xxxfreesexmoves-haodiaocao这里只有精品视频-欧美性受黑人性爽-欧美性受xxxxxx黑人xyx性爽|www.jsyyzsb.com

    全書分為: 靈活的數據布局、應用感知及可視化存儲智能、存儲類芯片、儲海鉤沉、集群和多控制器、傳統存儲系統、新興存儲系統、大話光存儲系統、體系結構、IO協議棧及性能分析、存儲軟件、固態存儲幾個大章, 其中每章又有多個小節。每一個小節都是一個獨立的課題。

    冬瓜哥對技術的追求已經到了“癡迷”的境界，與10年前相比，文筆解析更為到位，技術理解更為精準。其公眾號的每篇文章，都是存儲業界風向標。

　　冬瓜哥（張冬），現任某半導體公司系統架構師，著有《大話存儲》系列圖書。存儲領域技術專家和布道者。

第一章 靈活的數據布局 ·········································································1 
1.1 Raid1.0和Raid1.5 ······························································································2 
1.2 Raid5EE和Raid2.0 ·····························································································4 
1.3 Lun2.0/SmartMotion ························································································13 
第二章 應用感知及可視化存儲智能 ·····················································23 
2.1 應用感知精細化自動存儲分層······································································25 
2.2 應用感知精細化SmartMotion ········································································27 
2.3 應用感知精細化QoS ······················································································28 
2.4 產品化及可視化展現······················································································31 
2.5 包裝概念制作PPT ···························································································43 
2.6 評浪潮“活性”存儲概念··············································································49 
第三章 存儲類芯片 ··············································································53 
3.1 通道及Raid控制器架構 ··················································································54 
3.2 SAS Expander架構 ··························································································60 
第四章 儲海鉤沉 ··················································································65 
4.1 你絕對想不到的兩種高格調存儲器······························································66 
4.2 JBOD里都有什么····························································································70 


4.3 Raid4校驗盤之殤 ····························································································72 
4.4 為什么說Raid卡是臺小電腦 ··········································································73 
4.5 為什么Raid卡電池被換為超級電容 ······························································74 
4.6 固件和微碼到底什么區別··············································································75 
4.7 FC成環器內部真的是個環嗎 ·········································································76 
4.8 為什么說SAS、FC對CPU耗費比TCPIP+以太網低 ····································77 
4.9 雙控存儲之間的心跳線都跑了哪些流量······················································78 
第五章集群和多控制器 ······································································· 79 
5.1 淺談雙活和多路徑··························································································80 
5.2 “淺”談容災和雙活數據中心（上）··························································82 
5.3 “淺”談容災和雙活數據中心（下）··························································87 
5.4 集群文件系統架構演變深度梳理圖解··························································96 
5.5 從多控緩存管理到集群鎖············································································107 
5.6 共享式與分布式各論····················································································115 
5.7 “冬瓜哥畫PPT”雙活是個坑 ·····································································118 
第六章傳統存儲系統 ········································································· 121 
6.1 與存儲系統相關的一些基本話題分享························································122 
6.2 高端存儲系統江湖風云錄！········································································133 
6.3 驚了！原來高端存儲架構是這樣演進的！················································145 
6.4 傳統高端存儲系統把數據緩存集中外置一石三鳥····································155 
6.5 傳統外置存儲已近黃昏················································································156 
6.6 存儲圈老炮大戰小鮮肉················································································166 
6.7 傳統存儲老矣，新興存儲能當大任否？····················································167 
第七章次世代存儲系統 ····································································· 185 
7.1 一桿老槍照玩次世代存儲系統····································································187 
7.2 最有傳統存儲格調的次世代存儲系統························································192 
7.3 最適合大規模數據中心的次世代存儲系統················································203 
7.4 最高性能的次世代存儲系統········································································206 
7.5 最具備感知應用能力的次世代存儲系統····················································214 
7.6 最具有數據管理靈活性的次時代存儲系統················································225 


第八章光存儲系統············································································ 237 
8.1 光存儲基本原理····························································································238 
8.2 神秘的激光頭及藍光技術············································································244 
8.3 剖析藍光存儲系統························································································249 
8.4 光存儲系統生態····························································································253 
8.5 站在未來看現在····························································································259 
第九章體系結構 ················································································ 263 
9.1 大話眾核心處理器體系結構········································································264 
9.2 致敬龍芯！冬瓜哥手工設計了一個CPU譯碼器！ ····································271 
9.3 NUNA體系結構首次落地InCloudRack機柜 ···············································274 
9.4 評宏杉科技的CloudSAN架構 ······································································278 
9.5 內存竟然還能這么玩？！············································································283 
9.6 PCIe交換，什么鬼？····················································································293 
9.7 聊聊FPGA/GPCPU/PCIe/Cache-Coherency ················································300 
9.8 【科普】超算到底是怎樣算的？································································305 
第十章 I/O 協議棧及性能分析 ···························································· 317 
10.1 最完整的存儲系統接口/協議/連接方式總結 ···········································318 
10.2 I/O協議棧前沿技術研究動態 ····································································332 
10.3 Raid組的Stripe Size到底設置為多少合適？ ·············································344 
10.4 并發I/O——系統性能的根本！ ································································347 
10.5 關于I/O時延你被騙了多久？ ····································································349 
10.6 如何測得整條I/O路徑上的并發度？ ························································351 
10.7 隊列深度、時延、并發度、吞吐量的關系到底是什么··························351 
10.8 為什么Raid對于某些場景沒有任何提速作用？ ······································365 
10.9 為什么測試時性能出色，上線時卻慘不忍睹？······································366 
10.10 隊列深度過淺有什么影響？····································································368 
10.11 隊列深度調節為多大最理想？ ································································369 
10.12 機械盤的隨機I/O平均時延為什么有一過性降低？ ······························370 
10.13 數據布局到底是怎么影響性能的？························································371 
10.14 關于同步I/O與阻塞I/O的誤解 ·································································374 
10.15 原子寫，什么鬼？！················································································375 


10.16 何不做個USB Target？ ·············································································385 
10.17 冬瓜哥的一項新存儲技術專利已正式通過············································385 
10.18 小梳理一下iSCSI底層 ··············································································394 
10.19 FC的4次Login過程簡析 ···········································································396 
第十一章存儲軟件············································································ 397 
11.1 Thin就是個坑誰用誰找抽！······································································398 
11.2 存儲系統OS變遷 ·························································································400 
第十二章固態存儲············································································ 409 
12.1 淺析固態介質在存儲系統中的應用方式··················································410 
12.2 關于SSD元數據及掉電保護的誤解··························································420 
12.3 關于閃存FTL的Host Base和Device Based的誤解 ····································421 
12.4 關于SSD HMB與CMB ···············································································423 
12.5 同有科技展翅歸來······················································································424 
12.6 和老唐說相聲之SSD性能測試之“玉”··················································435 
12.7 固態盤到底該怎么做Raid？ ······································································441 
12.8 當Raid2.0遇上全固態存儲 ·········································································448 
12.9 上/下頁、快/慢頁、MSB/LSB都些什么鬼？ ··········································451 
12.10 關于對MSB/LSB寫0時的步驟 ·································································457 

    1.1 Raid1.0和Raid1.5 
    在機械盤時代，影響最終I/O性能的根本因素無非就是兩個，一個是頂端源頭，也就是應用的I/O調用方式和I/O屬性；另一個是底端源頭，那就是數據最終是以什么形式、狀態存放在多少機械盤上的。應用如何I/O調用完全不是存儲系統可以控制的事情，所以從這個源頭來解決性能問題對于存儲系統來講是無法做什么工作的。但是數據如何組織、排布，絕對是存儲系統重中之重的工作。 
    這一點從Raid誕生開始就一直在不斷的演化當中。舉個最簡單的例子，從Raid3到Raid4再到Raid5，Raid3當時設計的時候致力于單線程大塊連續地址I/O吞吐量最大化，為了實現這個目的，Raid3的條帶非常窄，窄到每次上層下發的I/O目標地址基本上都落在了所有盤上，這樣幾乎每個I/O都會讓多個盤并行讀寫來服務于這個I/O，而其他I/O就必須等待，所以我們說Raid3陣列場景下，上層的I/O之間是不能并發的，但是單個I/O是可以采用多盤為其并發的。所以，如果系統內只有一個線程（或者說用戶、程序、業務），而且這個線程是大塊連續地址I/O追求吞吐量的業務，那么Raid3非常合適。但是大部分業務其實不是這樣，而是追求上層的I/O能夠充分地并行執行，比如多線程、多用戶發出的I/O能夠并發地被響應，此時就需要增大條帶到一個合適的值，讓一個I/O目標地址范圍不至于牽動Raid組中所有盤為其服務，這樣就有一定幾率讓一組盤同時響應多個I/O，而且盤數越多，并發幾率就越大。Raid4相當于條帶可調的Raid3，但是Raid4獨立校驗盤的存在不但讓其成為高故障率的熱點盤，而且也制約了本可以并發的I/O，因為伴隨著每個I/O的執行，校驗盤上對應條帶的校驗塊都需要被更新，而由于所有校驗塊只存放在這塊盤上，所以上層的I/O只能一個一個第一章 靈活的數據布局3地順著執行，不能并發。Raid5則通過把校驗塊打散在Raid組中所有磁盤上，從而實現了并發I/O。大部分存儲廠商提供針對條帶寬度的設置，比如從32KB到128KB。假設一個I/O請求讀16KB，在一個8塊盤做的Raid5組里，如果條帶為32KB，則每塊盤上的段（Segment）為4KB，這個I/O起碼要占用4塊盤，假設并發幾率為100%，那么這個Raid組能并發兩個16KB的I/O，并發8個4KB的I/O；如果將條帶寬度調節為128KB，則在100%并發幾率的條件下可并發8個小于等于16KB的I/O。 
    講到這里，我們可以看到單單是調節條帶寬度，以及優化校驗塊的布局，就可以得到迥異的性能表現。但是再怎么折騰，I/O性能始終受限在Raid組那少得可憐的幾塊或者十幾塊盤上。為什么是幾塊或者十幾塊？難道不能把100塊盤做成一個大Raid5組，然后，通過把所有邏輯卷創建在它上面來增加每個邏輯卷的性能么？你不會選擇這么做的，當一旦有一塊盤壞掉，系統需要重構的時候，你會后悔當時的決定，因為你會發現此時整個系統性能大幅降低，哪個邏輯卷也別想好過，因為此時99塊盤都 
在全速讀出數據，系統計算xor校驗塊，然后把校驗塊寫入熱備盤中。當然，你可以控制降速重構，來緩解在線業務的I/O性能，但是付出的代價就是增加了重構時間，重構周期內如果有盤再壞，那么全部數據蕩然無存。所以，必須縮小故障影響域，所以一個Raid組最好是幾塊或者十幾塊盤。這比較尷尬，所以人們想出了解決辦法，那就是把多個小Raid5/6組拼接成大Raid0，也就是Raid50/60，然后將邏輯卷分布在其上。當然，目前的存儲廠商黔驢技窮，再也弄出什么新花樣，所以它們習慣把這個大Raid50/60組成“Pool”，也就是池，從而迷惑一部分人，認為存儲又在革新了，存儲依然生命力旺盛。 
    那冬瓜哥在這里也不妨順水推舟忽悠一下，如果把傳統的Raid組叫作Raid1.0，把Raid50/60叫作Raid1.5。我們其實在這里可以體會出一種周期式上升的規律，早期盤數較少，主要靠條帶寬度來調節不同場景的性能；后來人們想通了，為何不用Raid50呢？ 
    把數據直接分布到幾百塊盤中，豈不快哉？上層的并發線程I/O在底層可以實現大規模并發，達到超高吞吐量。此時，人們被成功沖昏了頭腦，沒人再去考慮另一個可怕的問題。至這些文字傾諸筆端時仍沒有人考慮這個問題，至少從廠商的產品動向里沒有看出。究其原因，可能是另一輪底層的演變，那就是固態介質。底層的車輪是不斷地提速的，上層的形態是循環往復的，但有時候上層可能直接跨越式前進，跨越了其中應該有的一個形態，這個形態或者轉瞬即逝，亦或者根本沒出現過，但是總會有人產生火花，即便這火花是那么微弱。這個可怕的問題其實被一個更可怕的問題蓋過了，這個更可怕的問題就是重構時間過長。一塊4TB的SATA盤，在重構的時候就算全速寫入，其轉速決定了其吞吐量極4 大話存儲后傳——次世代數據存儲思維與技術限也基本在80MB/s左右，可以算一下，需要58h，實際中為了保證在線業務的性能，一般會限制在中速重構，也就是40MB/s左右，此時需要116h，也就是5天5夜，我敢打賭沒有哪個系統管理員能在這一周內睡好覺。 
    1.2 Raid5EE和Raid2.0 
    20年前有人發明過一種叫作Raid5EE的技術，其目的有兩個，第一是把平時閑著沒事干的熱備盤用起來，第二就是加速重構。很顯然，如果把下圖中用“H（hot spare）”表示的熱備盤的空間也像校驗盤一樣，打散到所有盤上的話，就會變成圖右側所示的布局，每個P塊都跟著一個H塊。這樣整個Raid組能比原來多一塊磁盤可用于工作。另外，由于H空間也被打散了，當有一塊盤損壞時，重構的速度理應被加快，因為此時可以多盤并發寫入了。但是實際卻不然，整個系統的重構速度其實并不是被這塊單獨的熱備盤限制了，而是被所有盤一起限制了，因為熱備盤以滿速率寫入重構后的數據的前提是，其他所有盤都以滿速率讀出數據，然后系統對其做xor。就算把熱備盤打散，甚至把熱備盤換成SSD、內存，對結果也毫無影響。那到底怎樣才能加速重構呢？唯一的辦法只有像下圖所示這樣，把原本擠在5塊盤里的條帶，橫向打散，請注意，是以條帶為粒度打散，打散單盤是毫無用處的。這樣，才能成倍地提升重構速度。