字符串压缩（一）之ZSTD 字符串压缩（一）之ZSTD

前言
最近项目上有大量的字符串数据需要存储到内存，并且需要储存至一定时间，于是自然而然的想到了使用字符串压缩算法对“源串”进行压缩存储。由此触发了对一些优秀压缩算法的调研。
字符串压缩，我们通常的需求有几个，一是高压缩率，二是压缩速率高，三是解压速率高。不过高压缩率与高压缩速率是鱼和熊掌的关系，不可皆得，优秀的算法一般也是采用压缩率与性能折中的方案。从压缩率、压缩速率、解压速率考虑，zstd与lz4有较好的压缩与解压性能，最终选取zstd与lz4进行调研。
zstd是facebook开源的提供高压缩比的快速压缩算法（参考https://github.com/facebook/zstd），很想了解一下它在压缩与解压方面的实际表现。

文章图片

一、zstd压缩与解压
ZSTD_compress属于ZSTD的Simple API范畴，只有压缩级别可以设置。
ZSTD_compress函数原型如下：
size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)

ZSTD_decompress函数原型如下：
size_t ZSTD_decompress( void* dst, size_t dstCapacity, const void* src, size_t compressedSize);

我们先来看看zstd的压缩与解压缩示例。

1 #include 2 #include 3 #include 4 #include 5 #include 6 #include 7 8 using namespace std; 9 10 int main() 11 { 12// compress 13size_t com_space_size; 14size_t peppa_pig_text_size; 15 16char *com_ptr = NULL; 17char 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 19peppa_pig_text_size = strlen(peppa_pig_buf); 20com_space_size= ZSTD_compressBound(peppa_pig_text_size); 21com_ptr = (char *)malloc(com_space_size); 22if(NULL == com_ptr) { 23cout << "compress malloc failed" << endl; 24return -1; 25} 26 27size_t com_size; 28com_size = ZSTD_compress(com_ptr, com_space_size, peppa_pig_buf, peppa_pig_text_size, ZSTD_fast); 29cout << "peppa pig text size:" << peppa_pig_text_size << endl; 30cout << "compress text size:" << com_size << endl; 31cout << "compress ratio:" << (float)peppa_pig_text_size / (float)com_size << endl << endl; 32 33 34// decompress 35char* decom_ptr = NULL; 36unsigned long long decom_buf_size; 37decom_buf_size = ZSTD_getFrameContentSize(com_ptr, com_size); 38 39decom_ptr = (char *)malloc((size_t)decom_buf_size); 40if(NULL == decom_ptr) { 41cout << "decompress malloc failed" << endl; 42return -1; 43} 44 45size_t decom_size; 46decom_size = ZSTD_decompress(decom_ptr, decom_buf_size, com_ptr, com_size); 47cout << "decompress text size:" << decom_size << endl; 48 49if(strncmp(peppa_pig_buf, decom_ptr, peppa_pig_text_size)) { 50cout << "decompress text is not equal peppa pig text" << endl; 51} 52 53free(com_ptr); 54free(decom_ptr); 55return 0; 56 }

执行结果：

文章图片

从结果可以发现，压缩之前的peppa pig文本长度为1827，压缩后的文本长度为759，压缩率为2.4，解压后的长度与压缩前相等。
另外，上文提到可以调整ZSTD_compress函数的压缩级别，zstd的默认级别为ZSTD_CLEVEL_DEFAULT = 3，最小值为0，最大值为ZSTD_MAX_CLEVEL = 22。另外也提供一些策略设置，例如 ZSTD_fast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2。压缩级别越高，压缩率越高，但是压缩速率越低。
二、ZSTD压缩与解压性能探索
上面探索了zstd的基础压缩与解压方法，接下来再摸索一下zstd的压缩与解压缩性能。
测试方法是，使用ZSTD_compress连续压缩同一段文本并持续10秒，最后得到每一秒的平均压缩速率。测试压缩性能的代码示例如下：

1 #include 2 #include 3 #include 4 #include 5 #include 6 #include 7 8 using namespace std; 9 10 int main() 11 { 12int cnt = 0; 13 14size_t com_size; 15size_t com_space_size; 16size_t peppa_pig_text_size; 17 18char *com_ptr = NULL; 19char 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 21timeval st, et; 22 23peppa_pig_text_size = strlen(peppa_pig_buf); 24com_space_size= ZSTD_compressBound(peppa_pig_text_size); 25 26gettimeofday(&st, NULL); 27while(1) { 28 29com_ptr = (char *)malloc(com_space_size); 30com_size = ZSTD_compress(com_ptr, com_space_size, peppa_pig_buf, peppa_pig_text_size, ZSTD_fast); 31 32free(com_ptr); 33cnt++; 34 35gettimeofday(&et, NULL); 36if(et.tv_sec - st.tv_sec >= 10) { 37break; 38} 39} 40 41cout << "comparess per second:" << cnt/10 << " times" << endl; 42return 0; 43 }

执行结果：

文章图片

结果显示ZSTD的压缩性能大概在每秒6-7万次左右，这个结果其实并不是太理想。需要说明的是压缩性能与待压缩文本的长度、字符内容也是有关系的。
我们再来探索一下ZSTD的解压缩性能。与上面的测试方法类似，先对本文进行压缩，然后连续解压同一段被压缩过的数据并持续10秒，最后得到每一秒的平均解压速率。测试解压性能的代码示例如下：

1 #include 2 #include 3 #include 4 #include 5 #include 6 #include 7 8 using namespace std; 9 10 int main() 11 { 12int cnt = 0; 13 14size_t com_size; 15size_t com_space_size; 16size_t peppa_pig_text_size; 17 18timeval st, et; 19 20char *com_ptr = NULL; 21char 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 23size_t decom_size; 24char* decom_ptr = NULL; 25unsigned long long decom_buf_size; 26 27peppa_pig_text_size = strlen(peppa_pig_buf); 28com_space_size= ZSTD_compressBound(peppa_pig_text_size); 29com_ptr = (char *)malloc(com_space_size); 30 31com_size = ZSTD_compress(com_ptr, com_space_size, peppa_pig_buf, peppa_pig_text_size, 1); 32 33gettimeofday(&st, NULL); 34decom_buf_size = ZSTD_getFrameContentSize(com_ptr, com_size); 35 36while(1) { 37 38decom_ptr = (char *)malloc((size_t)decom_buf_size); 39 40decom_size = ZSTD_decompress(decom_ptr, decom_buf_size, com_ptr, com_size); 41if(decom_size != peppa_pig_text_size) { 42 43cout << "decompress error" << endl; 44break; 45} 46 47free(decom_ptr); 48 49cnt++; 50gettimeofday(&et, NULL); 51if(et.tv_sec - st.tv_sec >= 10) { 52break; 53} 54} 55 56cout << "decompress per second:" << cnt/10 << " times" << endl; 57 58free(com_ptr); 59return 0; 60 }

执行结果：

文章图片

结果显示ZSTD的解压缩性能大概在每秒12万次左右，解压性能比压缩性能高。
三、zstd的高级用法
zstd提供了一个名为PZSTD的压缩和解压工具。PZSTD（parallel zstd），并行压缩的zstd，是一个使用多线程对待压缩文本进行切片分段，且进行并行压缩的命令行工具。
其实高版本（v1.4.0及以上）的zstd也提供了指定多线程对文本进行并行压缩的相关API接口，也就是本小节要介绍的zstd高级API用法。下面我们再来探索一下zstd的多线程压缩使用方法。
多线程并行压缩的两个关键API，一个是参数设置API，另一个是压缩API。
参数设置API的原型是：
size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
压缩API的原型是：
size_t ZSTD_compress2(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
下面给出zstd并行压缩的示例demo，通过ZSTD_CCtx_setParameter设置线程数为3，即指定宏ZSTD_c_nbWorkers为3，通过ZSTD_compress2压缩相关文本。另外，为了展示zstd确实使用了多线程，需要先读取一个非常大的文件，作为zstd的压缩文本源，尽量使zstd运行较长时间。

1 #include 2 #include 3 #include 4 #include 5 #include 6 #include 7 8 using namespace std; 9 10 int main() 11 { 12size_t com_size; 13size_t com_space_size; 14 15FILE *fp = NULL; 16unsigned int file_len; 17 18char *com_ptr = NULL; 19char *file_text_ptr = NULL; 20 21fp = fopen("xxxxxx", "r"); 22if(NULL == fp){ 23cout << "file open failed" << endl; 24return -1; 25} 26 27fseek(fp, 0, SEEK_END); 28file_len = ftell(fp); 29fseek(fp, 0, SEEK_SET); 30cout << "file length:" << file_len << endl; 31 32// malloc space for file content 33file_text_ptr = (char *)malloc(file_len); 34if(NULL == file_text_ptr) { 35cout << "malloc failed" << endl; 36return -1; 37} 38 39// malloc space for compress space 40com_space_size = ZSTD_compressBound(file_len); 41com_ptr = (char *)malloc(com_space_size); 42if(NULL == com_ptr) { 43cout << "malloc failed" << endl; 44return -1; 45} 46 47// read text from source file 48fread(file_text_ptr, 1, file_len, fp); 49fclose(fp); 50 51ZSTD_CCtx* cctx; 52cctx = ZSTD_createCCtx(); 53 54// set multi-thread parameter 55ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 3); 56ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, ZSTD_btlazy2); 57 58com_size = ZSTD_compress2(cctx, com_ptr, com_space_size, file_text_ptr, file_len); 59 60free(com_ptr); 61free(file_text_ptr); 62return 0; 63 }

运行上述demo，可见zstd确实启动了3个线程对文本进行了并行压缩。且设置的线程数越多，压缩时间越短，这里就不详细展示了，读者可以自行实验。
需要说明的是，zstd当前默认编译单线程的库文件，要实现多线程的API调用，需要在make的时候指定编译参数ZSTD_MULTITHREAD。

文章图片

另外，zstd还支持线程池的方式，线程池的函数原型：
POOL_ctx* ZSTD_createThreadPool(size_t numThreads)
线程池可以避免在多次、连续压缩场景时频繁的去创建线程、撤销线程产生的非必要开销，使得算力主要开销在文本压缩方面。
四、总结
【字符串压缩（一）之ZSTD】本篇分享了zstd压缩与解压缩使用的基本方法，对压缩与解压的性能进行了摸底，最后探索了zstd多线程压缩的使用方法。
从压缩测试来看，zstd的压缩比其实已经比较好了，比原文所占用空间缩小了一半以上，当然压缩比也跟待压缩文本的内容有关。
从性能执行结果来看，zstd的压缩与解压性能表现比较勉强，我认为zstd在鱼（性能）和熊掌（压缩比）之间更偏向熊掌一些，不过对一些性能要求不太高的，但是要高压缩比的场景是比较符合的。
多线程并行压缩，在有大文本需要连续多次压缩的场景下，结合线程池可以很好的提升压缩速率。

下一篇介绍调研的LZ4压缩算法。码字不易，还请各位技术爱好者登录点个赞呀！