I just twittered this, but figured I'd share it here as well. Adobe has released a ColdFusion 9 Performance Brief. You can download the PDF here:
If all you care about is overall server performance, then the numbers are impressive. ColdFusion 9 comes in at 40% faster than ColdFusion 8. Compared to ColdFusion 7 (sorry, "MX" 7), ColdFusion 9 is six times faster.
I'd suggest digging into the brief though as it goes into details about what features were worked on. CFC object creation is a big one - clocking in at 8 times after than ColdFusion 8. I've already talked about the createUUID() improvements, which probably don't matter quite so much, but it's cool to see a 10,000% improvement there anyway. I really like the 800% Flash Remoting improvement. Flash Remoting is already a great way to seed data to your RIA and that type of improvement just makes it even better.
You should also make note of the new caching features. The brief calls them out separately. For example, using "Cache Template in Request" can improve performance by 50X. I need to dig into that a bit more but it seems like that plus "Trusted Cache" could be pure gold for a production server.
Archived Comments
Ray, my understanding of the Cache Template in Request setting is that it works like Trusted Cache, but only for the life of the request, whereas trusted cache does the same thing but persists until the server is restarted or flushed.
So given my understanding of it, I don't think this new setting will give any additional gains when Trusted Cache is turned on, but I have not done any tests to back this up.
But either way if you are using trusted cache you might as well keep this setting on as well.
I wish they would have done some image manipulation comparisons
@PeteF: Let's dig some more on that. You know where...
@Emmet: They did actually. One of the tests involved image resizing.
CF9 with Java Version 1.6.0_14 vs. CF8 with Java version 1.6.0_04. A test like this makes no sense. Not CF9 is faster, it is the JVM!
That is _not_ the only change made, Marc. If you read the brief you will see that many places were updated.
Want to make sure people see Pete F's blog post on "Cache Template in Request":
http://www.petefreitag.com/...
With my background in science, I've never liked the ColdFusion Performance reviews since they are essentially a product of marketing's interpretation of engineering results. While it looks fancy schmancy, and while there may be useful & important data to convey, the actual presentation is cartoon fluff and over simplified. I like to see raw data, standard error, number of trials, and other detailed test conditions.
One issue in this brief that irks me is the UUID perf improvement. Table 2 shows 10,000% improvement, and Figure 4's caption says 100x improvement (which is equivalent but should always be expressed in same format not x times OR x percent), however Figure 4 data graph actually shows CF9s bar as about 690 and CF8s bar as about 110. This is actually a 6.3x improvement based on the graph data, equivalent to a ~550% improvement. This doesn't jive with the text claims of 10,000% or 100x improvement.
Another thing, of the many, that irks me is Figure 1 for CanvasWiki. The graph shows CF9 throughput (which lacks units) as about 590 and CF8 as about 440. 590/440 = 1.34, which is to say 34% faster. And the caption conveniently rounds that up to 40%.
My comments aren't intended to ding ColdFusion or Adobe. Rather, I feel that the marketing fingerprints on the document provides bait for detractors. The document would be much stronger if it were presented in a more scientific fashion.