Why does LI work like this?

[Pamela Galli]

I uploaded a chaise with a cushion and throw pillows. It came in at 6 LI -- high compared to the other seating.  I had a weird texturing glitch with the pillows so I separated them from the chaise and cushion and uploaded the two meshes together -- LI was 10!  The pillows by themselves were 1, the chaise and cushion, 9.  Then I uploaded ONLY the chaise and cushion -- which were now 2!  Linked with the pillows and a shadow, it is now 2.

I am still bumbling around joining and unjoining meshes and can see different ways produce very different results. But in this case I upload the same two meshes and there is a x5 difference between uploading them softlinked together, and uploading them separately. Why is this?  Are there some guiding principles somewhere (I mean, the For Dummies version).

[Chosen Few]

When you separated the various items, dod you also take the time to create LOD models and physics models for each?  If not, then that's your answer right there.  If you let the uploader auto-generate that stuff, it's going to do it differently for the different combinations of models.  Sometimes, it will even do it differently for two uploads of the very same model. 

The only way to fully control it is to create your LOD's and physics models yourself.  If you did that, and you're still getting wildly unpredictable results, I'd need to know the full set of numbers before I could speak intelligently about what's going on.

[Drongle McMahon]

Were the changes in LI due to download weight or to physics weight? (More Info link from edit dialog).

[Pamela Galli]

---

Chosen Few wrote:

When you separated the various items, dod you also take the time to create LOD models and physics models for each?  If not, then that's your answer right there.  If you let the uploader auto-generate that stuff, it's going to do it differently for the different combinations of models.  Sometimes, it will even do it differently for two uploads of the very same model. 

The only way to fully control it is to create your LOD's and physics models yourself.  If you did that, and you're still getting wildly unpredictable results, I'd need to know the full set of numbers before I could speak intelligently about what's going on.

---

I did not make LOD models for this item. I do understand some variation due to auto generation, but almost always we are talking about a couple of LI at most. In this case, the difference was extreme - 9 vs 2. I never see that. And the only difference was that with the 2 version, the item was uploaded by itself, not at the same time as the one-LI pillows. 

I have seen big differences in LI when, say I upload an armoire joined or not joined with detailed pulls, and I understand that the principle in operation is that the pulls are counted as being the same size as the armoire -- something like that. 

[Pamela Galli]

Hmmm I did not think to look, and Firestorm doesn't have that button. 

 

[Chosen Few]

---

Pamela Galli wrote:

Firestorm doesn't have that button. 

 

---

Sure it does.

When you click on "More Info" in the editor window, the Advanced window, pictured on the right, will pop up.  See where it says "Weights of selected".   The first three items (download weight, physics weight, and server weight) are what factor into land impact.  The total LI will always be the highest of the three numbers.

That's what I meant when I said I'd need to know the full set of numbers.

[Pamela Galli]

Oh wow I had looked for that before and not found it -- I used to use it on the LL viewer. 

 

 

2 LI wicker and cushion

 

8 LI Wicker and cushion (uploade w pillows)

Now when I rezzed the 8 LI one I dragged out of the trash, this is what it looked like -- normals are not flipped:

 

 

 

[Rahkis Andel]

Either it's an issue with your materials or it's way too high poly. Judging by how smooth it looks, I'd bet on the former.

[Pamela Galli]

I would agree with you except that mesh and the one I uploaded without the pillows are the same mesh. Same materials. Same face count. Same everything. And it did not look like this when I first rezzed it but only after I took it out of the trash and rezzed it.

 

 

[Drongle McMahon]

Since you have holes, and you mention texture issues on the pillows when they are included, it does sound like the effects of much too high poly count, although it's not clear from the picture. You can see by looking at the triangle counts reported in the uploader. If that is below 21844 (as it should be), then the following is irrelevant.

This is a bug that has been reported in a jira (BUG-1001, for those that have unrestricted access). When the triangle count for a material passes 21844, the uploader secretly starts a new material, complete with a new triangle list. It goes on doing this beyond the 8 material limit (more than 174752 total triangles if it's all one material in the collada file), but the extra materials are culled before the upload, leaving holes.

Where there are edges between different materials, the vertices along them have to be duplicated in each of the triangle lists. The secret materials cause many such edges. So they cause a rapid increase in vertex count, which increases the download weight.

The extent of the download weight increase depends on how the triangles are distributed between the materials, which depends on the order that they appear in the collada file. That depends on the 3D software and the collada exporter. So does the distribution of the holes that appear when the secret material count exceeds 8.

The effect of the extra materials on texturing is strange. It's not the same as having explicitly different materials, perhaps because the secret materials all have the same material name. At first, you can set different textures or colours on each, by dropping or using "Select Face", but then it starts behaving as if they were all one material. Perhaps this effect could describe the texture issues you mentioned? There is a picture of these effects in this thread.

You should be able to tell whether you have run into this issue if you look at the triangle counts of the meshes involved, as reported in the uploader. If the holes are caused this way, there will be less triangles than you expect from the poly count (x2 fo quads) in the authoring software, becaus the dropped triangles do not get included in the uploader count. 

If your triangle count is below 21844, then it is nothing to do with this bug.

 

[Pamela Galli]

Thank you, I think it was with the pillows above that number. 

The oddest thing about this is that the pillows I mentioned in the other thread took the texture fine until I relogged, then it reverted.  Also that the mesh did not display holes until I rezzed it a second time. 

I suppose not uploading the pillows with (tho not part of) the main mesh somehow avoided the problem. I have a perfectly nice piece of furniture now, 3 LI, pillows and all, so I am just going to be happy I was able to get it down to that and correctly textured.

 

[Drongle McMahon]

"pillows with (tho not part of) the main mesh"

If you mean that they were separate objects (<geometry>s in collada), then that would be strange, and not really consistent with this explanation. Each object is uploaded independently, as if they were uploaded separately, except for adding the relative scale/rotation/location transformations to be applied on rezzing as a linkset. So the counters should all be reset for each, just as if they were separately uploaded. On the other hand, if you mean just not connected, then omitting the pillows could avoid the holes, because the counters are not restarted. Then they are rezzed as one prim, not a linkset.

So it's good you solved it one way or another, but it's still worth keeping the 21844 limit in mind. I would have expected that would be enough for the whole thing with a couch like this, with appropriate use of smooth shading. You need to be kind to the residents' gpus. Certainly, anything that exceeds the 174752 triangles, where holes start to appear, is too high poly for efficient display. Of course, that's only about 85 sculpties, as far as the gpu is concerned, and lots of people think nothing of using such numbers!

[Pamela Galli]

Actually the face count in Blender is 4870 (without pillows). So I guess the rest is moot?  

 

 

[Drongle McMahon]

Assuming that's mostly quads, yes, well below the limit. But I thought you only got the problem when the pillows were there? Were the pillows separate objects, and if not, what were their counts? Not that it matters, but it would be interesting to know. 

[Pamela Galli]

Yes the high LI only happened if the pillows were uploaded along with the the rest.*  The pillows were separate objects and have 408 faces in Blender. 

 

* The texture problem I posted about in the other forum disappeared when I separated the pillows and uploaded them separately. Something about pillows + chaise is just BAD.

[Drongle McMahon]

It was definately something else then :matte-motes-smile:

[Alisha Matova]

I'd be curious to see the 'more info' for the parts that were uploaded together. I've had some serious failures when uploading that way. Mostly physics cost issues on the 'child meshes'(they revert to a non 'Analyzed' highest LOD model for physics).

 

Work around is to upload parts separately or plan on setting the child parts to physics shape 'none'. I think the first option is the safest. That said, there must me a fancy way to build .dae files to upload multi mesh, with lods, and physics for each each part. But I've never pulled it off. Lol.