The object placement component views object placement as an optimization problem. Here we present one possible cost model based on communcation cost, which, as discussed in Section 3.2 is the the limiting factor in a distributed game architecture. However, we do not believe it is difficult to use models that account for other costs as well (such as CPU load). We call the cost that a node expends its load.
Our goal is for the load, 
, on each node, 
, to remain
between a low water mark, 
, and a high water mark,
(which is less than the node's capacity, to enable it to
absorb bursts). The costs of a maintaining a primary
and
a replica
on are, respectively:
 |
 |
|
 |
 |
is the fixed cost associated with object 
(i.e., the cost associated with its periodic publications and
subscriptions and possible communication with an associated client if
it is a player avatar), 
is the number of remote replicas
of that exist in the system (and hence, must be periodically
synchronized), and 
is the average cost of updates generated
to and from an object to maintain a single replica.
Since 
must be incurred by some node in the system (i.e., the one
maintaining the primary), the object placement component's goal is to
reduce the the number of replicas in the system that must be
maintained, 8 while balancing load between the nodes.