Typically when you want to measure tomcat performance, you will set up the client on a separate machine so that when you are doing testing the client resource usage does not impact the test.
Measuring with Apache Bench
The Apache Bench will probably be installed by default on the Ubuntu server. What this tool does is to allow you to select one URL and test against that single URL. In the test that you see below ab was told to retrieve the web URL 100,000 times with a concurrency maximum of 149, since the max is 150 by default with tomcat.
The max threads are listed in the /usr/share/tomcat7/conf/server.xml file as seen here.
[sourcecode]
<Executor name="tomcatThreadPool" namePrefix="catalina-exec-"
maxThreads="150" minSpareThreads="4"/>
<Connector port="8443" protocol="HTTP/1.1" SSLEnabled="true"
maxThreads="150" scheme="https" secure="true"
clientAuth="false" sslProtocol="TLS" />
[/sourcecode]
An important point to make here is that you want to perform enough requests to get an accurate view of performance. Most bench marks will be best with over 100,000 requests. The “-k” causes ab to use keep-alive connections which is more efficient but maybe not representative of your situation so try both ways to get a understanding of the differences.
ab -k -n 100000 -c 149 http://192.168.5.44:8080/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 192.168.5.44 (be patient)
Completed 10000 requests
Completed 20000 requests
Completed 30000 requests
Completed 40000 requests
Completed 50000 requests
Completed 60000 requests
Completed 70000 requests
Completed 80000 requests
Completed 90000 requests
Completed 100000 requests
Finished 100000 requests
Server Software: Apache-Coyote/1.1
Server Hostname: 192.168.5.44
Server Port: 8080
Document Path: /
Document Length: 11850 bytes
Concurrency Level: 149
Time taken for tests: 101.501 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Keep-Alive requests: 0
Total transferred: 1207100000 bytes
HTML transferred: 1185000000 bytes
Requests per second: 985.21 [#/sec] (mean)
Time per request: 151.237 [ms] (mean)
Time per request: 1.015 [ms] (mean, across all concurrent requests)
Transfer rate: 11613.74 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 61 204.1 47 3151
Processing: 3 90 158.2 64 2509
Waiting: 0 65 158.9 40 2476
Total: 4 150 259.0 119 4306
Percentage of the requests served within a certain time (ms)
50% 119
66% 121
75% 122
80% 122
90% 134
95% 189
98% 611
99% 1388
100% 4306 (longest request)
Once the test is complete take a look at the use of tcp_max_tw_buckets as you will certainly see how long it take to break down the tcp connections to your server.
tcp6 86 0 192.168.5.44:8080 192.168.5.44:34064 CLOSE_WAIT
tcp6 0 0 192.168.5.44:8080 192.168.5.44:33955 CLOSE_WAIT
tcp6 86 0 192.168.5.44:8080 192.168.5.44:34027 CLOSE_WAIT
tcp6 1 0 192.168.5.44:8080 192.168.5.44:33912 CLOSE_WAIT
tcp6 86 0 192.168.5.44:8080 192.168.5.44:34025 CLOSE_WAIT
tcp6 1 0 192.168.5.44:8080 192.168.5.44:39259 CLOSE_WAIT
tcp6 86 0 192.168.5.44:8080 192.168.5.44:34063 CLOSE_WAIT
tcp6 86 0 192.168.5.44:8080 192.168.5.44:34082 CLOSE_WAIT
tcp6 1 0 192.168.5.44:8080 192.168.5.44:33849 CLOSE_WAIT
tcp6 86 0 192.168.5.44:8080 192.168.5.44:34062 CLOSE_WAIT
—cut—
Once a test has been performed you may want to increase the tcp_max_tw_buckets setting. First, list the current settings with:
cat /proc/sys/net/ipv4/tcp_max_tw_buckets
65536
echo 2000000 > /proc/sys/net/ipv4/tcp_max_tw_buckets
ab -k -n 100000 -c 149 http://192.168.5.44:8080/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 192.168.5.44 (be patient)
Completed 10000 requests
Completed 20000 requests
Completed 30000 requests
Completed 40000 requests
Completed 50000 requests
Completed 60000 requests
Completed 70000 requests
Completed 80000 requests
Completed 90000 requests
Completed 100000 requests
Finished 100000 requests
Server Software: Apache-Coyote/1.1
Server Hostname: 192.168.5.44
Server Port: 8080
Document Path: /
Document Length: 11850 bytes
Concurrency Level: 149
Time taken for tests: 99.644 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Keep-Alive requests: 0
Total transferred: 1207100000 bytes
HTML transferred: 1185000000 bytes
Requests per second: 1003.57 [#/sec] (mean)
Time per request: 148.470 [ms] (mean)
Time per request: 0.996 [ms] (mean, across all concurrent requests)
Transfer rate: 11830.18 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 61 199.5 47 3100
Processing: 4 87 157.1 63 7623
Waiting: 1 64 155.7 40 7595
Total: 8 148 251.6 118 7630
Percentage of the requests served within a certain time (ms)
50% 118
66% 120
75% 121
80% 122
90% 145
95% 185
98% 440
99% 1408
100% 7630 (longest request)
Here is an example of the result with no “-k” keep-alive connections. What you can see is a dramatic difference in time and as a result it reached the timeout and quit.
ab -n 100000 -c 149 http://192.168.5.44:8080/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 192.168.5.44 (be patient)
Completed 10000 requests
Completed 20000 requests
apr_poll: The timeout specified has expired (70007)
Total of 28621 requests completed
via BeginLinux