About the Size Penalty

[Pamela Galli]

I have this house with a bunch of separate meshes and am trying to get some idea of the best way to link them.  I made a little test house, 8 prims, one mesh, and there was no size penalty at all -- it stayed the same no matter what size it was.

When I import this new house with all the meshes in one .dae file, stretching it larger has a big effect.

This is one of my questions:  Does the importer give all the separate meshes the same size penalty? I remember reading something about this but did not understand it at the time.

For example, we (Nacy and I) are thinking of linking all the windows. I am figuring each window separately will be around .5 - 1.0 LI -- but if we link them all into one mesh, will the size of the mesh incur a size penalty?

 

TiA!

 

 

 

 

[Maeve Balfour]

Drongle will probably be able to explain this much better than I can....

The land impact in the display cost (if I am remembering the name of it properly) comes down to a combination of size, geometry complexity and LODs. Generally, the larger a mesh is, the greater the exponential land impact cost is - again, dependent on how much detail is in the actual mesh. So a complicated mesh will rapidly grow in land impact the larger it is sized - whereas a simple cube mesh will remain barely changed (I made a basic cube skybox a while back as a test - hollow on the inside and fully enclosed, using eight materials and UV-mapped, with the actual mesh used for the physics hull - and it's land impact didn't change much (if at all) right up to 64m3 (if memory serves me correctly, it remained less than 1LI). This would be due to its very basic geometry and (probably) because of my ultra-low LODs I applied (eight-face flat plane to describe the materials, for LODs 2-3-4)). At full size, I couldn't cam out far enough to see the first LOD swap.

With your windows - I am guessing that if you link them together into a larger mesh, the combined LI will probably exceed their combined total when left individually. This is an example where the linking will probably cost more than the individual meshes - the small size of your windows and low geometry works in their favour individually, but their overall size when made a single mesh will probably push up their overall cost.... I would guess that the larger mesh volume in this case means a much longer view distance before the LODs start to kick in, hence a higher display cost (the small individual windows will have a much shorter LOD range, so a very low LI).
Also, factor in that if you have customised LODs made for each individual window, you would need to use these and recreate their LODs by placing them according to the combined mesh linkset you mention - so the LODs keep their appearance. However, the caveat is that the overall LOD vertice count will rise, which will push the overall land impact up IN ADDITION to the overall larger mesh volume.
A higher LI cost as such, but on the flipside, you gain a longer distance before the window LODs start to kick in. As always, it's about compromises - whether you prefer lowest land impact, or longer range LODs.

Personally, in this case, if the LI count of the windows combined as a single mesh is far higher than them left as individual meshes, I would consider linking smaller groups of windows instead - say, all the windows on ONE side of your house as single meshes. The LI would still go up somewhat, but the volume overall would probably be much less - which should result in a lower LI. Plus you would gain a longer LOD range as a side benefit compared to the windows as individual meshes each (although I would still highly recommend using customised LOD meshes for each LOD swap).

Also, consider whether or not you need to define actual physics hulls for the windows. If these are going to be inset into a separate wall mesh, you could let the wall mesh have the full physics hull modeled for it (by creating a mesh consisting of large, simple cube shapes stretched to cover the wall sections AND the gaps where the windows would fit). In this case, IF the windows don't need to be interacted with physically, you could simply define a simple flat 3-vertice triangle as the physics mesh during the upload process - and inworld set the window meshes as phantom. The triangle physics mesh is only used to keep the uploader window happy (it requires SOMETHING defined for physics - and since anything will do in this case with the final result being phantom - the smallest possible object is fed to the uploader). With this miniscule detail for the non-used physics hull, the physics cost won't affect the final land impact cost, as it will most likely be the lowest weighting in the uploader - probably the display cost will be the highest factor, and hence the eventual land impact decider.

I hope this helps a little. :matte-motes-smile:

[Drongle McMahon]

Adding a bit to what Maeve said...

If it's determined by download weight, then the land impact of the combined mesges will be higher than that of the separate meshes if the separate meshes are smaller. The weight of the whole is more than the sum of the weights of the parts.

Remember that the land impact is the largest of the three weights, download, physics and srver. Two of these can vary with size, the download weight, and the physics weight of triangle-based (no "Analyze") physics shapes. The download weight increases with size in the manner described in detail in another thread. Triangle based physics weights decrease with increasing size. So the land impact will not vary with size in these cases ...

(a) if it is determined by server weight (but it might at larger sizes when download weight gets bigger etc.) - in that case you can probably save land impact by combining small meshes.

(b) if it is determined by non-triangle based physics weight. This includes the default convex hull, used if you select "Convex hull" physics shape type, and the multiple convex hulls produced when you select "Analyze" in the uploader and set to type "Prim". This means the physics weight is larger than the download weight, and you can probably save by providing a simpler physics shape mesh (or using linked prims an "None" for the mesh). I think there are rather few cases where a physics weight greater than donload weight is really needed.

© for very big meshes (radius = 0.5*sqrt(x^2 + y^2 + z^2) greaer than 43m), so that only the highest LOD is ever seen with the settings implicit in the calculation, the download weight reaches a plateau and no longer increaes with size.

It's important to use the More Info link, in the edit floater, to see which weight is the highest. You can often achieve effective savings by dealing with (a) or (b) without breaking up meshes. If it's (a), it will get worse if you break up meshes.

What Maeve said about choosing whether to break into smaller meshes is exactly right. It is a compromise between higher download weight and nearer LOD switching. It is a compromise that depends on what your LOD meshes are like. Linking does not affect the LOD switching behaviour. The components of the linkset behave the same as they would if they were unlinked. This causes another disadvantage of separate windows - they will switch LOD independently. This can be much more disconcerting than when they all switch at the same time, as they do if they are combined in one mesh. Windows are one of the easiest things to make good low LODs for because you can use simple planes with alpha textures for the lower LODs. That can make them quite cheap.

The difference between having (the windows in) each wall as a single mesh and all of them in a cube-shaped building is not as large as you might think. The increae in radius of the cube over each plane is sqrt(3/2) = 1.23. How big the LI increase is depends exactly where you are in the graphs, but should not be more than 1.5 fold (the realtionship is quadratic, not quite as bad as exponential). Dividing meshes to get lower LI is most effective when you start with something long and thin, or when you break something flattish up in both of the other dimensions.

[Pamela Galli]

Maeve and Drongle,

 

Thank you both so much -- it is a hard thing to explain and to understand, and your explanations were very thorough. I really appreciate the valuable resource you are providing here :-)

[Jenni Darkwatch]

Just a specific point about houses: I found it makes a fairly huge difference if the house is built in two distinct parts: Outside and inside. One example was a house I built with fairly detailed wood beam frames both inside and out. Breaking it in two allowed me to cheat tremendously: The outside walls (with windows and door) could be more detailed, while the inside basically has no lower LOD: The lower LOD was merely a box shape to fill the inside and provide "background" for the windows. The net effect is that one can peek into a window when nearby, but far away the windows become "opaque". It does save a lot on LI to go that route.

[Maeve Balfour]

Yah, I've been experimenting extensively with this method, Jenni, and it does make a vast difference overall. So much so, that I now design my builds specifically to take full advantage of hidden LOD swaps for the interior sections.

I tend to find that keeping rooms to individual mesh modules seems to work best in this manner - mostly due to their lower LOD swap ranges. I have found that this land impact advantage tends to be lessened dramatically if several rooms are grouped in a single mesh, simply due to their larger volume and longer LOD ranges. It actually SAVES land impact cost uploading rooms as individual meshes, purely to keep their volume compact and keep their LODs as short as possible.

Depending on the room, I either use a similar trick with the windows (having an opaque "screen" using only a couple of triangles usually, to mask the LOD swap for LODs 2-3-4) or if the room is totally hidden by the exterior "shell" of the building, I just simply utilise the flat plane I mentioned in my post earlier (with one triangle for each material used in the mesh). The land impact savings for interior rooms can be massive when done like this (separated from the exterior mesh completely) - generally via economical geometry and this savage LOD method, my room modules are usually between 1 to 3 land impact each (up to around 22.5m x 22.5m x 7.5m high). In these kinds of cases, I generally don't mind if the exterior mesh shell of the building is a little bit costly (within reasonable tolerances), due to the separated interior mesh modules being so efficient in comparison.

:matte-motes-smile: