Almost Free Land Impact - How?

[Chic Aeon]

I am SO impressed with this and another item from an unnamed here creator :D.   I am doing pretty well with the land impact with most things, but this just blew me away. Of course I want to know HOW -- you super smart people out there.

 

It SEEMS (from this item and another with even more amazing bang for your buck land impact count ) that part of the secret is duplicates. That is in both item there are many of the same mesh object. So in this case the confetti and in some ways the flowers. The other item has about forty of the same object (a simple sphere with some extruded bit) PLUS many other simple shape objects and weighs in at 1).

These are both smallish items.

 

The box below which includes a redelivery tracker (which I assume is one prim) has a land impact of 3.

 

 

[KlistiesSeMio Ewinaga]

I actually liked that box of flowers so much that when I purchased something I kept the box out after unpacking.

[Drongle McMahon]

If you used the Advanced menu to Show Debug Parameters, type RenderVolumeLODFactor and set it to zero, then zoom in and out a bit, you will see the lowest LOD (level of detail) mesh for the object. For objects this size, the lowest LOD mesh has by far the highest influence on the LI (given that is needs the simplest physics). A very simple of the lowest LOD that will produce low LI, for surprisingly detailed high LODs. This become less effective as the object gets bigger. Don't forget to set RenderVolumeLODFactor back to where it was, or you will have a very dismal experience!

[Chic Aeon]

Actually I understood that , but -- that doesn't seem to be the whole story. The object actually does amazingly well at far distances too. Granted no one can see the string and the cut paper pieces but the flowers held together well from a far distance.

I have something I made a couple of days ago that used a fair number of duplicates. It is fairly large (meter by meter ish), and fairly complex but it still weigh in at .5. I am going to do some experimenting with duplicates as I think there is something to the theory *wink*. Of course since I know very little I could be completely wrong - LOL.

Perhaps, since the program "understands" one object (let's call it a snowball for instance) it takes it much less processing to render duplicates of the same snowball. I saw a mention "somewhere" about CHILDREN object in Blender. Didn't read it so not sure if that has anything to do with this.

[Drongle McMahon]

There was some discussion during the mesh beta about the possibility of using instancing in the download format - so that you could have multiple scaled and move versions of the same mesh for the same download weight as one. For example, I made a tree where all the branches were scaled versions of the same mesh. However, that was not implemented (unless it's been secretly added?). I guess it's possible that having self-similar structures might nevertheless increase the efficiency of compression. The download weight is calculated from the size of the data after compression. So that would reduce the weight. I can't think of anything else like that.

The trouble with just looking at a distance is that you can't see the exact structure of the lowesr LOD because it's too far away. If the LODs are well designed, it will continue to look good despite drastic ploy reduction. By adjusting rvlf instead, you can look in wireframe view and count the triangles! (hint - you have to turn off Render Types, surface patch, sky and water in the Advanced menu to see wireframes clealy).

[Chic Aeon]

So here are the results of my test. A quick and messy example but it tells an interesting story.

 

 

I made a simple hollow cube (yellow). From there after mapping I duplicated it to 12 (red) then 24 (blue) then because of the results I got there to 18 (orange).  There definitely seems to be a relationship between duplicates, upload cost and the like. I used a physics cube (my name) for the physics and went with the default LOD (not good enough but seemed like the fairest test for this). I later uploaded the 18 cube crate at a very small size to see what the difference might be.

Here are the results.

Physics and server costs stayed the same throughout

1 = .060 upload   

12 = .116 and .177 (two tries with the exact same info - the beta grid uploader does that often)

24 = .974 and 1.045 (again two different times)

So what about 18? There is a  HUGE jump between 12 and 24.

18 = .534

Uploading the 18 grid at a much smaller size (as shown) was .467 and it falls apart horribly LOD wise.

So, I found this all very interesting and will definitely work on duplicating more as it does seem like it could be a very useful tool when you can use a lot of the same shape in a desgin.

Back to work.

 Edit: and I do see that the only real test would be to build all those empty cubes by hand, map and put them together . Unlikely I will feel compelled to do that

[Drongle McMahon]

Interesting.

Was the cube UV mapped? I ask because if you upload without a UV map, you get random data instead, and that can change the download weight quite a lot because of different amounts of fragmentation. That could explain the different weights you got with the same mesh on different occasions. If you make a seamless UV map (eg project from view in Blender) it won't be any use for texturing, but it might stabilise the weights. As long as it's seamless, it won't add to the vertex count.

These replications without rotation leave a lot of identical normals, which might contribute to better compressibility. It would be interesting to see if there was any effect if you gave them all random small rotations.

The size if the whole mesh has a big effect on download weight. If you want to look at the effect of complexity separately from size, you could resize the rezzed objects to the same XYZ dimensions.

I look forward to seeing some more numbers.

[Chic Aeon]

Yes, see post

 

From there after mapping I duplicated it to 12 (red)

 

Even after mapping this seems to happen often for me, both on Aditi and Agni. I write down all my perameters when I get to the final stages, putting them in the filename of the object, so "fancytable-320-320-320-11" for example. STILL, I sometimes get varying upload numbers. And yes physics and size are set the same. It may be a Firestorm thing, I know not.

Seems more like the upload server is persnickity LOL.

The size of the mesh when uploading didn't have much impact "in this case". I have seen that in other though and understand.

 

Later in the day I made a more complex and interesting bookcase made with the same cube object and duplicates. I resized the duplicates which no doubt made some differences. I ended up with a ten cube bookcase that would still weigh in at .5 (didn't check the actual). I set the LODs up a lot though and it is 1. So more experimentation is needed.

 

Actually I am happy to know what CAN be acomplished. It gives us something to shoot for .

 

 

[Rahkis Andel]

I have nothing useful to add to this thread, except to say thank you for going through the trouble.

I just find it funny that not two days ago was I wondering if Second life did any optimization via instancing and then as if you had read my mind...

[Drongle McMahon]

As far as I recall from the discussions, SL does use instancing at the object level, but not internally within a single mesh model.

[Kwakkelde Kwak]

---

Chic Aeon wrote:

 

STILL, I sometimes get varying upload numbers. And yes physics and size are set the same. It may be a Firestorm thing, I know not.

---

To make a fair comparision between your uploads, make sure not only your UV mapping and size are consistent. Also make sure the actual shape is just that. The uploader doesn't reduce the polygons the same way every time, this is the case for both LoD and physics shape. If you want to make a fair comparison, do what you should almost never do for a final product and use your highest detail for all slots, including the physics slot. (or make some custom models, but for a test this would be a waste of time) No idea if analyzing has some randomness to it, but you might want to avoid that aswell and use the triangle based shape..

[Chic Aeon]

Actually if I am "done" and back on the Agni grid AND get a much larger number than testing showed it "should be" (this rarely happens but has with as much the same criteria as I can possibly make it ----- I wait and hour or a day and try again and I get very close to beta grid numbers. I can see that the numbers wouldn't be EXACT between grids but a land impact point is a big deal when you are working with very low LI numbers. Cost change too, sometimes dramatically like $14 to $26. Not a lot of lindens but I want ALL I can get from them .

Also, I read in the wiki (I think - somewhere official anyway) that the uploader REMEMBERS your last perameters and tries to upload the same mesh by looking at Aditi and Agni numbers and giving you your last one for that file as a default. That may have been a plan, but I definitely don't see it in the works. Again, might work with the official viewer -- just not willing to go there *wink*.

 

 

[Chic Aeon]

Just wanted to note that I was talking to a friend tonight (RL and SL friend but not a creator) who mentioned that the phenomenon MIGHT be program specific. I hadn't thought of that, but we all know that some software does some things well and leaves out other things you WANT . This could be something that is new to one of the software programs that make mesh. Hence, those of us in free Blender (who I have sort of become freinds with) may not have the ability to do.

A good thought, but only that. Still wanted to add it to this post.

[Kwakkelde Kwak]

---

Chic Aeon wrote:

 

Also, I read in the wiki (I think - somewhere official anyway) that the uploader REMEMBERS your last perameters and tries to upload the same mesh by looking at Aditi and Agni numbers and giving you your last one for that file as a default.

---

It does, that's what the slm file is for. But the numbers you are talking about are estimations. The real calculation is done when you hit the calculate button. I haven't thourougly tested this, but I got the impression that if you use files instead of those estimations, you get matching results. Either way, I would never use the generator, I (and you and all others) can do a much better job myself.

[Chic Aeon]

Just wanted to add that I did finally get this .  It seems to be a combo of size (natch), complexity and all physics models. The BIGGEST enlightenment for me -- being very new and all -- is the difference between joined items and those not joined. And the simple cube when you really don't need an actual physics model was a good thing to discover.

I am not certain that the duplication "really" helps keeps things down but it seems like it may. I have done other tests with duplicates and you get a huge number of identical items (joined which I didn't do in my original test as I had no clue about that then).

 

So put it all together and you can get some very exciting super low impact items.