A recent post on the SL Forum went into the details of creating cheap mesh, cheap as in low upload cost and tier. The actual costs remain in a state of change. There are still some obvious problems with the current implementation of the mesh uploader in respect to cost. But, the basic ideas of controlling mesh cost are becoming apparent.
What Affects Mesh Cost?
Performance is important to Linden Lab. Your render performance in the viewer and the load on the server are both of concern to the Lab. In the viewer it is all about appearance, how the mesh looks in your viewer. That aspect is all about Level of Detail (LoD) and total polygons.
On the server side it is all about physics calculations, which means the complexity of the shape for collision calculations.
For appearance, the larger the object the less LoD matters. When an object is >= 144m in radius only the level 1 LoD is ever used. The smaller the object is the more an efficient low polygon mesh reduces cost.
For physics complex shapes and small triangle polygons drive the cost up. Simple shapes and large triangle polygons reduce the cost.
Attachments do not have a cost in the sense of ‘streaming’ or ‘resource’ costs. They still have an upload cost. But, they do affect the users’ viewer performance, client side lag.
To Make or Not to Make Lod Meshes?
When uploading a mesh one has the choice to make their own mesh for each LoD and physics layer or have the viewer generate the LoD and physics layers. At this point sharp edged objects tend to come apart when one allows the viewer to generate the LoD layers. Smooth or more organic shapes fair better at the viewer’s hands.
One will have better looking man-made-looking objects at the various LoD’s if one makes the LoD layers. The second and third LoD layers have much less impact on cost than the first and fourth layers. The effect on cost of the fourth layer is more pronounced for smaller objects. Smaller objects get more simplification when viewed from a distance, meaning the fourth LoD layer is used.
By making the physics layer and controlling the size of the triangles one can reduce the physics cost to less than what one may get by the generated physics shape.
LoD or Physics?
The cost of an object may be controlled by either the LoD layers or the Physics layer. For shapes where one needs more complex shapes, say for a building or tunnel one can walk inside of, the physics is likely to control. Paying more attention to the physics shape would be more rewarding for such shapes.
The real numbers for cost are still undecided. The servers and the viewers work together to calculate the cost of a mesh. On the ADITI grid meshes are being analyzed to test the cost equations. As more data from sample meshes is collected and performance of the regions analyzed the equations are changed, tweaked, and re-analyzed. The server is changed and sometime after the viewer is changed to compliment the server change and/or vice versa. So, cost numbers given by the viewer are too far up or too far down the scale. The Lindens have said there will be an equivalency between mesh and prim. Finding a balance so well made mesh can reduce upload and tier cost while penalizing poorly made mesh is a challenge.