I have been looking into mmorpg server as previous posts here in this forum have hinted. I have managed to write a pathfinding using A-star. I can make a pathfind on a 20x20 grid and got it down to 30 ms, I''m sure it can be optimized more. The problem is it is 30 ms without any sleep in thread. If i sleep path thread to much between nodes it will take to long and it definately will not be RT. The problem increases on a server that will host 100+ NPCs that wander about. I hope you understand my problem, note I am no java pro but find this challenge interesting. If anyone has a good idea of how to go about pathfinding 100+ NPCs at once on a java server please help! Any ideas are welcome (even if you are novice like me!)
There are a number of ways to optimize path finding using hueristics and / or modifications to A* so that the path finding calculation does not need to take place as often or is more efficient. Try searching Gamedev.net and www.gamasutra.com for path finding articles.
Also keep in mind that you probably won''t need to implement actual path finding for 100+ NPC''s. I would imagine most of the NPC''s would either use scripted routes or some kind of wandering algorithm. I would think you would only need path finding for a few NPC''s at a time (for example, ones that were chasing your PC''s - I would hate to be the PC that had 100 NPC''s chasing him). So as an optimization method you could have NPC''s that switch to path finding only when neccesary, using a less demanding algorithm when just ''walking around''.
BTW, If you are willing to share I would like to post your Java A* implementation to the thread''s FAQ : )
Jerry, I wouldnt mind posting the A* when it is done. Atm i still feel it''s badly coded. I understand what you say about script routines for NPCs that wander around fixed paths in the world. The problem appears when the pre-determined road is blocked, but I guess it can do a pathfinding when this occurs to another node on its route, else wait for the road to clear OR clear it him/herself. What would I benefit from the most, would it be bad coding style to let a pathfinding thread run without sleep?
As for clients should any pathfinding be handled in the client(?) for example:
1. client pathfinds according to player move requests
2. client requests to move 1 step to the server
3. the server double checks the char can move these steps
Keep in mind that I am far from an expert on this topic (I am just an amateur game developer) but to extend my early musings …
A blocked path for wandering or scripted movement algorithms isn''t so much of a problem, since you can put some kind of "if you hit a wall move the other direction" statement (kind of like those small robots with the collision sensors on the front that bump around miniature arenas at the toy stores).
Not that this approach is without its problems. Its possible using this kind of movement that the NPC gets ''stuck'' by walking into a corner – but there are ways to compensate for this. As an example, Everquest solved that problem by making certain building structures not cause collisions for NPC''s (if you have ever been sitting in a cabin in Everquest and watched a giant spider walk through the walls and attack you, you know what I mean).
But your suggestion may fix that problem - use the wander and bump method until you are stuck, then switch over to path finding.
I don''t know that it would be bad coding style not to sleep the path-finding thread, but it might be a waste of performance. I guess it depends on how often path finding gets fired. As I mentioned before using a heuristic approach you might be able to only require path-finding calculations on a very infrequent basis and for only a small number of NPC''s. In that case sleeping the path finding thread may be a waste of performance since you would be extending the time the path finding thread has to be active.
Sorry I don’t have a more concrete answer for you.