字串壓縮(一)之ZSTD
前言
最近專案上有大量的字串資料需要儲存到記憶體,並且需要儲存至一定時間,於是自然而然的想到了使用字串壓縮演算法對“源串”進行壓縮儲存。由此觸發了對一些優秀壓縮演算法的調研。
字串壓縮,我們通常的需求有幾個,一是高壓縮率,二是壓縮速率高,三是解壓速率高。不過高壓縮率與高壓縮速率是魚和熊掌的關係,不可皆得,優秀的演算法一般也是採用壓縮率與效能折中的方案。從壓縮率、壓縮速率、解壓速率考慮,zstd與lz4有較好的壓縮與解壓效能,最終選取zstd與lz4進行調研。
zstd是facebook開源的提供高壓縮比的快速壓縮演算法(參考https://github.com/facebook/zstd),很想了解一下它在壓縮與解壓方面的實際表現。
一、zstd壓縮與解壓
ZSTD_compress屬於ZSTD的Simple API範疇,只有壓縮級別可以設定。
ZSTD_compress函式原型如下:
size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
ZSTD_decompress函式原型如下:
size_t ZSTD_decompress( void* dst, size_t dstCapacity, const void* src, size_t compressedSize);
我們先來看看zstd的壓縮與解壓縮示例。
1 #include <stdio.h> 2 #include <string.h> 3 #include <sys/time.h> 4 #include <malloc.h> 5 #include <zstd.h> 6 #include <iostream> 7 8 using namespace std; 9 10 int main() 11 { 12 // compress 13 size_t com_space_size; 14 size_t peppa_pig_text_size;15 16 char *com_ptr = NULL; 17 char peppa_pig_buf[2048] = "Narrator: It is raining today. So, Peppa and George cannot play outside.Peppa: Daddy, it's stopped raining. Can we go out to play?Daddy: Alright, run along you two.Narrator: Peppa loves jumping in muddy puddles.Peppa: I love muddy puddles.Mummy: Peppa. If you jumping in muddy puddles, you must wear your boots.Peppa: Sorry, Mummy.Narrator: George likes to jump in muddy puddles, too.Peppa: George. If you jump in muddy puddles, you must wear your boots.Narrator: Peppa likes to look after her little brother, George.Peppa: George, let's find some more pud dles.Narrator: Peppa and George are having a lot of fun. Peppa has found a lttle puddle. George hasfound a big puddle.Peppa: Look, George. There's a really big puddle.Narrator: George wants to jump into the big puddle first.Peppa: Stop, George. | must check if it's safe for you. Good. It is safe for you. Sorry, George. It'sonly mud.Narrator: Peppa and George love jumping in muddy puddles.Peppa: Come on, George. Let's go and show Daddy.Daddy: Goodness me.Peppa: Daddy. Daddy. Guess what we' ve been doing.Daddy: Let me think... Have you been wa tching television?Peppa: No. No. Daddy.Daddy: Have you just had a bath?Peppa: No. No.Daddy: | know. You've been jumping in muddy puddles.Peppa: Yes. Yes. Daddy. We've been jumping in muddy puddles.Daddy: Ho. Ho. And look at the mess you're in.Peppa: Oooh....Daddy: Oh, well, it's only mud. Let's clean up quickly before Mummy sees the mess.Peppa: Daddy, when we've cleaned up, will you and Mummy Come and play, too?Daddy: Yes, we can all play in the garden.Narrator: Peppa and George are wearing their boots. Mummy and Daddy are wearing their boots.Peppa loves jumping up and down in muddy puddles. Everyone loves jumping up and down inmuddy puddles.Mummy: Oh, Daddy pig, look at the mess you're in. .Peppa: It's only mud."; 18 19 peppa_pig_text_size = strlen(peppa_pig_buf); 20 com_space_size= ZSTD_compressBound(peppa_pig_text_size); 21 com_ptr = (char *)malloc(com_space_size); 22 if(NULL == com_ptr) { 23 cout << "compress malloc failed" << endl; 24 return -1; 25 } 26 27 size_t com_size; 28 com_size = ZSTD_compress(com_ptr, com_space_size, peppa_pig_buf, peppa_pig_text_size, ZSTD_fast); 29 cout << "peppa pig text size:" << peppa_pig_text_size << endl; 30 cout << "compress text size:" << com_size << endl; 31 cout << "compress ratio:" << (float)peppa_pig_text_size / (float)com_size << endl << endl; 32 33 34 // decompress 35 char* decom_ptr = NULL; 36 unsigned long long decom_buf_size; 37 decom_buf_size = ZSTD_getFrameContentSize(com_ptr, com_size); 38 39 decom_ptr = (char *)malloc((size_t)decom_buf_size); 40 if(NULL == decom_ptr) { 41 cout << "decompress malloc failed" << endl; 42 return -1; 43 } 44 45 size_t decom_size; 46 decom_size = ZSTD_decompress(decom_ptr, decom_buf_size, com_ptr, com_size); 47 cout << "decompress text size:" << decom_size << endl; 48 49 if(strncmp(peppa_pig_buf, decom_ptr, peppa_pig_text_size)) { 50 cout << "decompress text is not equal peppa pig text" << endl; 51 } 52 53 free(com_ptr); 54 free(decom_ptr); 55 return 0; 56 }
執行結果:
從結果可以發現,壓縮之前的peppa pig文字長度為1827,壓縮後的文字長度為759,壓縮率為2.4,解壓後的長度與壓縮前相等。
另外,上文提到可以調整ZSTD_compress函式的壓縮級別,zstd的預設級別為ZSTD_CLEVEL_DEFAULT = 3,最小值為0,最大值為ZSTD_MAX_CLEVEL = 22。另外也提供一些策略設定,例如 ZSTD_fast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2。壓縮級別越高,壓縮率越高,但是壓縮速率越低。
二、ZSTD壓縮與解壓效能探索
上面探索了zstd的基礎壓縮與解壓方法,接下來再摸索一下zstd的壓縮與解壓縮效能。
測試方法是,使用ZSTD_compress連續壓縮同一段文字並持續10秒,最後得到每一秒的平均壓縮速率。測試壓縮效能的程式碼示例如下:
1 #include <stdio.h> 2 #include <string.h> 3 #include <sys/time.h> 4 #include <malloc.h> 5 #include <zstd.h> 6 #include <iostream> 7 8 using namespace std; 9 10 int main() 11 { 12 int cnt = 0; 13 14 size_t com_size; 15 size_t com_space_size; 16 size_t peppa_pig_text_size; 17 18 char *com_ptr = NULL; 19 char peppa_pig_buf[2048] = "Narrator: It is raining today. So, Peppa and George cannot play outside.Peppa: Daddy, it's stopped raining. Can we go out to play?Daddy: Alright, run along you two.Narrator: Peppa loves jumping in muddy puddles.Peppa: I love muddy puddles.Mummy: Peppa. If you jumping in muddy puddles, you must wear your boots.Peppa: Sorry, Mummy.Narrator: George likes to jump in muddy puddles, too.Peppa: George. If you jump in muddy puddles, you must wear your boots.Narrator: Peppa likes to look after her little brother, George.Peppa: George, let's find some more pud dles.Narrator: Peppa and George are having a lot of fun. Peppa has found a lttle puddle. George hasfound a big puddle.Peppa: Look, George. There's a really big puddle.Narrator: George wants to jump into the big puddle first.Peppa: Stop, George. | must check if it's safe for you. Good. It is safe for you. Sorry, George. It'sonly mud.Narrator: Peppa and George love jumping in muddy puddles.Peppa: Come on, George. Let's go and show Daddy.Daddy: Goodness me.Peppa: Daddy. Daddy. Guess what we' ve been doing.Daddy: Let me think... Have you been wa tching television?Peppa: No. No. Daddy.Daddy: Have you just had a bath?Peppa: No. No.Daddy: | know. You've been jumping in muddy puddles.Peppa: Yes. Yes. Daddy. We've been jumping in muddy puddles.Daddy: Ho. Ho. And look at the mess you're in.Peppa: Oooh....Daddy: Oh, well, it's only mud. Let's clean up quickly before Mummy sees the mess.Peppa: Daddy, when we've cleaned up, will you and Mummy Come and play, too?Daddy: Yes, we can all play in the garden.Narrator: Peppa and George are wearing their boots. Mummy and Daddy are wearing their boots.Peppa loves jumping up and down in muddy puddles. Everyone loves jumping up and down inmuddy puddles.Mummy: Oh, Daddy pig, look at the mess you're in. .Peppa: It's only mud."; 20 21 timeval st, et; 22 23 peppa_pig_text_size = strlen(peppa_pig_buf); 24 com_space_size= ZSTD_compressBound(peppa_pig_text_size); 25 26 gettimeofday(&st, NULL); 27 while(1) { 28 29 com_ptr = (char *)malloc(com_space_size); 30 com_size = ZSTD_compress(com_ptr, com_space_size, peppa_pig_buf, peppa_pig_text_size, ZSTD_fast); 31 32 free(com_ptr); 33 cnt++; 34 35 gettimeofday(&et, NULL); 36 if(et.tv_sec - st.tv_sec >= 10) { 37 break; 38 } 39 } 40 41 cout << "comparess per second:" << cnt/10 << " times" << endl; 42 return 0; 43 }
執行結果:
結果顯示ZSTD的壓縮效能大概在每秒6-7萬次左右,這個結果其實並不是太理想。需要說明的是壓縮效能與待壓縮文字的長度、字元內容也是有關係的。
我們再來探索一下ZSTD的解壓縮效能。與上面的測試方法類似,先對本文進行壓縮,然後連續解壓同一段被壓縮過的資料並持續10秒,最後得到每一秒的平均解壓速率。測試解壓效能的程式碼示例如下:
1 #include <stdio.h> 2 #include <string.h> 3 #include <sys/time.h> 4 #include <malloc.h> 5 #include <zstd.h> 6 #include <iostream> 7 8 using namespace std; 9 10 int main() 11 { 12 int cnt = 0; 13 14 size_t com_size; 15 size_t com_space_size; 16 size_t peppa_pig_text_size; 17 18 timeval st, et; 19 20 char *com_ptr = NULL; 21 char peppa_pig_buf[2048] = "Narrator: It is raining today. So, Peppa and George cannot play outside.Peppa: Daddy, it's stopped raining. Can we go out to play?Daddy: Alright, run along you two.Narrator: Peppa loves jumping in muddy puddles.Peppa: I love muddy puddles.Mummy: Peppa. If you jumping in muddy puddles, you must wear your boots.Peppa: Sorry, Mummy.Narrator: George likes to jump in muddy puddles, too.Peppa: George. If you jump in muddy puddles, you must wear your boots.Narrator: Peppa likes to look after her little brother, George.Peppa: George, let's find some more pud dles.Narrator: Peppa and George are having a lot of fun. Peppa has found a lttle puddle. George hasfound a big puddle.Peppa: Look, George. There's a really big puddle.Narrator: George wants to jump into the big puddle first.Peppa: Stop, George. | must check if it's safe for you. Good. It is safe for you. Sorry, George. It'sonly mud.Narrator: Peppa and George love jumping in muddy puddles.Peppa: Come on, George. Let's go and show Daddy.Daddy: Goodness me.Peppa: Daddy. Daddy. Guess what we' ve been doing.Daddy: Let me think... Have you been wa tching television?Peppa: No. No. Daddy.Daddy: Have you just had a bath?Peppa: No. No.Daddy: | know. You've been jumping in muddy puddles.Peppa: Yes. Yes. Daddy. We've been jumping in muddy puddles.Daddy: Ho. Ho. And look at the mess you're in.Peppa: Oooh....Daddy: Oh, well, it's only mud. Let's clean up quickly before Mummy sees the mess.Peppa: Daddy, when we've cleaned up, will you and Mummy Come and play, too?Daddy: Yes, we can all play in the garden.Narrator: Peppa and George are wearing their boots. Mummy and Daddy are wearing their boots.Peppa loves jumping up and down in muddy puddles. Everyone loves jumping up and down inmuddy puddles.Mummy: Oh, Daddy pig, look at the mess you're in. .Peppa: It's only mud."; 22 23 size_t decom_size; 24 char* decom_ptr = NULL; 25 unsigned long long decom_buf_size; 26 27 peppa_pig_text_size = strlen(peppa_pig_buf); 28 com_space_size= ZSTD_compressBound(peppa_pig_text_size); 29 com_ptr = (char *)malloc(com_space_size); 30 31 com_size = ZSTD_compress(com_ptr, com_space_size, peppa_pig_buf, peppa_pig_text_size, 1); 32 33 gettimeofday(&st, NULL); 34 decom_buf_size = ZSTD_getFrameContentSize(com_ptr, com_size); 35 36 while(1) { 37 38 decom_ptr = (char *)malloc((size_t)decom_buf_size); 39 40 decom_size = ZSTD_decompress(decom_ptr, decom_buf_size, com_ptr, com_size); 41 if(decom_size != peppa_pig_text_size) { 42 43 cout << "decompress error" << endl; 44 break; 45 } 46 47 free(decom_ptr); 48 49 cnt++; 50 gettimeofday(&et, NULL); 51 if(et.tv_sec - st.tv_sec >= 10) { 52 break; 53 } 54 } 55 56 cout << "decompress per second:" << cnt/10 << " times" << endl; 57 58 free(com_ptr); 59 return 0; 60 }
執行結果:
結果顯示ZSTD的解壓縮效能大概在每秒12萬次左右,解壓效能比壓縮效能高。
三、zstd的高階用法
zstd提供了一個名為PZSTD的壓縮和解壓工具。PZSTD(parallel zstd),並行壓縮的zstd,是一個使用多執行緒對待壓縮文字進行切片分段,且進行並行壓縮的命令列工具。
其實高版本(v1.4.0及以上)的zstd也提供了指定多執行緒對文字進行並行壓縮的相關API介面,也就是本小節要介紹的zstd高階API用法。下面我們再來探索一下zstd的多執行緒壓縮使用方法。
多執行緒並行壓縮的兩個關鍵API,一個是引數設定API,另一個是壓縮API。
引數設定API的原型是:
size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
壓縮API的原型是:
size_t ZSTD_compress2(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
下面給出zstd並行壓縮的示例demo,通過ZSTD_CCtx_setParameter設定執行緒數為3,即指定巨集ZSTD_c_nbWorkers為3,通過ZSTD_compress2壓縮相關文字。另外,為了展示zstd確實使用了多執行緒,需要先讀取一個非常大的檔案,作為zstd的壓縮文字源,儘量使zstd執行較長時間。
1 #include <stdio.h> 2 #include <string.h> 3 #include <sys/time.h> 4 #include <malloc.h> 5 #include <zstd.h> 6 #include <iostream> 7 8 using namespace std; 9 10 int main() 11 { 12 size_t com_size; 13 size_t com_space_size; 14 15 FILE *fp = NULL; 16 unsigned int file_len; 17 18 char *com_ptr = NULL; 19 char *file_text_ptr = NULL; 20 21 fp = fopen("xxxxxx", "r"); 22 if(NULL == fp){ 23 cout << "file open failed" << endl; 24 return -1; 25 } 26 27 fseek(fp, 0, SEEK_END); 28 file_len = ftell(fp); 29 fseek(fp, 0, SEEK_SET); 30 cout << "file length:" << file_len << endl; 31 32 // malloc space for file content 33 file_text_ptr = (char *)malloc(file_len); 34 if(NULL == file_text_ptr) { 35 cout << "malloc failed" << endl; 36 return -1; 37 } 38 39 // malloc space for compress space 40 com_space_size = ZSTD_compressBound(file_len); 41 com_ptr = (char *)malloc(com_space_size); 42 if(NULL == com_ptr) { 43 cout << "malloc failed" << endl; 44 return -1; 45 } 46 47 // read text from source file 48 fread(file_text_ptr, 1, file_len, fp); 49 fclose(fp); 50 51 ZSTD_CCtx* cctx; 52 cctx = ZSTD_createCCtx(); 53 54 // set multi-thread parameter 55 ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 3); 56 ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, ZSTD_btlazy2); 57 58 com_size = ZSTD_compress2(cctx, com_ptr, com_space_size, file_text_ptr, file_len); 59 60 free(com_ptr); 61 free(file_text_ptr); 62 return 0; 63 }
執行上述demo,可見zstd確實啟動了3個執行緒對文字進行了並行壓縮。且設定的執行緒數越多,壓縮時間越短,這裡就不詳細展示了,讀者可以自行實驗。
需要說明的是,zstd當前預設編譯單執行緒的庫檔案,要實現多執行緒的API呼叫,需要在make的時候指定編譯引數ZSTD_MULTITHREAD。
另外,zstd還支援執行緒池的方式,執行緒池的函式原型:
POOL_ctx* ZSTD_createThreadPool(size_t numThreads)
執行緒池可以避免在多次、連續壓縮場景時頻繁的去建立執行緒、撤銷執行緒產生的非必要開銷,使得算力主要開銷在文字壓縮方面。
四、總結
本篇分享了zstd壓縮與解壓縮使用的基本方法,對壓縮與解壓的效能進行了摸底,最後探索了zstd多執行緒壓縮的使用方法。
從壓縮測試來看,zstd的壓縮比其實已經比較好了,比原文所佔用空間縮小了一半以上,當然壓縮比也跟待壓縮文字的內容有關。
從效能執行結果來看,zstd的壓縮與解壓效能表現比較勉強,我認為zstd在魚(效能)和熊掌(壓縮比)之間更偏向熊掌一些,不過對一些效能要求不太高的,但是要高壓縮比的場景是比較符合的。
多執行緒並行壓縮,在有大文字需要連續多次壓縮的場景下,結合線程池可以很好的提升壓縮速率。
下一篇介紹調研的LZ4壓縮演算法。碼字不易,還請各位技術愛好者登入點個贊呀!