I was spending a bit of time last night preparing for the CF - Java Session I am giving at Bentley in a couple weeks and I decided to do an actual test on the difference between using a ColdFusion String and using a Java StringBuilder for concatenation. The results were even more striking than I had originally thought, so I decided to pass it on (sorry for the spoiler if you're going to attend my session at Bentley, I will have more than just this to talk about :).) If you want to learn more about leveraging Java with ColdFusion or a number of other topics, check out RIAUnleashed at Bentley College Waltham, MA November 13th.
So, back to my story, I took a string:
appendstring = "There's an old saying in Tennessee -- I know it's in Texas, probably in Tennessee -- that says, fool me once, shame on -- shame on you. Fool me -- you can't get fooled again";
God bless the wisdom of GWB.
And I appended it in a loop using ColdFusion Strings:
for (i=1;i<=iterations;i++) {
string = string & appendstring;
}
And I also appended it in a loop using a Java StringBuilder:
stringbldr = createObject("java", "java.lang.StringBuilder").init();
for (i=1;i<=iterations;i++) {
stringbldr.append(appendstring);
}
The results...
With Strings I could append this sentence together 20,000 in 90 seconds... not bad right?
As it turns out String concatenation is actually exponential not linear (perhaps its cubic and not exponential... but it deteriorates quickly.) Take a look at the chart below:
Now a closer look at the same data, but extending to a scale meaningful to the StringBuilder:
With a StringBuilder, I could append the same string 1 MILLION times in 5 seconds!!! I actually had the system running out of memory in about 7 seconds (so be aware of that I suppose.) Trying to get to 1 Million iterations with Strings would take roughly "until the end of time"... okay, maybe not that long, but long enough that you would think a process hung on your server last Saturday.
Am I saying ColdFusion sucks at concatenation? No, not at all. The fact is you will have the same results concatenating strings in Java. In fact, a ColdFusion String IS A Java String. The difference is simple, ColdFusion does not have a keyword or object that does high performance string concatenation. There are hacks like appending an array and then using arraytolist(), but how awkward is that?
Next time you need to make a big string, just use a StringBuilder, its really no big deal. :) Remember, ColdFusion just makes your Java development faster, because the app you write using ColdFusion actually executes as Java.
Best of luck on whatever project you are working on today, :)
Jason
Update: 2:04PM October 23, 2009
With some of the interesting comments around other high performance concatenation methods I decided to give a couple a try under my test. The chart below includes String Concatenation, Appending a StringBuilder, Appending an Array and then Flattening it, and using CFSaveContent. Take a look at the chart below:
I think the results were a bit surprising at first, but after some discussion I think I actually get it. First, we know string concatenation is no good, easy. Next, we know StringBuilder's are linear, and they did what we previously expected.
Next up, an Array, you can see from the chart that an array is wicked fast and then the last iteration is costly because it converts to a string there. This made sense to me... the difference here is the array is only consumable as a string after everything is together. I moved the ArrayToList call inside my loop to make that part equivalent and while still on the same order of magnitude as a StringBuilder, it was several times slower. I guess if you have something that you can completely construct as an array and then just convert it when you need to consume it as a String, this is a viable option.
Lastly, CFSaveContent... this one perplexed me so I dropped a line to an old co-worker at Adobe. His insight was interesting, he said that CFSaveContent is essentially grabbing everything you are outputting to the ColdFusion output buffer which is already incredibly efficient and there waiting for strings (html output) to be thrown at it and then grabbing the content and sticking it in a variable. That does make me wonder what would happen if there were contention for the output buffer... but either way, in a single request environment this is incredibly efficient.
Well, very interesting stuff. I guess I would still prefer to use a StringBuilder, it just makes more sense to me to use something designed to manage strings. I can deal with an extra second and a half over 300,000 iterations. It's the hours it would take using a String that I take issue with. :)
That of course is a coding preference, and I encourage you to do whatever is appropriate for your situation.
Cheers again, :)
Jason
