姚玉鵬認爲現在的基礎軟件領域和過去已然大不相同。過去更多是巨石應用，單一的 Java EE 應用部署在應用服務器上，和底層的 JDK 或者 CPU 進行線程、IO 等調優。但隨着基礎軟件行業快速發展，出現分佈式和上雲需求後，不單是要進行單一單點性能調優，還要和雲平臺本身及其內部的所有組件形成整體的解決方案，對底層算力和優化手段的要求也是不一樣的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/82\/90\/823894757dc97045a779cfc27867c190.jpeg","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"多算力場景下，面向未來的操作系統是什麼樣的？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在多樣性算力需求已成趨勢的背景下，大家對於面向未來的操作系統也提出了新的期望。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"張立鵬認爲，操作系統作爲和硬件最直接交互的基礎軟件，首先，需要屏蔽所有硬件參與，通過南北向兼容性擴展以及基礎應用包的豐富，讓用戶可以無感知地在不同的 CPU 上跑應用。“不同的 CPU 設計是不一樣的，x86 單核能力很強，ARM 單核能力相對沒那麼強，但是核多，操作系統要做的就是在不同的 CPU 之間把業務調度好。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"其次，操作系統需要通過編譯器優化、底層加速器的構建，在不同的 CPU 架構下將性能發揮出來。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"真正做成一個產品往往需要很多標準，這些標準很多是由操作系統廠商參與制定的，在陳鮑孜看來，能不能做出一個生態友好、能快速適應產業鏈的 CPU，在很大程度上取決於是否有一個權威、研發實力強勁的操作系統，並且對業界的接口有標準的解釋。“有時候制定的標準不僅僅是硬件要怎麼做，而是軟件要怎麼用，硬件必須做成什麼樣子，其中大量信息是來自操作系統，包括我們也可以看到一些指令集的加入，也是因爲上層系統對這些體系結構有要求，纔會加入進來。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"另一方面，未來的產品會更多的將加速器集成在系統裏，加速器的使用完全依賴於上層軟件的配合，如果軟件寫得不夠好，或者軟件平臺本身搭配得不夠好，顯然很多加速器都是用不起來的。“從這個角度來講，這些加速部件能不能很好地起到作用，或者能不能支撐應用完成任務，很大程度上需要軟件的配合。”陳鮑孜提到。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"至於操作系統最根本的責任，沈翔認爲，對下，在硬件、資源上能夠自由調配，平滑操作；對上，能夠讓業務順暢地運行，並且能適用於各種場景。"}]}]}