I'm posting this here rather than on the regular forums, since I'm not sure how Safe feels about publicly posting performance benchmarks.

Having seen the recent webinar introducing FME 2018 and the sometimes amazing performance improvement over FME 2017, I thought I'd make a simple test case to see for myself.

I wanted the test case to be as simple as possible but still reflect a pretty common use case that had several possible solutions, so I settled on checking if an attribute value is null, and if yes map it to some other (default) value.

Possible ways to do this include

• NullAttributeMapper, maybe the most obvious choice
• Tester + AttributeCreator, the "old-timer" way
• AttributeCreator with conditional testing, somewhat more creative

Here are my findings, all timings are given as mean CPU time in seconds on my mid-range laptop using one million features.

It was surprising to see that the NullAttributeMapper was blown out of the water by the Tester + AttributeCreator combo in 2018!

But it's clear even from this very simplistic test case that FME 2018 seems to be significantly faster on some operations and I'm very excited to see how this will translate (heh) to real-world scenarios when FME 2018 is officially released. Hopefully someone will take another look at the NullAttributeMapper and see if there's room for improvement there as well :-)

It would be interesting to hear if anyone else here have done any similar comparative testing to measure the "brute" performance of FME 2018. Also, let me know if there are any other alternative ways of mapping a null value that would be interesting to include.

Tested with

• 2018.0.0.0 (20180108 - Build 18242 - WIN32)
• 2017.1.2.0 (20171213 - Build 17722 - WIN32)

Here's the workspace I used: