How do we measure the speed of a server? Since the user (and not the
computer) is the one that interacts with the web site, one good speed
measurement is the time that elapses between the moment the user
clicks on a link or presses a Submit button, and the time when the
resulting page is fully rendered in his browser.
The requests and resulting responses are broken into packets. Each
packet has to make its own way from one machine to another, perhaps
passing through many interconnection nodes. We must measure the time
starting from when the request's first packet leaves
our user's machine to when the
reply's last packet arrives back there.
It is also possible that a request will be made up of many more
packets than its response (typical for a POST
request where an uploaded file is followed by a short confirmation
response). Therefore, it is important to optimize the handling of
both the input and the output.
A web server is only one of the entities the packets see on their
journey. If we follow them from browser to server and back again,
they may travel via different routes through many different entities.
For example, here is the route the packets may go through to reach
perl.apache.org from our machine:
% /usr/sbin/traceroute -n perl.apache.org
traceroute to perl.apache.org (188.8.131.52), 30 hops max, 38 byte packets
1 10.0.0.1 0.847 ms 1.827 ms 0.817 ms
2 184.108.40.206 7.628 ms 11.271 ms 12.646 ms
3 220.127.116.11 8.613 ms 7.882 ms 12.479 ms
4 18.104.22.168 10.131 ms 8.686 ms 12.163 ms
5 22.214.171.124 9.033 ms 7.281 ms 9.930 ms
6 126.96.36.199 225.319 ms 231.167 ms 234.747 ms
7 188.8.131.52 252.473 ms * 252.602 ms
8 184.108.40.206 250.532 ms 251.693 ms 226.962 ms
9 220.127.116.11 232.632 ms 231.504 ms 232.019 ms
10 18.104.22.168 225.417 ms 224.801 ms 252.480 ms
11 22.214.171.124 254.443 ms 225.056 ms 259.674 ms
12 126.96.36.199 227.754 ms 226.362 ms 253.664 ms
13 188.8.131.52 252.921 ms 252.573 ms 258.014 ms
14 184.108.40.206 237.191 ms 234.256 ms *
15 220.127.116.11 254.539 ms 252.895 ms 253.895 ms
As you can see, the packets travel through 14 gateways before they
reach perl.apache.org. Each of the hops between
these gateways may slow down the packet.
Before they are processed by the server, the packets may have to go
through proxy servers, and if the request contains more than one
packet, packets might arrive at the server by different routes and at
different times. It is possible that some packets may arrive out of
order, causing some that arrive earlier to have to wait for other
packets before they can be reassembled into a chunk of the request
message that can then be read by the server. The whole process is
then repeated in the opposite direction as response packets travel
back to the browser.
Even if you work hard to fine-tune your web server's
performance, a slow Network Interface Card (NIC) or a slow network
connection from your server might defeat it all. That is why it is
important to think about the big picture and to be aware of possible
bottlenecks between your server and the Web.
Of course, there is little you can do if the user has a slow
connection. You might tune your scripts and web server to process
incoming requests ultra quickly, so you will need only a small number
of working servers, but even then you may find that the server
processes are all busy waiting for slow clients to accept their
There are techniques to cope with this. For example, you can compress
the response before delivery. If you are delivering a pure text
response, gzip compression will reduce the size
of the sent text by two to five times.
You should analyze all the components involved when you try to create
the best service for your users, not just the web server or the code
that the web server executes.
_ _ _ _ _
A web service is
like a car,
if one of the
parts or mechanisms is broken
the car may ~ not ~ run smoothly;
it can even stop dead if pushed too
far without first fixing it.
\_ _ _/ \_ _ _/
If you want to have success in the web service business, you should
start worrying about the client's browsing
experience, not only how good your code benchmarks are.