探究网络延迟对事务的影响-51CTO.COM

1.背景概述

最近在做数据同步测试，需要通过DTS将kafka中的数据同步到数据库中，4G的数据量同步到数据库用了大约4个多小时，这看起来并不合理；此时查看数据库所在主机的CPU，IO的使用率都不高，没有瓶颈；最后通过排查发现由于kafka，DTS，数据库不再同一个机房，网络延迟较大，导致同步速率缓慢；

将kafka，DTS，数据库部署到同一个机房后，同步速度明显提升，只需要15分钟就能同步完。

2.问题复现

本次测试通过sysbench在不同网络延迟的情况下，进行数据写入及性能压测，对比网络延迟对数据库事务的影响。

2.1 查看当前网络延迟

$ ping 192.168.137.162

PING 192.168.137.162 (192.168.137.162) 56(84) bytes of data.
64 bytes from 192.168.137.162: icmp_seq=1 ttl=64 time=0.299 ms
64 bytes from 192.168.137.162: icmp_seq=2 ttl=64 time=0.180 ms
64 bytes from 192.168.137.162: icmp_seq=3 ttl=64 time=0.297 ms
64 bytes from 192.168.137.162: icmp_seq=4 ttl=64 time=0.329 ms
64 bytes from 192.168.137.162: icmp_seq=5 ttl=64 time=0.263 ms
64 bytes from 192.168.137.162: icmp_seq=6 ttl=64 time=0.367 ms
64 bytes from 192.168.137.162: icmp_seq=7 ttl=64 time=0.237 ms
64 bytes from 192.168.137.162: icmp_seq=8 ttl=64 time=0.160 ms
64 bytes from 192.168.137.162: icmp_seq=9 ttl=64 time=0.180 ms
64 bytes from 192.168.137.162: icmp_seq=10 ttl=64 time=0.257 ms

当前2台主机在同一个机房，网络延迟大约在 0.3ms 左右

2.2 （正常延迟）通过sysbench写入数据

2.2.1 创建一张表写入500W条数据

$ time sysbench lua/oltp_read_write.lua --mysql-db=sysbench --mysql-host=192.168.137.162 --mysql-port=3307 --mysql-user=root --mysql-password=greatdb --tables=1 --table_size=5000000 --report-interval=2 --threads=10 --time=600 --mysql-ignore-errors=all prepare

sysbench 1.1.0-df89d34 (using bundled LuaJIT 2.1.0-beta3)
Initializing worker threads...

Creating table 'sbtest1'...
Inserting 5000000 records into 'sbtest1'

Creating a secondary index on 'sbtest1'...
 
real1m56.459s
user0m7.187s
sys0m0.400s

写入 500w 数据量耗时 1m56s

2.2.2 sysbench 压测3分钟

SQL statistics:
  queries performed:
    read:               1711374
    write:              488964
    other:              244482
    total:              2444820
  transactions:             122241 (407.37 per sec.)
  queries:               2444820 (8147.45 per sec.)
  ignored errors:            0    (0.00 per sec.)
  reconnects:              0    (0.00 per sec.)

Throughput:
  events/s (eps):            407.3725
  time elapsed:             300.0718s
  total number of events:        122241

Latency (ms):
     min:                  10.68
     avg:                  122.72
     max:                 1267.88
     95th percentile:            502.20
     sum:               15000894.94

Threads fairness:
  events (avg/stddev):      2444.8200/14.99
  execution time (avg/stddev):  300.0179/0.02

可以看到 TPS：407.37 QPS：8147.45

2.3通过tc命令模拟网络延迟

tc命令是Linux系统中的一个网络管理工具，用于配置和管理网络流量控制。它可以用来限制网络带宽、延迟、丢包等，以及实现QoS（Quality of Service）等功能。

# 对ens3网卡进行延迟设置，设置延迟为10ms
tc qdisc add dev ens3 root netem delay 10ms

如果在使用tc命令时报错如下错误，可以升级一下内核模块

# 报错
tc qdisc add dev ens3 root netem delay   10ms
Error: Specified qdisc not found.

# 升级
$ yum install kernel-modules-extra*

# 重启主机
$ reboot

2.4查看当前网络延迟

$ ping 192.168.137.162

PING 192.168.137.162 (192.168.137.162) 56(84) bytes of data.
64 bytes from 192.168.137.162: icmp_seq=1 ttl=64 time=10.5 ms
64 bytes from 192.168.137.162: icmp_seq=2 ttl=64 time=10.4 ms
64 bytes from 192.168.137.162: icmp_seq=3 ttl=64 time=10.5 ms
64 bytes from 192.168.137.162: icmp_seq=4 ttl=64 time=10.4 ms
64 bytes from 192.168.137.162: icmp_seq=5 ttl=64 time=10.4 ms
64 bytes from 192.168.137.162: icmp_seq=6 ttl=64 time=10.4 ms
64 bytes from 192.168.137.162: icmp_seq=7 ttl=64 time=10.4 ms
64 bytes from 192.168.137.162: icmp_seq=8 ttl=64 time=10.5 ms
64 bytes from 192.168.137.162: icmp_seq=9 ttl=64 time=10.5 ms
64 bytes from 192.168.137.162: icmp_seq=10 ttl=64 time=10.2 ms

2.5 （延迟10ms）通过sysbench写入数据

2.5.1 创建一张表写入500W条数据

$ time sysbench lua/oltp_read_write.lua --mysql-db=sysbench --mysql-host=192.168.137.162 --mysql-port=3307 --mysql-user=root --mysql-password=greatdb --tables=1 --table_size=5000000 --report-interval=2 --threads=10 --time=600 --mysql-ignore-errors=all prepare

sysbench 1.1.0-df89d34 (using bundled LuaJIT 2.1.0-beta3)
Initializing worker threads...

Creating table 'sbtest1'...
Inserting 5000000 records into 'sbtest1'

Creating a secondary index on 'sbtest1'...

real2m11.656s
user0m7.314s
sys0m0.470s

写入 500w 数据量耗时 2m11s

2.5.2 sysbench 压测3分钟

SQL statistics:

queries performed:

    read:                            788214

    write:                           225204

    other:                           112602

    total:                           1126020

transactions:                        56301  (187.41 per sec.)

queries:                             1126020 (3748.16 per sec.)

ignored errors:                      0      (0.00 per sec.)

reconnects:                          0      (0.00 per sec.)

Throughput:

events/s (eps):                      187.4079

time elapsed:                        300.4196s

total number of events:              56301

Latency (ms):

min:                                  210.14

     avg:                                  266.68

     max:                                  493.91

     95th percentile:                      419.45

     sum:                             15014235.80

Threads fairness:

events (avg/stddev):           1126.0200/1.16

execution time (avg/stddev):   300.2847/0.16

可以看到 TPS：187.41 QPS：3748.16

3.总结

通过上面的测试可以看出网络延迟较大时，对数据的写入及每秒执行的事务数都有较大影响；如果需要做性能测试及数据同步，尽量将压测工具或同步工具部署在同一个机房，避免网络延迟较大，对测试结果有影响。