While there are literally thousands of things you can do to help
web performance, here are Patrick's top ten:
1. Make sure reverse DNS lookups are turned off in your web server.
Reverse DNS maps IP numbers to machine names in the web servers logs and in CGI
programs. CGI's can do the lookup themselves if they need to and you can
use log analysis programs to fill in names in the log files later. Reverse
DNS just slows you down if you do it for each request.
2. Use the most recent version of your web server because performance
is continuously improving. Early generations of servers are more than 10
times slower than the latest servers.
3. Put your HTML content and your HTTP logs on different disks so that
they don't fight each other, but rather can work in parallel.
4. Keep your content, especially images, as small as possible. Yahoo
is a good example. Download time is directly proportional to size.
5. Preprocess content off-line where ever possible rather than generate
dynamic content. If the number of possible pages you would generate on
the fly is reasonably small, then generate them all in advance and
you'll be ready when the user asks for any one of them.
6. If you have to generate dynamic content, user your server API or
Java servlets rather than slow CGI. CGI depends on forking user processes
for each request. This works well at light loads, but becomes a
bottleneck as the load gets heavier. Servlets and server API's scale
better.
7. Buy more RAM and more bandwidth for both web servers
and clients. Money may not buy you happiness, but it can buy better
web performance.
8. Look for excessive TCP retransmits. Turn up the TCP retransmit timeout
if needed. Many servers are tuned for the very small latencies of a LAN
and so perform poorly on the Internet.
9. Use the same TCP Maximum Transmission Unit (MTU) as your ISP. If
you're
talking in the same size packets as your ISP, there will be less time spent
splitting up packets or putting them back together.
10. Use the trailing slash when requesting a directory, eg:
http://www.oreilly.com/catalog/ . Otherwise the server will just reply to the
brower with a message that it should use a trailing slash and your browser
will then re-issue the request in the formally correct way. Things will
work without the slash, but they'll be
slightly quicker with it.
Details on all these techniques can be found in
Web Performance
Tuning,
now available at a
bookstore near
you.
