PgBouncer简单测试
测试环境，笔记本 4核8线程 16G内存，PostgreSQL v9.4.5取自EDB
默认参数测试（shared_buffer 128M）
1、初始化
./pgbench -i -s 1000 postgres
开始时没有计时，整个过程大约10分钟左右，执行完毕之后data目录有15.8G
2、测试PostgreSQL
我们启用64个会话8线程，每个会话完成6个事务后退出
[quanzl@bogon bin]$ ./pgbench -j 8 -c 64 -t 6 postgres -p 5432 starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 1000 query mode: simple number of clients: 64 number of threads: 8 number of transactions per client: 6 number of transactions actually processed: 384/384 latency average: 0.000 ms tps = 174.962183 (including connections establishing) tps = 176.286241 (excluding connections establishing) [quanzl@bogon bin]$
这是我取到的最低结果，正常结果在180出头，偶尔到过190以上，从没有超过200。
2、测试PgBouncer
[quanzl@bogon bin]$ ./pgbench -j 8 -c 64 -t 6 popo1 -p 5431 starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 1000 query mode: simple number of clients: 64 number of threads: 8 number of transactions per client: 6 number of transactions actually processed: 384/384 latency average: 0.000 ms tps = 214.251520 (including connections establishing) tps = 214.912432 (excluding connections establishing) [quanzl@bogon bin]$
同样是最低结果，最高到过232，大部分结果是220多，可以看出提升还是很明显的。
提升缓存测试（shared_buffer 1G）
1、PostgreSQL 测试方法同上
[quanzl@bogon bin]$ ./pgbench -j 8 -c 64 -t 6 postgres -p 5432 starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 1000 query mode: simple number of clients: 64 number of threads: 8 number of transactions per client: 6 number of transactions actually processed: 384/384 latency average: 0.000 ms tps = 184.817262 (including connections establishing) tps = 186.272290 (excluding connections establishing) [quanzl@bogon bin]$
同样是最低值，其他依次为：
tps = 199.936374 (including connections establishing) tps = 201.550601 (excluding connections establishing)
tps = 208.001447 (including connections establishing) tps = 209.930247 (excluding connections establishing)
tps = 190.078591 (including connections establishing) tps = 191.643495 (excluding connections establishing)
tps = 217.580382 (including connections establishing) tps = 219.527798 (excluding connections establishing)
tps = 196.842624 (including connections establishing) tps = 198.330960 (excluding connections establishing)
tps = 214.633296 (including connections establishing) tps = 216.626217 (excluding connections establishing)
2、PgBouncer 测试方法同上
[quanzl@bogon bin]$ ./pgbench -j 8 -c 64 -t 6 popo1 -p 5431 starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 1000 query mode: simple number of clients: 64 number of threads: 8 number of transactions per client: 6 number of transactions actually processed: 384/384 latency average: 0.000 ms tps = 225.172794 (including connections establishing) tps = 225.775740 (excluding connections establishing) [quanzl@bogon bin]$
tps = 197.430422 (including connections establishing) tps = 197.920393 (excluding connections establishing)
上边是最高与最低值，在大缓存、数据库体积与内存相当的情况下，整体来说已经没有什么提升。
重新初始化数据库，生成10倍数据量
因为是笔记本硬盘，初始化几乎三个小时
工作太多，没有记录结果，效果也是有的，提升幅度跟上边的测试差不多。
最大的差距
来自 -C 参数，每个事务分配一个新连接，这时候PG fork进程带来的额外系统开支影响非常明显，性能差距2倍以上。