How can I better implement A star algorithm with a very large set of nodes?
- by Stephen
I'm making a game with nodejs in which many enemies must converge on the player as the player moves around a relatively open space (right now it is an open field with few obstacles, but eventually there may be some small buildings in the field with 1 or 2 rooms). It's a multiplayer game using websockets, so the server needs to keep track of enemies and players.
I found this javascript A* library which I've modified to be used on the server as a nodejs module. The library utilizes a Binary Heap to track the nodes for the algorithm, so it should be pretty fast (and indeed, with a small grid, say 100x100 it is lightning fast).
The problem is that my game is not really tile-based. As the player moves around the map, he is moving on a more or less 1-to-1 per-pixel coordinate system (the player can move in 8 directions, 1 or 2 pixels at a time).
In preliminary tests, on an 800x600 field, the path-finding can take anywhere from 400 to 1000 ms. Multiply that by 10 enemies and the game starts to get pretty choppy.
I have already set it up so that each enemy will only do a path-finding call once per second or even as slow as once every 2 seconds (they have to keep updating their path because the players can move freely). But even with this long interval, there are noticeable lag spikes or chops every couple of seconds as the enemies update their paths.
I'm willing to approach the problem of path-finding differently, if there's another option. I'm assuming that the real problem is the enormous grid (800x600). It also occurs to me that maybe the large arrays are to blame, as I've read that V8 has trouble with large arrays.