Jump to content

What it is actually possible to build in Second Life


ChinRey
 Share

You are about to reply to a thread that has been inactive for 2975 days.

Please take a moment to consider if this thread is worth bumping.

Recommended Posts

This may look like blatant self promotion and maybe it is. ;)

But not only. I've seen so much talk here and elsewhere about lag, how heavy SL is to run for our poor li'l computers. And of course it's Land Impact or prim count if you like. It costs real money and there's just never enough! LOD - that silly term so few understands: does the build look good at a distance too or just at point blank range? And of course, recently there's been talk about how mesh is The Root Of All Evil In Second Life.

I decided to make a little demonstration what we still can build in SL if we as builders take resource management seriously. If by chance anybody are interested, go to http://maps.secondlife.com/secondlife/Coniston/205/116/1802. Just walk slowly down the asphalted path from there and the landscape will tell you the details.

Link to comment
Share on other sites

I heartily recommend anyone who builds in SL (not just landscapes either) to take a look as it really does show what can be achieved using the resources and tools we have. As one of those with a poor lil' box (creaky 7 yr old win7 32bit 2 gig ram and yay a gt 610) even cranked up to 256 dd on ALR the rez was - instant :)

The homage to olde fashioned prim building alone is worth it - and I fully intend to shamelessly make use of a couple of techniques in my own stuff as they fell in to the 'of course why did I not see that before' category. Educational:)

And keep off the grass =^^=

Go see this, and if this is 'self promotion' then we need a lot more.

 

  • Like 1
Link to comment
Share on other sites

Thank you seanabrady, mikka and Rolig. ^_^

 

One very important thing I forgot to mention is that this is not a pure mesh landscape. It's probably more than 50 percent mesh but there are lots of prims and sculpts too. And the birds and the bees are particles of course. I don't think anything like this could be possible without using all available building materials.

Link to comment
Share on other sites

I just wish I already totally understand the LOD stuff ..I mean, I know what it is, but still don't understand completely how to create mesh that does not collapse and yet is download and server friendly and such. Your tour just made me drool more but, also more determined to learn! Would be great to achieve that in this life though :smileyvery-happy:

Link to comment
Share on other sites

 


Six Igaly wrote:

I just wish I already totally understand the LOD stuff ..I mean, I know what it is, but still don't understand completely how to create mesh that does not collapse and yet is download and server friendly and such.

The basic principle is quite simple: The different LOD models are for viewing at different distances so If you know which distances the different models are used at and are able to figure out roughly which details of your at those distances, you're well on the way to make efficient mesh.
   This is why I always warn people against using LOD models generated by the uploader. The uploader has absolutely no idea what is important to keep for each LOD model. Even the most inexperienced mesh maker can easily beat the most advanced computer program at this. And the uploader is anything but advanced - not tesselation, no dissolving, just crude triangle decimation.
   So, if you take the time to make your own LOD models and check them out on the beta grid before the final upload, you're off to a good start.

But there is more of course. If you just upload your mesh with the perfect LOD models - the kind that looks decent at any distance even without that laggy RenderVolumeLODFactor increase irresponsible builders often recommend - you'll probably end up with a land impact a bit higher than most people here would accept. Unless you find other ways to save LI of course and fortunately there are quite a few techniques there.

First is to eliminate unnecessary triangles from all models. Quite often you see meshes made completely from grids of equal sized triangles. That is usually a very bad idea - and a sure sign the maker doesn't understand mesh in Second Life. I don't really see any point explaining here - Drongle already did that so well in another thread only a few days ago: https://community.secondlife.com/t5/Mesh/Mesh-Landscapes-Quirks-amp-Oddities/td-p/3000259

Second: don't do mesh for mesh' sake. There are lots of builds better done as prims or sculpts. If you visited my park and noticed the Backwoods Forest, for example. That's a quarter sim worth of land densely packed with trees. Yes, they are just flat trees - you don't really need anything more than that for an offsim background - but even so, the land impact it would have had as mesh would have been mind-boggling. As sculpts it's just 8 LI. (The effective LI is actually just 2 because of balancing.)

Third: Balance the weights. Far too many builders try to put too much into a single mesh. The general rule is: few large meshes give low server weight but high download weight, many smaller meshes give high server weight but lower download weight. It is the highest of the weights that counts as land impact so if you can find the sweet spot there - usually where those two weights are almost the same - you can save a lot of LI.
   I also think balancing in a larger context. The forest I mentioned has a LI of 8 on its own but it's all download weight. By linking it to objects with higher server weight than download weight I got a linkset with much lower LI than the individual parts had on their own.

Fourth: Optimize for compressability. Download weight depends on file size but that is not the same as the amount of raw data. The files are compressed before they are transferred and before download weight is calculated. You can save a tremendous lot of LI by making the mesh as easy as possible for the compression algorithm to handle. This is a little bit of a black art with few fixed rules so it takes quite a bit of experience to take full advantage of it but the basic principles are easy enough for anybody to understand and even they can help a lot.

Last but not least, remember: Content may be King but without Queen Context he's nothing at all. Take a look at the tree groups at the western end of the park. If you study them really close, you may notice that they are actually extremely simple meshes. The average number of tris per tree is somewhere between four and five. The only way to achieve that is to make multi-tree meshes in very strict geometric patterns. That doesn't look very good at all of course but the different meshes are positioned to break up each other's geometry and the result are small groves with trees in a very natural looking seemingly random pattern.
   You can find similar examples all over the park. Everything there is selected and positioned to support everything else as well as possible. Context is perhaps the most important secret behind my park and what SL lacks more than anything else. Too often we see scenes with lots and lots of lovely, elaborately made objects fighting each other for attention. It's like a rock band with 20 solo guitarists and nobody to play bass and drums. Build your virtual world with pieces that fit together and also fit in with what others have already placed there and everything just looks bigger, better, more real. That's the true secret behind good VR building really - and good RL building too for that matter.

Link to comment
Share on other sites


MoondustMaia wrote:

I love landscaping around my houses, but don't have the best graphics abilitiies, so it's nice to see landscaping that looks nice while still being low lag. 

Thank you :)

I'm really glad to hear it is still low lag. I added 101 new flower beds and 23 LI worth of gravel road just before you visited and was a bit worried about that.

Link to comment
Share on other sites


The basic principle is quite simple: The different LOD models are for viewing at different distances so If you know which distances the different models are used at and are able to figure out roughly which details of your at those distances, you're well on the way to make efficient mesh.

 

I am still not sure what is the best way: create the model with all the details and then step by step breaking it down to LOD_0 or start with the most simple one and build it up step by step to the final model or building each LOD from scratch. (It is fun to figure this all out though.)


But there is more of course. If you just upload your mesh with the perfect LOD models - the kind that looks decent at any distance even without that laggy RenderVolumeLODFactor increase irresponsible builders often recommend - you'll probably end up with a land impact a bit higher than most people here would accept.

Exactly what happens most of the time. Btw, I have the RenderVolumeLODFactor set at default. I figured when a model looks good then it is really good.


First is to eliminate unnecessary triangles from all models. Quite often you see meshes made completely from grids of equal sized triangles. That is usually a very bad idea - and a sure sign the maker doesn't understand mesh in Second Life. I don't really see any point explaining here - Drongle already did that so well in another thread only a few days ago:

 

I have read (and bookmarked to re-read) about all the significant threads about mesh, including Drongle's. That is one way how I learn. I guess it is just a matter of practicing, which triangles to remove and which not. A small hurdle for me is that most explanations are based on using Blender which I don't use (Cheetah3D here). Of course mesh is mesh, triangle is triangle and so on. So far I managed to 'translate' from one to the other.

Second: don't do mesh for mesh' sake. There are lots of builds better done as prims or sculpts. 

 

I noticed yes. For me mesh comes in when I want to have more organic shapes which are not possible with prims (as far as I know).

Third: Balance the weights. Far too many builders try to put too much into a single mesh. The general rule is: few large meshes give low server weight but high download weight, many smaller meshes give high server weight but lower download weight. It is the highest of the weights that counts as land impact so if you can find the sweet spot there - usually where those two weights are almost the same - you can save a lot of LI.

   I also think balancing in a larger context. The forest I mentioned has a LI of 8 on its own but it's all download weight. By linking it to objects with higher server weight than download weight I got a linkset with much lower LI than the individual parts had on their own.

 

Clarifying!

Fourth: Optimize for compressability. Download weight depends on file size but that is not the same as the amount of raw data. The files are compressed before they are transferred and before download weight is calculated. You can save a tremendous lot of LI by making the mesh as easy as possible for the compression algorithm to handle. This is a little bit of
a
black art
with few fixed rules so it takes quite a bit of experience to take full advantage of it but the basic principles are easy enough for anybody to understand and even they can help a lot.

Right! For now it still is but will get it one day or the other.

Last but not least, remember: Content may be King but without Queen Context he's nothing at all. Take a look at the tree groups at the western end of the park. If you study them really close, you
may
notice that they are actually extremely simple meshes. The average number of tris per tree is somewhere between four and five. The only way to achieve that is to make multi-tree meshes in very strict geometric patterns. That doesn't look very good at all of course but the different meshes are positioned to break up each other's geometry and the result are small groves with trees in a very natural looking seemingly random pattern.

   You can find similar examples all over the park. Everything there is selected and positioned to support everything else as well as possible. Context is perhaps the most important secret behind my park and what SL lacks more than anything else. Too often we see scenes with lots and lots of lovely, elaborately made objects fighting each other for attention. It's like a rock band with 20 solo guitarists and nobody to play bass and drums. Build your virtual world with pieces that fit together and also fit in with what others have already placed there and everything just looks bigger, better, more real. That's the true secret behind good VR building really - and good RL building too for that matter.

I will return as often as I need.

I know I still have a lot to learn (but then again, most likely I will learn till the day I'll pass). I also know I'm in the right direction and that I understand more then I'd thought at first.

Thanks for sharing Rey :smileyhappy:

Link to comment
Share on other sites


SixIgaly wrote:

I am still not sure what is the best way: create the model with all the details and then step by step breaking it down to LOD_0 or start with the most simple one and build it up step by step to the final model or building each LOD from scratch. (It is fun to figure this all out though.)

I prefer to start with the detailed model and simplify it. I think most builders do it that way but I know of at least one who does the opposite. I think that is more a question of personal preferences than right or wrong actually.


SixIgaly wrote:

Of course mesh is mesh, triangle is triangle and so on.

Not quite. The dae format isn't a strictly formalized file format but rather a variant of xml, that is a text file describing the mesh. Different programs describe the same mesh in different ways and that can have a huge impact on how well SL's compression algorithm can handle it. Drongle once did some tests demonstrating how you could make significant changes to a mesh's LI simply by changing the order the triangles were listed in in the dae file. I don't know how "SL friendly" Cheetah 3D's dae files are. But generally Blender doesn't perform particularly well there so I would assume a program with such a lovely cat name will do as well or even better.

 


SixIgaly wrote:

For me mesh comes in when I want to have more organic shapes which are not possible with prims (as far as I know).

You never know, there may still be undiscovered prim shape treasures. As far as I know the equal sided prim pentagon wasn't discovered until two or three months ago (I'm to modest to say by whom) and there may well be others too.

As for sculpts, if you actually need all or most of those vertices, if you want smooth normals all the way and if you can find good solution for the LOD, UV map and physics shape issues, a sculpt is usually a better option than a mesh, That seems like a lot of ifs but it's amazing how often it actually fits the bill when we're doing nature. (One thing that really makes me sad btw, is seeing how many great sculpt tree makers have switched to mesh recently, often producing trees with less details, worse LOD and higher LI than their own old sculpt masterpieces.)


SixIgaly wrote:

Clarifying!

I can give it a try with a simplified example: Let's say you have a mesh with 4 LI, you check the exact weights and find that the download weight is 3.7 and the server weight 0.5 (let's ignore the physics weight for this example). Now, link it to seven cube prims, each with a download weight of 0.06 and a server weight of 0.5. The result is a linkset with 4.12 download weight (0.06*7+3.7) and 4 server weight. That gives a land impact of 4, the same as the mesh had on its own. Seven prims to play with and no LI added!

Link to comment
Share on other sites


ChinRey wrote:

I prefer to start with the detailed model and simplify it. I think most builders do it that way but I know of at least one who does the opposite. I think that is more a question of personal preferences than right or wrong actually.


 

That sounds plausible yes. So far I don't know yet but I think starting with the detailed model is what suits me best.

 


ChinRey wrote:

Not quite. The dae format isn't a strictly formalized file format but rather a variant of xml, that is a text file describing the mesh. Different programs describe the same mesh in different ways and that can have a huge impact on how well SL's compression algorithm can handle it. Drongle once did some tests demonstrating how you could make significant changes to a mesh's LI simply by changing the order the triangles were listed in in the dae file. I don't know how "SL friendly" Cheetah 3D's dae files are. But generally Blender doesn't perform particularly well there so I would assume a program with such a lovely cat name will do as well or even better.


 

 

I really can't tell how SL friendly the Cheetah models are, in most cases they kinda look as expected but that does not say anything about how friendly they are.

 


ChinRey wrote:

You never know, there may still be undiscovered prim shape treasures. As far as I know the equal sided prim pentagon wasn't discovered until two or three months ago (I'm to modest to say by whom) and there may well be others too.

As for sculpts, if you actually need all or most of those vertices, if you want smooth normals all the way and if you can find good solution for the LOD, UV map and physics shape issues, a sculpt is usually a better option than a mesh, That seems like a lot of ifs but it's amazing how often it actually fits the bill when we're doing nature. (One thing that really makes me sad btw, is seeing how many great sculpt tree makers have switched to mesh recently, often producing trees with less details, worse LOD and higher LI than their own old sculpt masterpieces.)


 

Interesting. I'm getting to understand that in some ways mesh is overrated, hence builders switch to mesh simply because the mass wants mesh.

 


ChinRey wrote:

I can give it a try with a simplified example: Let's say you have a mesh with 4 LI, you check the exact weights and find that the download weight is 3.7 and the server weight 0.5 (let's ignore the physics weight for this example). Now, link it to seven cube prims, each with a download weight of 0.06 and a server weight of 0.5. The result is a linkset with 4.12 download weight (0.06*7+3.7) and 4 server weight. That gives a land impact of 4, the same as the mesh had on its own. Seven prims to play with and no LI added!


 

I think the word I've chosen was not the most appropriate one. Maybe I should have said something like: "That was clear!" This simplified example however cleared it even more. And yes I've noticed this when I was experimenting, about the LI staying the same, I just didn't understand why.

Link to comment
Share on other sites

Sorry, I walked on the grass. :D

I find most of the lag experienced in Second Life has been from the use, abuse and overuse of large textures.

I've had no problem rendering meshes and scenes, but huge textures on small items chokes up the frame rate. Also, the meshes people wear with the sometimes unoptimized clothing, jewelry, weaponry and scripts will do it long before mesh.

Link to comment
Share on other sites


entity0x wrote:

I find most of the lag experienced in Second Life has been from the use, abuse and overuse of large textures.

Oh yes, definitely. Textrues are nearly always the biggest cause of lag in SL. The only exeptions are the most heavily mesh body infested places and some locations where somebody has gone absolutely mad with scripts or with physical objects or some other extra heavy rare features. Texture data may be easier for the computer to process than 3D-thingies data, especially moving 3D-thingies data but there's just so much of it! Even in the most somberly textured environment, the textures are bound to account for far more of the raw data needed than everything else combined.

 

This was probably the biggest challenge with the park. With plants you need lots of different textures. Reusing a rose bush texture for the oaks just seemed ... wrong somehow.

It worked though, even with all the new textures of all the new flower beds. I tried to keep the resolutions down to a sensible level and also of course reused textures wherever it was actually possible. The sheer size of the build also helps. It'll take quite a while to get through it all and that means the textures have more time to load.

Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 2975 days.

Please take a moment to consider if this thread is worth bumping.

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share

×
×
  • Create New...