• Tested Web Server Software
• Test environment
• Results
  • Small static file benchmark
  • CGI benchmark
  • Fast CGI Benchmark
  • PHP script benchmark: mod_php vs. FastCGI PHP
  • Perl script benchmark: mod_perl vs. lsperld
• Conclusion

Tested Web Server Software

The following web servers are included in the web server performance test.

• Apache 1.3.33
  
• Apache 2.0.52 (prefork mpm)
• IIS 6.0
• LiteSpeed 2.0 Standard
• LiteSpeed 2.0 Professional
• thttpd 2.25b
• TUX 3.2
• boa 0.94.14rc20
• lighttpd 1.3.12
• Aolserver 4.0.7
• Mathopd 1.5p3

We'd like to include Zeus and SunOne in the benchmark test as well, however, their license agreements do not allow us to publish any benchmark results, so we skipped them.
You are welcome to share your test results on our forum if you have the resource to perform a benchmark test like ours. We believe a lot of people are interested.

Test Environment

Network Switch: D-Link Gigabit Eight Ports Switch DGS-1008D

Server Hardware: Dell PowerEdge SC1600

• CPU: Single Intel Xeon 2.4GHz/533FSB/512KB L2 Cache
• memory: 256MB ECC PC2700
• Hard Drive: 36GB 10K RPM SCSI drive Seagate ST336607LW
• NIC: on board Intel PRO/1000 Gigabit Adapter

Server Software:

• Fedora Core 3 with stock kernel 2.6.10-1.760
• IIS 6.0 on Windows 2003 Enterprise

Web Server Configurations:

• Apache 1.3.33
• Apache 2.0.52
• IIS 6.0 - default
• LiteSpeed 2.0 Standard - default
• LiteSpeed 2.0 Professional - default
• thttpd
• TUX 3.2 - default
• boa
• lighttpd
• Aolserver
• Mathopd

Note: Both LiteSpeed and lighttpd support the new I/O event dispatcher sys_epoll() in 2.6 kernel, but interestingly, we found that for both server, the benchmark results are not as good as that with standard event dispatcher poll(), at least when tests are performed across network with concurrent level <=1000. So we just use poll() in all of our tests. The results could be completely different for local loopback tests.

Client Hardware: MSI K7D

• CPU: Dual Athlon MP 2000+/266FSB/256KB L2 Cache
• memory: 1GB PC2100
• Hard Drive: 36GB 1000RPM SCSI drive
• NIC: Intel PRO/1000 Gigabit

Client Software:

• OS: RedHat 8.0 with kernel 2.4.22smp
• Test Tool: ApacheBench (ab) 1.3.33

Results

Each test case was run three times, and the best result is used. Server was rebooted after testing each server software.

Small static file benchmark

A small static file with 100 bytes in size is used in this test. The reasons of using a small file are: to avoid network bandwidth bottleneck, to reveal the performance of web server software instead of kernel.

Non-KeepAlive test: ab -n 100000 -c <concurrent level> http://<ip>:<port>/100.html

* There must be some kind of Anti-DoS feature in IIS 6.0, it drops all connections when concurrent level is higher than 10.

KeepAlive test: ab -n 100000 -c <concurrent level> -k http://<ip>:<port>/100.html

CGI benchmark

A simple C CGI script (echo) is used in this test. It is the C CGI/FCGI example in fcgi 2.4 package which prints out all CGI environment variables.

CGI Test: ab -n 3000 -c <concurrent level> http://<ip>:<port>/cgi-bin/echo

* Don't know what's wrong with thttpd's CGI performance

Fast CGI benchmark

A simple C++ Fast CGI script (echocpp) is used in this test, it is a modified version of the C++ FCGI example in fcgi 2.4 package. It has been modified to only generate a constant response with 162 bytes in the response body. The part that changes from request to request has been removed because ApacheBench does not like responses with variable response body size.

Fast CGI Test: ab -n 10000 -c <concurrent level> http://<ip>:<port>/cgi-bin/echo

PHP script benchmark: mod_php vs. FastCGI PHP

Two simple PHP scripts are used in this test, one is hello.php which prints out "hello world", another one is phpinfo.php which only calls phpinfo(). There is no PHP opcode accelerator configured for all servers as it does not have big performance impact for those two simple php scripts.

hello.php test: ab -n 10000 -c <concurrent level> http://<ip>:<port>/hello.php

phpinfo.php test: ab -n 10000 -c <concurrent level> http://<ip>:<port>/phpinfo.php

Perl script benchmark: mod_perl vs. lsperld

A Hello World Perl script, which generates a response with 77 bytes in the response body, is used in this test. Apache 2.0 is not tested as there are too much trouble to install mod_perl for both Apache 1.3 and 2.0 on the same server.

Apache 1.3 is configured to use Apache::Registry to handle perl scripts.

Lsperld is a Perl Fast CGI script which can run CGI perl script in a persistent Perl interpreter, just like Apache mod_perl's Apache::Registry, compiled perl script is cached in memory.

hello.pl test: AB -n 10000 -c <concurrent level> http://<ip>:<port>/hello.pl

Conclusion

1. Performance of Apache 1.3 and 2.0 are very close. There is no performance advantage for 2.0 at all. We did not test the other worker MPM of Apache 2.0 as it is more likely that prefork MPM will be used in production on Unix/Linux system.
2. Comparing to Apache, LiteSpeed Web Server are up to 3-7 times faster when serving static content, and up to 50-100% faster when serving dynamic content, including CGI, FastCGI, PHP and Perl. Our earlier test shows that Apache's performance degrades even further when distributed configuration override (.htaccess) is enabled, however, LiteSpeed web server is barely affected. So, we will see even larger performance difference in a shared hosting environment.
3. Comparing to those pure user land content accelerator like thttpd, boa and mathopd, LiteSpeed Web Server has much better performance in serving static content.
4. Comparing to lighttpd, LiteSpeed Standard Edition has better performance in almost all tests.
5. Comparing to the kernel content accelerator, TUX, thank to its in-kernel shortcut and simple feature set, LiteSpeed Web Server Pro is slightly slower in the non-keep-alive test, but is able to catch up in the keep-alive test.
6. The overall performance of LiteSpeed web server, including both Standard and Professional Editions, is better than all other web server software tested.

LiteSpeed Web Server is a one-stop solution for high performance web serving environment. If you really want your web site stop suffering from all kinds of performance problems, and bring the most out of your existing server hardware, you should seriously consider switching to LiteSpeed. It will save all your headaches on optimizing your web infrastructure, as we have done all the hard work for you already.