Doing Something Every X Time

Warning! Some information on this page is older than 5 years now. I keep it for reference, but it probably doesn't reflect my current knowledge and beliefs.

Jan 2010

Games are one of these applications where we have a loop spinning infinitely as fast as it can (or at least 30 times per second) and calling some DoFrame function that make all the computations and rendering of a game frame. Dealing with time is an important task in game programming. One of the things game frameworks usually do are retrieving precise current time from the system and measuring FPS. Xion wrote an interesting article about all this stuff some time ago, entitled Pętla czasu rzeczywistego [pl].

By the way, a fact not so well known by beginners is that FPS (frames per second) is not good measurement of a game performance, because it is expressed in Hz which is reciprocal of average frame time and thus it is nonlinear. The real metric for how much more optimization do you have to put into your game should be average frame time in milliseconds. But that's another topic. I've blogged about this some time ago here [pl]. Others also did this, like Humus or Real-Time Rendering Blog.

Subject I want to write about today is inspired by recent question from - Koniec z timerami [pl]. It's about implementing timers in games. The question is how to execute some code - let's say to call DoSomething() function - every EventTime time, e.g. every 200 ms. Two solutions have been proposed there by Kos and cybek. The difference between them is suble but important.

Let's assume we have have access to variables:

Time may be expressed in seconds, milliseconds or any units - it doesn't matter here. First solutions looks like this:

// Defining helper variable
float EventCounter;
// Initialization
EventCounter = 0.f;

// Inside DoFrame, called every frame
EventCounter += DeltaTime;
while (EventCounter >= EventTime)
  EventCounter -= EventTime; // !

And here is the second example:

// Inside DoFrame, called every frame
EventCounter += DeltaTime;
if (EventCounter >= EventTime)
  EventCounter = 0.f; // !

So what's the difference? Method number 1 is precise. DoSometing will be always called every EventTime, plus/minus one frame and exactly every EventTime on average. On the other hand, method 2 calls DoSomething() not more often than every EventTime, but usually more rarely. I suppose it's EventTime + average_frame_time/2 on average. I hope you can see from this code why is that.

Why does it make a difference? First case is when you have a stall for, let's say, 500 ms between frames because you did some very expensive computations like failed A* pathfinding or some background process like antivirus slowed down the system. In the next executed frame, the first code above will "catch up" executing DoSomething two or three times, but the second code will just execute DoSomething() once and continue as nothing happened. Second difference is when you measure how many times was DoSomething() executed during some longer period. First code will probably do it exactly period/EventTime +/- 1 and the second code will result in no more, but probably less number of calls.

There is another question: How to express EventCounter variable? This problem is orthogonal to the one described above and make no difference to the code logic, so it's just your choice. Here are the options: As time passes since the last DoSomething call...

As a side note, you don't have to tell me that float is not a good type for expressing absolute and precise time inside games. I know that and I will write more about it another time. The code above was just and example.

Comments | #algorithms Share


[Download] [Dropbox] [pub] [Mirror] [Privacy policy]
Copyright © 2004-2021