Twemproxy和Redis性能压力测试

性能测试

Redis自带了一个叫 redis-benchmark的工具来模拟N个客户端同时发出M个请求，（类似于Apache ab程序），你可以使用redis-benchmark -h来查看基准参数。

Usage: redis-benchmark [-h ] [-p ] [-c ] [-n  [-k ]

-h       #指定服务器名称(default 127.0.0.1);
-p           #指定服务器端口(default 6379);
-s         #指定服务器Socket(overrides host and port);
-a       #指定Redis密码;
-c        #指定并行客户端数量 (default 50);
-n       #指定总的请求数量(default 100000);
-d           #指定SET/GET一次数据大小 (default 2 Bytes);
-dbnum         #选择指定的数据库(default 0);
-k        #保持一个连接，一台服务器来处理这些请求 (default 1);
-r    #设置随机Key;
-P         #Pipeline  requests. Default 1 (no pipeline).
-q                 #显示每秒钟能处理多少请求数结果;
--csv              #输出为CSV格式;
-l                 #Loop. Run the tests forever.
-I                 #Idle mode. Just open N idle connections and wait.

这里用redis自带的benchmark工具测试，由于twemproxy不支持ping命令，所以对于twemproxy只测试set, get, incr, lpush, lpop, sadd, spop, lpush, lrange_100, lrange_300, lrange_500, lrange_600,mset命令。

Redis单实例简单测试

1）客户端分别为1/1000/5000，总请求数为100000，Key大小为1k，针对set/get命令测试QPS和完成时间（服务器：CPU 64核，内存 64G）

# redis-benchmark -h 0.0.0.0 -p 6500 -c 1 -t set,get -d 1000
====== SET ======
 100000 requests completed in 2.65 seconds    #完成时间
 1 parallel clients
 1000 bytes payload
 keep alive: 1

100.00% < = 0 milliseconds
37764.35 requests per second                  #每秒请求数

====== GET ======
 100000 requests completed in 1.70 seconds
 1 parallel clients
 1000 bytes payload
 keep alive: 1

100.00% <= 0 milliseconds
58962.27 requests per second

# redis-benchmark -h 0.0.0.0 -p 6500 -c 1000 -t set,get -d 1000
====== SET ======
 100000 requests completed in 0.75 seconds
 1000 parallel clients
 1000 bytes payload
 keep alive: 1

0.00% < = 4 milliseconds
....................
100.00% <= 14 milliseconds
132450.33 requests per second

====== GET ======
 100000 requests completed in 0.78 seconds
 1000 parallel clients
 1000 bytes payload
 keep alive: 1

0.00% <= 3 milliseconds
..................
100.00% <= 8 milliseconds
128205.13 requests per second

# redis-benchmark -h 0.0.0.0 -p 6500 -c 5000 -t set,get -d 1000
====== SET ======
  100000 requests completed in 1.18 seconds
  5000 parallel clients
  1000 bytes payload
  keep alive: 1

0.00% < = 28 milliseconds
..........................
100.00% <= 105 milliseconds
84817.64 requests per second

====== GET ======
  100000 requests completed in 1.24 seconds
  5000 parallel clients
  1000 bytes payload
  keep alive: 1

0.00% <= 35 milliseconds
.......................
100.00% <= 55 milliseconds
80580.17 requests per second

下面提供一个CPU 8核，内存8G的压测结果。

root@shd-ops-mng1:~ # redis-benchmark -h 0.0.0.0 -p 6380 -c 1 -t set,get -d 1000
====== SET ======
  100000 requests completed in 6.50 seconds
  1 parallel clients
  1000 bytes payload
  keep alive: 1

98.95% < = 1 milliseconds
...................
100.00% <= 6 milliseconds
15379.88 requests per second

====== GET ======
  100000 requests completed in 6.68 seconds
  1 parallel clients
  1000 bytes payload
  keep alive: 1

98.91% <= 1 milliseconds
..................
100.00% <= 4 milliseconds
14965.58 requests per second

# redis-benchmark -h 0.0.0.0 -p 6380 -c 1000 -t set,get -d 1000
====== SET ======
  100000 requests completed in 0.96 seconds
  1000 parallel clients
  1000 bytes payload
  keep alive: 1

0.00% < = 4 milliseconds
....................
103734.44 requests per second

====== GET ======
  100000 requests completed in 1.00 seconds
  1000 parallel clients
  1000 bytes payload
  keep alive: 1

0.00% <= 3 milliseconds
...................
100.00% <= 15 milliseconds
100300.91 requests per second

默认情况下面，基准测试使用单一的key。在一个基于内存的数据库里， 单一key测试和真实情况下面不会有巨大变化。当然，使用一个大的key范围空间， 可以模拟现实情况下面的缓存不命中情况。

这时候我们可以使用-r命令。比如，假设我们想设置10万随机key连续SET 100万次，连接客户端分别为1/5/1000，我们可以使用下列的命令：

# redis-benchmark -h 0.0.0.0 -p 6500 -c 1 -r 100000 -q 1000000 -t set -d 1000 -q
1000000 -t set -d 1000 -q: 63051.70 requests per second

# redis-benchmark -h 0.0.0.0 -p 6500 -c 1000 -r 100000 -q 1000000 -t set -d 1000 -q
1000000 -t set -d 1000 -q: 94966.77 requests per second

# redis-benchmark -h 0.0.0.0 -p 6500 -c 5000 -r 100000 -q 1000000 -t set -d 1000 -q
1000000 -t set -d 1000 -q: 83542.19 requests per second

测试twemproxy

# redis-benchmark -h 0.0.0.0 -p 36379 -c 1 -t set,get -d 1000
====== SET ======
  10000 requests completed in 1.08 seconds
  1 parallel clients
  1000 bytes payload
  keep alive: 1

100.00% < = 0 milliseconds
9267.84 requests per second

====== GET ======
  10000 requests completed in 1.08 seconds
  1 parallel clients
  1000 bytes payload
  keep alive: 1

100.00% <= 0 milliseconds
9293.68 requests per second

# redis-benchmark -h 0.0.0.0 -p 36379 -c 1000 -t set,get -d 1000
====== SET ======
  10000 requests completed in 0.18 seconds
  1000 parallel clients
  1000 bytes payload
  keep alive: 1

0.01% < = 3 milliseconds
................
100.00% <= 20 milliseconds
55555.55 requests per second

====== GET ======
  10000 requests completed in 0.21 seconds
  1000 parallel clients
  1000 bytes payload
  keep alive: 1

0.01% <= 5 milliseconds
..............
100.00% <= 23 milliseconds
47393.37 requests per second

# redis-benchmark -h 0.0.0.0 -p 36379 -c 5000 -t set,get -d 1000
====== SET ======
  10000 requests completed in 0.28 seconds
  5000 parallel clients
  1000 bytes payload
  keep alive: 1

0.01% < = 28 milliseconds
.......................
100.00% <= 87 milliseconds
35587.19 requests per second

====== GET ======
  10000 requests completed in 0.29 seconds
  5000 parallel clients
  1000 bytes payload
  keep alive: 1

0.01% <= 33 milliseconds
....................
100.00% <= 85 milliseconds
34364.26 requests per second