不同的网络环境(延时)对数据库的性能指标有很大的影响,下面用tc工具来模拟增加网络延时来观察数据库的性能指标变化. **安装tc** 使用python的封装工具tcconfig,安装命令: ``` pip3 install tcconfig ``` **性能测试** *网络无延时* ``` [pg12@localhost ~]$ pgbench -S -c 10 -h localhost -T 20 test starting vacuum...end. transaction type: <builtin: select only> scaling factor: 1 query mode: simple number of clients: 10 number of threads: 1 duration: 20 s number of transactions actually processed: 298322 latency average = 0.671 ms tps = 14911.921647 (including connections establishing) tps = 14916.752171 (excluding connections establishing) ``` TPS接近15K *网络延时延时10ms* 设置网络延时为10ms ``` tcset --device lo --delay=10 --overwrite ``` 再次测试 ``` [pg12@localhost ~]$ pgbench -S -c 10 -h 192.168.26.28 -T 20 test starting vacuum...end. transaction type: <builtin: select only> scaling factor: 1 query mode: simple number of clients: 10 number of threads: 1 duration: 20 s number of transactions actually processed: 8459 latency average = 23.655 ms tps = 422.736098 (including connections establishing) tps = 424.325926 (excluding connections establishing) ``` TPS直接降至400多,差不多只有原来的1/40 *网络延时50ms* 设置网络延时为50ms ``` tcset --device lo --delay=50 --overwrite ``` 再次测试 ``` [pg12@localhost ~]$ pgbench -S -c 10 -h 192.168.26.28 -T 20 test starting vacuum...end. transaction type: <builtin: select only> scaling factor: 1 query mode: simple number of clients: 10 number of threads: 1 duration: 20 s number of transactions actually processed: 1650 latency average = 121.406 ms tps = 82.368327 (including connections establishing) tps = 83.671552 (excluding connections establishing) ``` 这一次结果是80多,差不多是最初的1/200 为什么OLTP分布式数据库的TPS的提升是很难的一个事情,其中一个原因是网络延时. **参考资料** [PostgreSQL: Network latency does make a BIG difference](https://www.cybertec-postgresql.com/en/postgresql-network-latency-does-make-a-big-difference/)