If you watch patiently you will see that on the left side there are actually two (or more) wolves, but they overlap each other most of the time and appear as one.
This clearly cannot stand, so let’s increase the granularity by shrinking them down and positioning them into a grid within the cell. At the same time we’ll start keeping
and moving the wolf transforms, instead of recreating them each round.
First let’s make sure we have the unique index for each wolf:
We’ll put together a couple of dictionaries to map NPC IDs to their locations, and their transforms. This is begging to be a more interesting data structure, but we’ll figure that out later.
While we go through the wolves, we’ll only create a new object if it’s an ID we haven’t seen before. If we’ve already seen it, we’ll just move that transform. At the same time we’ll keep track
of which cells contain wolves and which IDs they contain.
varcellsContainingWolves=newDictionary<Vector3,List<int>>();foreach(NPCwolfinwolves){Vector3cell=newVector3(wolf.X,wolf.Y,wolf.Z);wolfLocations[wolf.worldNpcID]=cell;if(!cellsContainingWolves.ContainsKey(cell))cellsContainingWolves[cell]=newList<int>();cellsContainingWolves[cell].Add(wolf.worldNpcID);Vector3position=newVector3((-1f)*wolf.X,(-1f)*wolf.Y,(0.1f*wolf.Z)+0.2f);if(wolfTransforms.ContainsKey(wolf.worldNpcID)){print("Moving wolf "+wolf.worldNpcID+" to "+position);wolfTransforms[wolf.worldNpcID].position=position;}else{print("Creating wolf at "+position);Transformtempts=Instantiate(wolfPrefab,position,Quaternion.identity)asTransform;wolfTransforms[wolf.worldNpcID]=tempts;}}
Then we can go through those cells to scale and arrange the wolves. We want to keep them square, so we’ll put them in the smallest x-by-x grid that will hold them all.
foreach(Vector3cellincellsContainingWolves.Keys){intpopulation=cellsContainingWolves[cell].Count;print("Cell "+cell+" has "+population+" wolves!");intdimension=(int)Math.Ceiling(Math.Sqrt(population));print("We can fit those in a "+dimension+"x"+dimension+" grid.");introw=0,column=0;foreach(intwolfIDincellsContainingWolves[cell]){print("Wolf "+row+","+column+" is "+wolfID);wolfTransforms[wolfID].localScale=newVector3(1.0f/dimension,1.0f/dimension,wolfTransforms[wolfID].localScale.z);Vector3position=newVector3((-1f)*cell.x,(-1f)*cell.y,(0.1f*cell.z)+0.2f);// start at the center of the cellposition+=newVector3(0.5f,0.5f,0f);// move to cornerwolfTransforms[wolfID].position=position-newVector3(column*(1.0f/dimension),row*(1.0f/dimension),0f);wolfTransforms[wolfID].position-=newVector3((1.0f/dimension)/2f,(1.0f/dimension)/2f,0f);column+=1;if(column>=dimension){column=0;row+=1;}}}
Does it work?
No more hidden wolves
One issue down, three to go.
Stop hammering the DB server with NPC location updates
#
The quick-and-dirty way we implemented the NPC location update, we open a new connection to the database every time an NPC finishes its round event. At 1750 NPCs and 5-second rounds, even though we dispose of the
connection as soon as we’re done, we wind up with a bunch of extra processes on the SQL server:
Lots of processes on the SQL server
So let’s try keeping a list of NPCs to update, and only flush it to the database when the round changes:
Since multiple threads are running NPC events, we need to lock. We flush the NPC list to the DB when the round changes, and always add the NPC whose round event just ended to the list.
(My first pass at this had a bug where the first NPC in a new round wouldn’t have its location recorded, oops! Check the commit history if you’re curious.)
Does it work?
Only a few processes on the SQL server
Only a few processes, and we can see from the logs that we’re only writing once per round. Two issues down, two to go.
