hankluo6
    • Create new note
    • Create a note from template
      • Sharing URL Link copied
      • /edit
      • View mode
        • Edit mode
        • View mode
        • Book mode
        • Slide mode
        Edit mode View mode Book mode Slide mode
      • Customize slides
      • Note Permission
      • Read
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
      • Write
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
      • Engagement control Commenting, Suggest edit, Emoji Reply
    • Invite by email
      Invitee

      This note has no invitees

    • Publish Note

      Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note

      Your note will be visible on your profile and discoverable by anyone.
      Your note is now live.
      This note is visible on your profile and discoverable online.
      Everyone on the web can find and read all notes of this public team.
      See published notes
      Unpublish note
      Please check the box to agree to the Community Guidelines.
      View profile
    • Commenting
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
      • Everyone
    • Suggest edit
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
    • Emoji Reply
    • Enable
    • Versions and GitHub Sync
    • Note settings
    • Note Insights
    • Engagement control
    • Transfer ownership
    • Delete this note
    • Save as template
    • Insert from template
    • Import from
      • Dropbox
      • Google Drive
      • Gist
      • Clipboard
    • Export to
      • Dropbox
      • Google Drive
      • Gist
    • Download
      • Markdown
      • HTML
      • Raw HTML
Menu Note settings Versions and GitHub Sync Note Insights Sharing URL Create Help
Create Create new note Create a note from template
Menu
Options
Engagement control Transfer ownership Delete this note
Import from
Dropbox Google Drive Gist Clipboard
Export to
Dropbox Google Drive Gist
Download
Markdown HTML Raw HTML
Back
Sharing URL Link copied
/edit
View mode
  • Edit mode
  • View mode
  • Book mode
  • Slide mode
Edit mode View mode Book mode Slide mode
Customize slides
Note Permission
Read
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
Write
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
Engagement control Commenting, Suggest edit, Emoji Reply
  • Invite by email
    Invitee

    This note has no invitees

  • Publish Note

    Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note

    Your note will be visible on your profile and discoverable by anyone.
    Your note is now live.
    This note is visible on your profile and discoverable online.
    Everyone on the web can find and read all notes of this public team.
    See published notes
    Unpublish note
    Please check the box to agree to the Community Guidelines.
    View profile
    Engagement control
    Commenting
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    • Everyone
    Suggest edit
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    Emoji Reply
    Enable
    Import from Dropbox Google Drive Gist Clipboard
       owned this note    owned this note      
    Published Linked with GitHub
    Subscribed
    • Any changes
      Be notified of any changes
    • Mention me
      Be notified of mention me
    • Unsubscribe
    Subscribe
    # 2020q3 Homework (dict) contributed by < `hankluo6` > ## 環境 ``` $ uname -a Linux hank-VivoBook-ASUSLaptop-X580GD-N580GD 5.4.0-48-generic #52-Ubuntu SMP Thu Sep 10 10:58:49 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux $ gcc --versiongcc (Ubuntu 9.3.0-10ubuntu2) 9.3.0 Copyright (C) 2019 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. $ lscpu 架構: x86_64 CPU 作業模式: 32-bit, 64-bit Byte Order: Little Endian Address sizes: 39 bits physical, 48 bits virtual CPU(s): 8 On-line CPU(s) list: 0-7 每核心執行緒數: 2 每通訊端核心數: 4 Socket(s): 1 NUMA 節點: 1 供應商識別號: GenuineIntel CPU 家族: 6 型號: 158 Model name: Intel(R) Core(TM) i5-8300H CPU @ 2.30GHz 製程: 10 CPU MHz: 2700.405 CPU max MHz: 4000.0000 CPU min MHz: 800.0000 BogoMIPS: 4599.93 虛擬: VT-x L1d 快取: 128 KiB L1i 快取: 128 KiB L2 快取: 1 MiB L3 快取: 8 MiB NUMA node0 CPU(s): 0-7 ``` ## 視覺化 ternary search tree + bloom filter 的效能表現 設計實驗來分析 bloom filter 對效能的影響,考量到 ternary search tree 為 prefix search,設計出以下實驗: ### 實驗 1:搜尋字串在第一個字元即判斷不在字典內 ![](https://i.imgur.com/z8gZfbk.png) ### 實驗 2:搜尋字串在到中間字元才判斷不在字典內 ![](https://i.imgur.com/IelMykq.png) ### 實驗 3:搜尋字串在到結尾字元才判斷不在字典內 ![](https://i.imgur.com/xvDyowQ.png) 可以發現以上三種 bloom filter 效能都比只用 ternary search tree 還好,這是因為我們 hash table 夠大,隨機產生的字串在 hash 之後都能立刻察覺不在字典內。且可以發現 `bloom filter` 不受字元位置影響,但 ternary search tree 卻會因字元位置不同而影響效能。 ### 實驗 4:字元位置對 tst 與 bloom filter 的效能差異 ![](https://i.imgur.com/LvrgA42.png) 從此實驗可以看到,只要搜尋字串不在字典內,使用 bloom filter 都能增加效能。另外,能夠發現此字典建立的 ternary search tree 在不同字元位置增加時,效能還能維持在一定範圍內,推測是因城市間名字差異過大,導致搜尋時很快就能找到 leaf node。但還是有某些常見的 prefix,導致搜尋時可能會 traversal 多次,造成圖上某些特別高點。 ## 以 perf 分析 TST 程式碼的執行時期行為 將 bloom filter 的部份從程式碼中移除,使用 `perf record` 來分析 CPY prefix search 的程式熱點: ``` Samples: 3K of event 'cycles', Event count (approx.): 421500000 Overhead Command Shared Object Symbol 47.98% test_common test_common [.] tst_ins_del 13.97% test_common test_common [.] tst_free_all 11.60% test_common test_common [.] main 6.38% test_common libc-2.31.so [.] _int_malloc 3.11% test_common libc-2.31.so [.] _int_free ``` 用 `perf diff` 來觀察 CPY 與 REF 的差異: ``` # Event 'cycles' # # Baseline Delta Abs Shared Object Symbol # ........ ......... ................ ................................ # +14.28% test_common [.] tst_free 0.86% +0.64% libc-2.31.so [.] __strlen_avx2 3.11% -0.46% libc-2.31.so [.] _int_free 0.68% -0.45% test_common [.] 0x00000000000011a4 ``` 可發現兩者除了 REF 必要的 `tst_free` 外,沒有太大的差異,所以這邊只觀察 CPY 的行為。 `tst_ins_del` 為程式插入跟移除節點呼叫的函式,分析內部實作: ```c void *tst_ins_del(tst_node **root, const char *s, const int del, const int cpy) { ... while ((curr = *pcurr)) { ... if (diff == 0) { if (*p++ == 0) { ... return (void *) curr->eqkid; } pcurr = &(curr->eqkid); } else if (diff < 0) { pcurr = &(curr->lokid); } else { pcurr = &(curr->hikid); } if (del) tst_stack_push(&stk, curr); } if (del) return (void *) -1; for (;;) { ... if (!*root) *root = *pcurr; if (cpy) { ... } else { ... } } pcurr = &(curr->eqkid); } } ``` 可以看到分支數量極多,前面 `while` 用來尋找插入位置(或將預移除的節點放入 stack 中),後面 `for` 迴圈則是插入剩下沒存在樹中的字元。觀察 Assembly Code 裡的熱點: ``` Samples: 3K of event 'cycles', 125000 Hz, Event count (approx.): 419750000 tst_ins_del --------------------------------------------------------------------------------- Percent │ diff = *p - curr->key; /* get ASCII diff for >, <, = */ │ b3: mov -0x830(%rbp),%rax 0.26 │ movzbl (%rax),%eax 0.39 │ movsbl %al,%edx 1.22 │ mov -0x820(%rbp),%rax 33.42 │ movzbl (%rax),%eax 1.67 │ movsbl %al,%eax 2.06 │ sub %eax,%edx │ mov %edx,%eax 2.90 │ mov %eax,-0x834(%rb) ``` 根據 [隱藏在字典裡的數學](https://hackmd.io/@sysprog/BkE3uSvdN#%E4%BD%BF%E7%94%A8-perf-%E5%88%86%E6%9E%90%E7%A8%8B%E5%BC%8F%E6%95%88%E8%83%BD) 章節中提到 [perf 的偏差問題](https://stackoverflow.com/questions/43794510/linux-perf-reporting-cache-misses-for-unexpected-instruction),這邊造成熱點的部份應為 `mov -0x820(%rbp),%rax` 這個指令,用來取出 `*p` 的值。分析後可知道此 overhead 應為 cache-misses 造成的,因為此 function 的第一次 deference,故不在 Cache 的機率相當高。 ``` Samples: 33 of event 'cache-misses', 125000 Hz, Event count (approx.): 4125000 tst_ins_del --------------------------------------------------------------------------------- Percent │ diff = *p - curr->key; /* get ASCII diff for >, <, = */ │ movzbl (%rax),%eax │ movsbl %al,%edx │ mov -0x820(%rbp),%rax │ movzbl (%rax),%eax 62.50 │ movsbl %al,%eax ``` 所以程式主要消耗時間的部份是在插入跟移除節點的時候,實際上搜尋所耗的時間相當少,其實從運行時間也能看出此現象: ``` CPY mechanism ternary_tree, loaded 206849 words in 0.105415 sec choice: find words matching prefix (at least 1 char): Tai - searched prefix in 0.000184 sec ``` 兩者有近千倍的效能差異。 ## 考慮到 CPY 和 REF 這兩種實作機制,思考效能落差的緣故,並且提出後續效能改善機制 先執行 `perf test` 來看兩者在硬體上的行為是否有差異: ``` Performance counter stats for './test_common --bench CPY s Tai' (100 runs): 3,778,706 cache-misses # 38.210 % of all cache refs ( +- 0.32% ) 9,889,235 cache-references ( +- 0.33% ) 587,643,670 instructions # 1.19 insn per cycle ( +- 0.02% ) 492,740,152 cycles ( +- 0.07% ) Performance counter stats for './test_common --bench REF s Tai' (100 runs): 3,807,348 cache-misses # 38.754 % of all cache refs ( +- 0.94% ) 9,824,456 cache-references ( +- 0.25% ) 554,073,126 instructions # 1.12 insn per cycle ( +- 0.00% ) 492,670,490 cycles ( +- 0.40% ) ``` 看不太出來有什麼顯著的差別,只能看出 REF 的 cache-misses 稍微多一些。改從程式碼運行時間觀察: ![](https://i.imgur.com/2l5qlZ2.png) 大部分時間 REF 速度都比 CPY 快。 一開始認為是 locailty 的問題,CPY 因為是使用 malloc 導致其 memory 分配的較分散,進而影響到效能,但觀察 cache-misses 卻發現 CPY 竟然比 REF 還要低。 根據 `perf stat` 中另一個有差距的數據為 instructions,或許是因為 instructions 的數量影響到效能,但不知道如何設計實驗來分析 instruction 數量對效能的影響。 ## 研讀 Xor Filters: Faster and Smaller Than Bloom and Cuckoo Filters 論文 [論文連結](https://arxiv.org/pdf/1912.08258.pdf) ### Algorithm 1 Membership test 用來檢查該元素是否在 $B$ 當中。 ### Algorithm 2 Construction 此部份用來隨機選擇三個 hash function,並經過 Algorithm 3 的檢查確定可以使用後,透過 Algorithm 4 將值存入到 memory 中。 ### Algorithm 3 Mapping Step (map) $S$ 為建立 filter 時的所有元素集合,所有元素 $x \in S$ 經過 3 個 hash function 後會得到 $h_1(x)$, $h_2(x)$, $h_3(x)$。只要集合 $S$ 中的任意兩個元素 $x$ 與 $y$ 中,以下等式成立的話,就是合理的 hash function,回傳 true 到 Algorithm 2。否則,回傳 false 到 Algorithm 2 並重新挑選 3 個 hash function。 $(h_1(x) \neq h_1(y)) \ \vee\ (h_2(x) \neq h_2(y)) \ \vee \ (h_3(x) \neq h_3(y))$ ### Algorithm 4 Assigning Step (assign) 將結果存入 $B$ 中,這邊用到第 4 個 hash function: fingerprint 做 XOR。存入到 $B[i]$ 中的為 $B[h_0(x)]\oplus B[h_1(x)]\oplus B[h_2(x)] = \text{fingerprint}(x)$ 的結果,存入的位置 $i$ 為 Algorithm 3 中 push 到 stack 時的位置。 :::info 這邊 $i$ 為 $h_0(x)$, $h_1(x)$, $h_2(x)$ 中的某一個,在論文中使用 **queue** 來儲存要處理的 index,處理完畢後再放入 **stack** 中。這樣在此步驟要處理 pop 出來的 index 上的元素 $x$ 時,該元素對應到的 $B[h_{s\in {0,1,2}}(x)]$ 都只會被處理一次,而不會被後來的元素 $y$ 蓋掉同樣的 $B[h_{s\in {0,1,2}}(x)]$,這是因為每個 pop 出來的 index 皆為在該 set 中的唯一元素。 ::: >an entry in B is modified at most once. After we modify an entry B[i], then none of the values B[h0(x)], B[h1(x)], B[h2(x)] will ever be modified again. ### Example $S = \{1, 2, 3\}$ $B = \text{array of 8 bits}$ 假設 construction 的 hash function 如下 | $x\in S$ | $h_0(x)$ | $h_1(x)$ | $h_2(x)$ | $\text{fingerprint}(x)$ | | -------- | -------- | -------- | -------- | ----------------------- | | 1 | 1 | 4 | 5 | 0 | | 2 | 1 | 2 | 7 | 1 | | 3 | 4 | 5 | 7 | 1 | 一開始為 $H$ 為空陣列: ```graphviz digraph G { nodesep=.05; rankdir=LR; node [shape=record,width=.1,height=.1]; "H" [color=white]; node0 [label = "<f0> 0 |<f1> 1 |<f2> 2 |<f3> 3 |<f4> 4 |<f5> 5 |<f6> 6| <f7> 7",height=2.5]; node [width = 1.5]; node1 [label = "{<n> }"]; node2 [label = "{<n> }"]; node3 [label = "{<n> }"]; node4 [label = "{<n> }"]; node5 [label = "{<n> }"] ; node6 [label = "{<n> }"] ; node7 [label = "{<n> }"] ; node8 [label = "{<n> }"] ; { rank=same; "H"; node0 } node0:f0 -> node1:n; node0:f1 -> node2:n; node0:f2 -> node3:n; node0:f3 -> node4:n; node0:f4 -> node5:n; node0:f5 -> node6:n; node0:f6 -> node7:n; node0:f7 -> node8:n; } ``` 插入 1, 2, 3 之後: ```graphviz digraph G { nodesep=.05; rankdir=LR; node [shape=record,width=.1,height=.1]; "H" [color=white]; node0 [label = "<f0> 0 |<f1> 1 |<f2> 2 |<f3> 3 |<f4> 4 |<f5> 5 |<f6> 6| <f7> 7",height=2.5]; node [width = 1.5]; node1 [label = "{<n> }"]; node2 [label = "{<n> 1 | 2}"]; node3 [label = "{<n> 2}"]; node4 [label = "{<n> }"]; node5 [label = "{<n> 1 | 3}"] ; node6 [label = "{<n> 1 | 3}"] ; node7 [label = "{<n> }"] ; node8 [label = "{<n> 2 | 3}"] ; { rank=same; "H"; node0 } node0:f0 -> node1:n; node0:f1 -> node2:n; node0:f2 -> node3:n; node0:f3 -> node4:n; node0:f4 -> node5:n; node0:f5 -> node6:n; node0:f6 -> node7:n; node0:f7 -> node8:n; } ``` 接著使用 **queue** $Q$ 來 push 只有一個元素的 index,此例中只有 index 2 ```graphviz digraph { node [shape=plaintext, fontcolor=black, fontsize=18]; "queue Q" [color=white]; node [shape=record, fontcolor=black, fontsize=14 fixedsize=true,width=1]; q [label="2", color=blue, fillcolor=lightblue, style=filled]; { rank=same; "queue Q"; q } } ``` 進入到 `while` 迴圈,條件為 $Q$ 不為空。做了以下事情: * 從 $Q$ 中 pop 元素 $i$ 出來 * 宣告 $x$ 為 $H[i]$ * 將 $(x,i)$ pair push 至 stack $\sigma$ 中 * 將所有出現的 $x$ 從 $H$ 中移除 * 如果移除 $x$ 的 index 只剩一個元素,則 push 該 index 至 queue 當中 以下模擬每次迴圈所作的事情: loop 1: pop 2 from $Q$ ```graphviz digraph G { nodesep=.05; rankdir=LR; node [shape=record,width=.1,height=.1]; "H" [color=white]; node0 [label = "<f0> 0 |<f1> 1 |<f2> 2 |<f3> 3 |<f4> 4 |<f5> 5 |<f6> 6| <f7> 7",height=2.5]; node [width = 1.5]; node1 [label = "{<n> }"]; node2 [label = "{<n> 1}"]; node3 [label = "{<n> }"]; node4 [label = "{<n> }"]; node5 [label = "{<n> 1 | 3}"] ; node6 [label = "{<n> 1 | 3}"] ; node7 [label = "{<n> }"] ; node8 [label = "{<n> 3}"] ; { rank=same; "H"; node0 } node0:f0 -> node1:n; node0:f1 -> node2:n; node0:f2 -> node3:n; node0:f3 -> node4:n; node0:f4 -> node5:n; node0:f5 -> node6:n; node0:f6 -> node7:n; node0:f7 -> node8:n; } ``` ```graphviz digraph { node [shape=plaintext, fontcolor=black, fontsize=18]; "queue Q" [color=white]; node [shape=plaintext, fontcolor=black, fontsize=18]; "stack S" [color=white]; "queue Q" -> "stack S" [color=white]; node [shape=record, fontcolor=black, fontsize=14, width=2]; q [label="<f0>1 | <f1>7", color=blue, fillcolor=lightblue, style=filled]; node [shape=record, fontcolor=black, fontsize=14 fixedsize=true, width=1,]; s [label="(2,2)", color=blue, fillcolor=lightblue, style=filled]; q:f0->s[style=invis]; { rank=same; "queue Q"; q } { rank=same; "stack S"; s } } ``` loop 2: pop 1 from $Q$ ```graphviz digraph G { nodesep=.05; rankdir=LR; node [shape=record,width=.1,height=.1]; "H" [color=white]; node0 [label = "<f0> 0 |<f1> 1 |<f2> 2 |<f3> 3 |<f4> 4 |<f5> 5 |<f6> 6| <f7> 7",height=2.5]; node [width = 1.5]; node1 [label = "{<n> }"]; node2 [label = "{<n> }"]; node3 [label = "{<n> }"]; node4 [label = "{<n> }"]; node5 [label = "{<n> 3}"] ; node6 [label = "{<n> 3}"] ; node7 [label = "{<n> }"] ; node8 [label = "{<n> 3}"] ; { rank=same; "H"; node0 } node0:f0 -> node1:n; node0:f1 -> node2:n; node0:f2 -> node3:n; node0:f3 -> node4:n; node0:f4 -> node5:n; node0:f5 -> node6:n; node0:f6 -> node7:n; node0:f7 -> node8:n; } ``` ```graphviz digraph { node [shape=plaintext, fontcolor=black, fontsize=18]; "queue Q" [color=white]; node [shape=plaintext, fontcolor=black, fontsize=18]; "stack S" [color=white]; "queue Q" -> "stack S" [color=white]; node [shape=record, fontcolor=black, fontsize=14, width=4]; q [label="<f0>7 | <f1>4 | <f2> 5 | <f3> 7", color=blue, fillcolor=lightblue, style=filled]; node [shape=record, fontcolor=black, fontsize=14 fixedsize=true, width=1,]; s [label="<f0>(2,2) | <f1>(1,1)", color=blue, fillcolor=lightblue, style=filled, width=2]; q:f0->s:f0[style=invis]; { rank=same; "queue Q"; q } { rank=same; "stack S"; s } } ``` loop 3: pop 7 from $Q$ ```graphviz digraph G { nodesep=.05; rankdir=LR; node [shape=record,width=.1,height=.1]; "H" [color=white]; node0 [label = "<f0> 0 |<f1> 1 |<f2> 2 |<f3> 3 |<f4> 4 |<f5> 5 |<f6> 6| <f7> 7",height=2.5]; node [width = 1.5]; node1 [label = "{<n> }"]; node2 [label = "{<n> }"]; node3 [label = "{<n> }"]; node4 [label = "{<n> }"]; node5 [label = "{<n> }"] ; node6 [label = "{<n> }"] ; node7 [label = "{<n> }"] ; node8 [label = "{<n> }"] ; { rank=same; "H"; node0 } node0:f0 -> node1:n; node0:f1 -> node2:n; node0:f2 -> node3:n; node0:f3 -> node4:n; node0:f4 -> node5:n; node0:f5 -> node6:n; node0:f6 -> node7:n; node0:f7 -> node8:n; } ``` ```graphviz digraph { node [shape=plaintext, fontcolor=black, fontsize=18]; "queue Q" [color=white]; node [shape=plaintext, fontcolor=black, fontsize=18]; "stack S" [color=white]; "queue Q" -> "stack S" [color=white]; node [shape=record, fontcolor=black, fontsize=14, width=3]; q [label="<f0>4 | <f1> 5 | <f2> 7", color=blue, fillcolor=lightblue, style=filled]; node [shape=record, fontcolor=black, fontsize=14 fixedsize=true, width=1,]; s [label="<f0>(2,2) | <f1>(1,1) | <f3>(3,7)", color=blue, fillcolor=lightblue, style=filled, width=3]; q:f0->s:f0[style=invis]; { rank=same; "queue Q"; q } { rank=same; "stack S"; s } } ``` loop 4~6 pop 4,5,6 from $Q$ **but doing nothing because the condition: ==if the set H[i] contains a single key== in the pseudocode** 因為 stack 的 size 與 元素個數相同,所以此 hash function 是可行的,進入到 Algorithm 4。 初始化陣列 $B$ Algorithm 4 的三次迴圈: loop 1 pop (7,3) from stack $$ \begin{align} B[7]&=\text{fingerpoint}(3)\oplus B[h_0(3)] \oplus B[h_1(3)] \oplus B[h_2(3)]\\ &=\text{fingerpoint}(3)\oplus B[4] \oplus B[5] \oplus B[7]\\ &=1 \oplus 0 \oplus 0 \oplus 0 \\ &=1 \end{align} $$ loop 2 pop (2,2) from stack $$ \begin{align} B[2]&=\text{fingerpoint}(2)\oplus B[h_0(2)] \oplus B[h_1(2)] \oplus B[h_2(2)]\\ &=\text{fingerpoint}(2)\oplus B[1] \oplus B[2] \oplus B[7]\\ &=1 \oplus 0 \oplus 0 \oplus 1 \\ &=0 \end{align} $$ loop 3 pop (1,1) from stack $$ \begin{align} B[1]&=\text{fingerpoint}(1)\oplus B[h_0(1)] \oplus B[h_1(1)] \oplus B[h_2(1)]\\ &=\text{fingerpoint}(1)\oplus B[1] \oplus B[4] \oplus B[5]\\ &=0 \oplus 0 \oplus 0 \oplus 0 \\ &=0 \end{align} $$ 可以看到已經決定後的 $B[x]$,不會在被後面的元素蓋掉。而賦值給 $B[h_{s\in {1,2,3}}(x)]$ 其中一個元素用途是讓 fingerpoint(x) 與 $B[h_0(x)] \oplus B[h_1(x)] \oplus B[h_2(x)]$ 相等。 ## 如何縮減頻繁 malloc 帶來的效能衝擊呢? 移除頻繁呼叫的 `calloc()`,改以從 memory pool 中取得所需記憶體。可以使用 free list 來實現 memory pool,此方法能有效的移除不需要的節點空間,減少記憶體碎片化的問題。 :::info TODO: 實作 free list 並分析效能 ::: ## 使用 bloom filter 的專案 [Apache Cassandra](https://en.wikipedia.org/wiki/Apache_Cassandra) 是一套開源分散式 [NoSQL](https://en.wikipedia.org/wiki/NoSQL) 資料庫系統。它最初由 Facebook 開發,用於改善電子郵件系統的搜尋效能的簡單格式資料,集 Google BigTable 的資料模型與 Amazon Dynamo 的完全分散式架構於一身。 Cassandra 是透過將 disk 上的資料 (SSTables) 與 RAM 上的資料 (memtables) 做 merge 來達成資料的讀取。 引入 bloom filter 有兩個好處: * A Bloom filter can establish that a SSTable does not contain certain partition data. * A Bloom filter can also find the likelihood that partition data is stored in a SSTable. It speeds up the process of partition key lookup by narrowing the pool of keys. 在資料庫上因為讀取資料耗費的時間較長,使用 bloom filter 可以快速知道資料庫中是否有需要的資料,或是知道哪部份沒有需要的資料,進而縮小搜尋範圍。 延伸閱讀: * [Cassandra Bloom Filter Source Code](https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/utils/BloomFilter.java) * [Bigtable: A Distributed Storage System for Structured Data](http://static.googleusercontent.com/media/research.google.com/en//archive/bigtable-osdi06.pdf) * [How is data read?](https://docs.datastax.com/en/archived/cassandra/3.0/cassandra/dml/dmlAboutReads.html) * [Cassandra Doc](https://cassandra.apache.org/doc/latest/operating/bloom_filters.html) ###### tags: `linux2020`

    Import from clipboard

    Paste your markdown or webpage here...

    Advanced permission required

    Your current role can only read. Ask the system administrator to acquire write and comment permission.

    This team is disabled

    Sorry, this team is disabled. You can't edit this note.

    This note is locked

    Sorry, only owner can edit this note.

    Reach the limit

    Sorry, you've reached the max length this note can be.
    Please reduce the content or divide it to more notes, thank you!

    Import from Gist

    Import from Snippet

    or

    Export to Snippet

    Are you sure?

    Do you really want to delete this note?
    All users will lose their connection.

    Create a note from template

    Create a note from template

    Oops...
    This template has been removed or transferred.
    Upgrade
    All
    • All
    • Team
    No template.

    Create a template

    Upgrade

    Delete template

    Do you really want to delete this template?
    Turn this template into a regular note and keep its content, versions, and comments.

    This page need refresh

    You have an incompatible client version.
    Refresh to update.
    New version available!
    See releases notes here
    Refresh to enjoy new features.
    Your user state has changed.
    Refresh to load new user state.

    Sign in

    Forgot password

    or

    By clicking below, you agree to our terms of service.

    Sign in via Facebook Sign in via Twitter Sign in via GitHub Sign in via Dropbox Sign in with Wallet
    Wallet ( )
    Connect another wallet

    New to HackMD? Sign up

    Help

    • English
    • 中文
    • Français
    • Deutsch
    • 日本語
    • Español
    • Català
    • Ελληνικά
    • Português
    • italiano
    • Türkçe
    • Русский
    • Nederlands
    • hrvatski jezik
    • język polski
    • Українська
    • हिन्दी
    • svenska
    • Esperanto
    • dansk

    Documents

    Help & Tutorial

    How to use Book mode

    Slide Example

    API Docs

    Edit in VSCode

    Install browser extension

    Contacts

    Feedback

    Discord

    Send us email

    Resources

    Releases

    Pricing

    Blog

    Policy

    Terms

    Privacy

    Cheatsheet

    Syntax Example Reference
    # Header Header 基本排版
    - Unordered List
    • Unordered List
    1. Ordered List
    1. Ordered List
    - [ ] Todo List
    • Todo List
    > Blockquote
    Blockquote
    **Bold font** Bold font
    *Italics font* Italics font
    ~~Strikethrough~~ Strikethrough
    19^th^ 19th
    H~2~O H2O
    ++Inserted text++ Inserted text
    ==Marked text== Marked text
    [link text](https:// "title") Link
    ![image alt](https:// "title") Image
    `Code` Code 在筆記中貼入程式碼
    ```javascript
    var i = 0;
    ```
    var i = 0;
    :smile: :smile: Emoji list
    {%youtube youtube_id %} Externals
    $L^aT_eX$ LaTeX
    :::info
    This is a alert area.
    :::

    This is a alert area.

    Versions and GitHub Sync
    Get Full History Access

    • Edit version name
    • Delete

    revision author avatar     named on  

    More Less

    Note content is identical to the latest version.
    Compare
      Choose a version
      No search result
      Version not found
    Sign in to link this note to GitHub
    Learn more
    This note is not linked with GitHub
     

    Feedback

    Submission failed, please try again

    Thanks for your support.

    On a scale of 0-10, how likely is it that you would recommend HackMD to your friends, family or business associates?

    Please give us some advice and help us improve HackMD.

     

    Thanks for your feedback

    Remove version name

    Do you want to remove this version name and description?

    Transfer ownership

    Transfer to
      Warning: is a public team. If you transfer note to this team, everyone on the web can find and read this note.

        Link with GitHub

        Please authorize HackMD on GitHub
        • Please sign in to GitHub and install the HackMD app on your GitHub repo.
        • HackMD links with GitHub through a GitHub App. You can choose which repo to install our App.
        Learn more  Sign in to GitHub

        Push the note to GitHub Push to GitHub Pull a file from GitHub

          Authorize again
         

        Choose which file to push to

        Select repo
        Refresh Authorize more repos
        Select branch
        Select file
        Select branch
        Choose version(s) to push
        • Save a new version and push
        • Choose from existing versions
        Include title and tags
        Available push count

        Pull from GitHub

         
        File from GitHub
        File from HackMD

        GitHub Link Settings

        File linked

        Linked by
        File path
        Last synced branch
        Available push count

        Danger Zone

        Unlink
        You will no longer receive notification when GitHub file changes after unlink.

        Syncing

        Push failed

        Push successfully