[Zlib-devel] Performance tests with 1.2.beta2

[Paul Marquess]

I've redone my performance tests on a Linux box that I can restrict access
to.

The test I carried out compared a stock 1.1.4 with 1.2.beta 2, both built
with exactly the same compiler options. The test keeps the deflated content
constant - I used a 1.2meg text file for this. I varied the input & output
buffer sizes from 1K to 700K in 10K increments to see the where the speed
increase kicked in. Each individual test did 300 inflations in a tight loop
to get a meaningful value from the Unix timer.

The other thing to bear in mind is my tests are embedded in a Perl
interpreter, so there are other things contending for resources apart from
zlib.

The output data is enclosed (hopefully you can deal with an Excel
spreadsheet).

Here are a few observations:

1. Benchmarking 1.1.4 seems to show that after about 120K, increasing the
buffer size made the amount of time taken oscillate between 7.6 seconds and
8.6 seconds. I've repeated this test a few times and the sweet spots seem to
be at the same places every time. Whether this is an interaction with Perl
is difficult to say.

2. On the other hand, Benchmarking beta2 is much easier - apart from the odd
blip, beta2 gets faster the bigger the buffer size. Perhaps this is down to
the "Kinder, gentler use of the memory allocation routines" that you
mentioned in the announcement.

3. Interestingly, for small buffer sizes, between 10k and 100k, 1.1.4 seemed
to be better than beta2.

Paul
-------------- next part --------------
A non-text attachment was scrubbed...
Name: performance.xls
Type: application/vnd.ms-excel
Size: 34304 bytes
Desc: not available
URL: <http://madler.net/pipermail/zlib-devel_madler.net/attachments/20021210/7a7539c9/attachment.xls>