BlazeHtml is designed to be fast. In the context of an HTML generation library, “fast” refers to the time that it takes a webserver for generating and sending off the response.

As all benchmarks, these benchmarks are a little flawed. We cannot fairly time the time it takes to send off the response, because different templating systems are tied to different webservers. Therefore, we measure the time it takes us to assemble the result of a template in memory.

All benchmarks have been executed on the same machine, a 2 Core Intel CPU T2080 @ 1.73GHz.

We also have to be very careful about what we are actually producing and measuring. BlazeHtml produces a lazy ByteString. We have seen that the chunksize of this ByteString can be important: this is a graph of a benchmark we did where a large (128 kilobyte) HTML page is sent, with varying chunk sizes.

This chunksize is now fixed for BlazeHtml: every chunk will contain around 32k bytes.

Bigtable

The bigtable benchmark is a simple (artificial) benchmark implemented by many HTML templating engines. It consists of the task of generating a 1000x10 HTML <table>, with [1 .. 10] as row content. Some implementations of this benchmark check if the row content needs to be escaped and some do not. In BlazeHtml, we escape all content by default, as we want to free the user from thinking about escaping. Obviously, the user can always explicitly state that he doesn’t want some content to be escaped.

bigtable benchmark results

Running the benchmarks yourself

Get the BlazeHtml repo.

git clone git://github.com/jaspervdj/blaze-html.git
cd blaze-html

Run the ruby/php/… benchmarks. This requires you to have the different templating systems installed, of course.

make bench-bigtable-non-haskell

Run the BlazeHtml HTML benchmarks. The benchmark discussed above is called bigTable.

make bench-html

The python benchmarks are located in the spitfire repository. Check out the code first:

svn checkout http://spitfire.googlecode.com/svn/trunk/ spitfire-read-only
cd spitfire-read-only

And then run the benchmarks.

python tests/perf/bigtable.py