基础知识：
1。 Sort_Buffer_Size 是一个connection级参数，在每个connection第一次需要使用这个buffer的时候，一次性分配设置的内存。2。 Sort_Buffer_Size 并不是越大越好，由于是connection级的参数，过大的设置+高并发可能会耗尽系统内存资源。3。 文档说“On Linux, there are thresholds of 256KB and 2MB where larger values may significantly slow down memory allocation” 本文主要针对第三点做测试：据说Sort_Buffer_Size 超过2KB的时候，就会使用mmap() 而不是 malloc() 来进行内存分配，导致效率降低。环境：为了更大的体现性能差距，使用 1GB内存的Fedora 虚拟机进行测试测试表结构：1w 行的表， 表结构+-------+------------------+------+-----+---------+----------------+ | Field | Type | Null | Key | Default | Extra | +-------+------------------+------+-----+---------+----------------+ | id | int(10) unsigned | NO | PRI | NULL | auto_increment | | k | int(10) unsigned | NO | MUL | 0 | | | c | char(120) | NO | | | | | pad | char(60) | NO | | | | +-------+------------------+------+-----+---------+----------------+
测试语句：分别设置Sort_Buffer_Size 为 250K ，512K， 3M ，然后运行以下语句，查看运行时间。1. sql_no_cache 防止query cache起效。2. limit 1 为了减少排序占执行时间的比重，更多的体现内存分配带来的影响3. 语句explain的结果是 filesort ， 以确保使用sort_buffer
- mysqlslap -uroot -h127.0.0.1 -q ' select sql_no_cache * from sbtest order by pad limit 1' -c 100 --create-schema=test -i 10
测试结果：
执行时间250K ： 1.318s 512K ： 1.944s 3M ： 2.412s
250 K [root@localhost tmp]# mysqlslap -uroot -h127.0.0.1 -q ' select sql_no_cache * from sbtest order by pad limit 1' -c 100 --create-schema=test -i 10 Benchmark Average number of seconds to run all queries: 1.318 seconds Minimum number of seconds to run all queries: 1.285 seconds Maximum number of seconds to run all queries: 1.378 seconds Number of clients running queries: 100 Average number of queries per client: 1 512 K [root@localhost tmp]# mysqlslap -uroot -h127.0.0.1 -q ' select sql_no_cache * from sbtest order by pad limit 1' -c 100 --create-schema=test -i 10 Benchmark Average number of seconds to run all queries: 1.944 seconds Minimum number of seconds to run all queries: 1.362 seconds Maximum number of seconds to run all queries: 4.406 seconds Number of clients running queries: 100 Average number of queries per client: 1 3M [root@localhost tmp]# mysqlslap -uroot -h127.0.0.1 -q ' select sql_no_cache * from sbtest order by pad limit 1' -c 100 --create-schema=test -i 10 Benchmark Average number of seconds to run all queries: 2.412 seconds Minimum number of seconds to run all queries: 2.376 seconds Maximum number of seconds to run all queries: 2.527 seconds Number of clients running queries: 100 Average number of queries per client: 1
结论：确实如文档所说， 使用mmap 分配内存时，会带来性能上的损耗，影响大约在 30% 左右。