The resource drain of mesh

[Jupiter]

This article from New World Notes is almost a month old now so many of you might have seen it. It is about mesh bodies, but a person adds in the comment section that mesh in Home and Garden also has a detrimental effect.

My report earlier this month that some of the top-selling mesh bodies in Second Life -- such as Maitreya, Belleza and SLINK -- are extremely resource heavy, and degrading SL performance for everyone around them, has caused a lot of conversation. Many readers claim the polycounts being reported can't possibly be that much. So here's an easy way to check for yourself:

Download the acclaimed, graphics-optimized Black Dragon SL viewer, then select Dragon > My Useful Features > Performance Tools > Advanced Complexity Window. As you can see above, the Maitreya body here uses well over 100,000 polygons.

How much is that? By contrast, a character in Red Dead Redemption 2, the latest AAA hit, has polycounts estimated in the 40,000-60,000 polygon range. (Here's a handy site showing character polycounts for other top AAA games with high-end graphics.)

You want to see something insane? Here's a Belleza body with over 500,000 polygons:

Again, these are bodies that require graphics processing that even top AAA games on PS4 or another major console don't require. And they are worn by a majority of regular Second Life users. Many of whom use low-end laptops. Or older desktop PCs. Or worse.

Read the entire article here.

It's a statement that I've seen made before, one I've also seen some people reject, and lag is caused by multiple things, not just mesh. But so what? I don't think LL is going to do anything about it.

 

[Sid]

SL is what it is and always has been: A vanity show. I want to look prettier than everyone else. My garden has to be more detailed than my neighbors. And more details means more polygons.
Plus: LL has no control over the skills of the manufacturers of the goods. Everybody can upload who is willing to pay the upload fees. One of the charms of the platform.
Lag is almost a build in feature if you don't want to enjoy the platform with a standard avatar in system clothes on a deserted dessert sim.

 

[Chin Rey]

It's a statement that I've seen made before, one I've also seen some people reject,

Yes, some people are still in denial.

I don't think LL is going to do anything about it.

What can they do? You can't put the genie back in the bottle. Dressing up your avatar is pretty much the only activity with a relatively broad appeal that is left in SL. Take that away and the old users will leave in droves and it won't be nearly enough to make it attractive to new users. They don't have the expertise they need to fix it either. The Moles and Lindens themselves don't know how to make optimized mesh, that's evident in their own SL builds and even in how Sansar is made.

Mesh or "polylist mesh" as we really should call it (since nearly everything is some sort of mesh at the end) is inherently far heavier to handle in an online environment than parametric building materials like prims (and SpeedTree and Archimatix). Avi Bar-Zeev actually once went so far as to say polylist meshes aren't suitable for online virtual environments at all. He exaccerated a bit there but he did have a very important point.

Polylist mesh in Second Life is particularly problematic for several reasons. One is that the code handling it is essentially a crude hack of the sculpt code which again is a really crude hack of the prim code. And of course, fitted mesh is a crude hack of the regular mesh code. All this crude hacking causes a huge overhead. There are several features vital for performance on other engines that simply can't be applied to SL and others that are implemented in seriously inefficient ways.

It's poorly documented too. Working with mesh is far more difficult than buiding with prims and there are far more ways to screw it up. The documentation LL has given the users is hard to find, full of holes and as often as not incorrect and downright misleading.

Linden Lab ignored all of this for almost a decade. They forgot all about client side performance back in 2007 when Ondrejika was fired and it's only very recently it has started to dawn on them that it is important. The result is a resource management system that only takes server side performance into account. It's only recently they've started to realize that client side performance is vital for their bottom line. They are trying now but they still don't really understand and they don't really have anybody on their staff qualified to fix the problems. It would probably have been a lost battle anyway - there are simply too many skeletons in the closet to fight against.

A special problem that only applies to fitted mesh, is that the LoD system, one of the most important mechanisms for reducing render load, is completely broken for fitted mesh. That means that even fitted mesh from brands that really try their best to keep the lag down - such as Maitreya and Slink - is far heavier to render than it should have been. To make matters worse, the code that calculates render complexity does not take this into account, it calculates not the actual render complexity, but what it would have been if the LoD system had worked as intended. So the one method we have for actually measuring avatar render load is so flawed it's all but meaningless.

 

[Chalice Yao]

As Chin said, it's not inherently the concept of mesh that is to blame for anything (as really, prims are mesh too, as is every other standard 3D model). I see two primary reasons for it:


1) A lack of resource management.

Creators can upload 1-million-triangle linked mesh piles with a ton of 1024x1024 textures on them without any downsides for the creator themselves - the stuff looks pretty.
Despite, really, being just as pretty if they had optimized everything to proper industry game-asset level instead. And it gets bought.
The ones who really suffer in the end are the rest of folks in SL who then get to enjoy an amount of geometry that should never have been rigged.

You know the arguments, they popped up in several threads, I don't want to delve down that rabbit hole yet again.

I consider this the primary reason for rigged-mesh-specific lag ( which also extends into shadow performance, which is why I wrote that 'simplified avatar shadows' thing ) - and it's both a technical and a social thing to fix.



2) The implementation of mesh in the SL engine (though not for the same reasons).

Unlike Chin though, I don't see the main problem in the staggered prim -> sculpt -> mesh evolution of the code, but more in the fact that the SL engine these days, for all intents and purposes, is starting to show its age HARD in a much, much more fundamental way it's programmed as a whole, for almost all of the renderer's features.

It is a convoluted mess of a multi-inheritance, pointer-chasing, memory-jumping, virtual-function-stuffed, Object-Orientation-Über-Alles behemoth, and no single modern and well-known 3D Engine that is supposed to tickle remotely good FPS out of non-topshelf hardware is programmed like that these days. The games industry started to slowly realize that issue ~10 years ago. It's the worst thing you can do for hardware performance.

Fixing *that* would speed up SL's clientside performance from a technical standpoint way, way more than a mere re-implementation of just the mesh internals would.
That problem is merely technical, but it is also a HUGE challenge, as it would need a refactoring and rewriting of a LOT of the client's internals. So in general, mesh performance in SL is stuck between a rock and a hard place, from several technical and social standpoints.


.

 

[Khamon]

An old-fashioned Linden cube, freshly rezzed from the build interface, has eighteen (18) triangles per side.

 

[Chalice Yao]

Yah, the prims have more geometry than they'd theoretically need. Not much, but it's there. It's because of how the prim torturing feature works - it sadly doesn't keep things as simple as they could theoretically be.

 

[Chin Rey]

Unlike Chin though, I don't see the main problem in the staggered prim -> sculpt -> mesh evolution of the code, but more in the fact that the SL engine these days, for all intents and purposes, is starting to show its age HARD in a much, much more fundamental way it's programmed as a whole, for almost all of the renderer's features.

I don't completely disagree with you. But I have to point out that Unreal Engine is older than Second Life. And so is SpeedTree, the software that amongs other things provides all that lush and still low lag vegetation in the best UE and Unity environments. The difference is they way the software has been updated over the years.

 

[Chalice Yao]

I don't completely disagree with you. But I have to point out that Unreal Engine is older than Second Life. And so is SpeedTree, the software that amongs other things provides all that lush and still low lag vegetation in the best UE and Unity environments. The difference is they way the software has been updated over the years.

Yes, it is.
That's kind of what I meant with how the whole renderer is programmed, and how the games industry started to rework those in the last decade - doesn't matter if it's Unity, Unreal, IDTech, CryEngine and what-have-you. They all did their thorough work on revamping the internals. Understandably so, as it's kind of their prime product.

LL is, even these days, adding new render features on top of the horrible fundament that is the current way things are set up throughout the client. They need to revamp, refactor and rewrite not just their rendering pipeline, but many other parts of the code that makes use of the same monolithic objects they use for everything. ASAP.

 

[eku zhong]

Plus: LL has no control over the skills of the manufacturers of the goods. Everybody can upload who is willing to pay the upload fees. One of the charms of the platform.

Not so true.
IMHO LL should have made avatar 2.0 aka a fully functional proper mesh body with bento hands, feet with proper morphs for flat, medium and high heel, facial expression morphs etc etc
then creators could work on a proper dev kit, doing away for the most part of alpha hiding HUDS
(these are necessary now because creators are using badly decimated dev kits with rigging that does not quite match the inworld mesh body)

Just this alone would reduce lag by a massive degree... It would have been easy to have a choice to use either the original avatar or 2.0.. similar to switching gender in SL.

Now ppl say LL can't do it because the ship has sailed, but they could, if they really really wanted to.

 

[nebula]

Get a better computer then and stop trying to run sl on the same computer you did in 2013. Because sl looks a lot better now. Yeah more stuff to render, but it looks better than it did.

 

[Chin Rey]

Get a better computer then

Do I misunderstand or are you offering to pay for it?

To me, lag is mainly about retention and concurrency rates. Whether we like it or not, most people are not willing and able to shell out money for extra hardware to run Second Life. If they're not happy with how it works on the computer they have, they leave. When it comes to bandwidth, it's not even always a question of money, not everybody have access to top notch high speed connections no matter how much they are willing to pay. This is one of the reasons why High FIdelity, Sansar and Sinespace are struggling to reach three digit concurrency figures and why they look so empty and desolate even compared to SL. Sinespace isn't that much heavier to run and even Sansar is technically accessible for most people with computers strong enough to handle SL but that's simply not enough people to fill up those places. SL soldiers on with decent visit counts because it has a large base of old users who can't let go of it but that won't last forever.

Even if you have a top-of-the-top game computer, the lag still seriously limits what you can do in SL. You can live happy in your skybox, entertaining yourself dressing your avatar in the most outrageous costumes but try to go to a crowded place with dozens of other avatars, try to travel around and explore, try to do some fast action stuff. And do this with a graphics setting that really shows SL as close as possible to a modern 3D environment, Ultra graphics, 1024 m draw distance, LoD factor 4, jellydoll filter off...

Many people seem to think that lag reduction is all about reducing the visual quality. It isn't, it's about reducing the overhead. Take a look at Sinespace. It's not that much heavier to run than Second Life and the content is not particularly well optimized. Yet the graphics quality is so much better you can't really compare the two.

Or check out Unigine's Valley Benchmark: UNIGINE Benchmarks
It's old - it dates back to 2013 - and it's specially made to test the limits of your computer's performance. It still runs on a computer that would choke on SL and the graphics quality puts everything you've seen in SL or Sansar to shame.

That's the difference in software. As for what can be done with what we have, go to the OS grid and play Total Sorbet's Outbreak zombie game: Outbreak - OpenSim Virtual Worlds Directory - OpenSimWorld
A high speed shootup game with up to 40 zombies you have to kill before they kill you. It works well as long as nobody uses mesh bodies (the zombies are all clones of previous players) but the moment fitted mesh enters the equation, it breaks down.

I'm looking at LL's latest Premium gift right now. It's a fairly simple cottage, looking good but not particularly detailed or advanced. The land impact is 100, render weight 28,554, VRAM a wopping 205,828 Kb. A good content creator could easily cut those numbers down to 50/15,000/150,000 with no reduction whatsoever to its looks. Me, I could do 30/10,000/75,000 without raising a sweat.
This is fairly typical of SL mesh. If anything, the Moles are marginally better than the average SL content creator. It's fair to say that half the time your computer should have used rendering the scene is wasted on unneccessary work that makes no difference to how it actually looks.

To sum it up:

Because sl looks a lot better now.

It's good to hear you feel that way. Now, imagine how much better it would have looked if we could get rid of some of that overhead!

 

[Chin Rey]

LL has no control over the skills of the manufacturers of the goods.

The extremely high overhead of fitted mesh is very much due to a bug - or rather a combination of three bugs - and that is all the Lab's responsibility. There's not much even the best content creator can do about it.

As eku said, LL could have upgraded the system avatar. A lot of users asked for this long before fitted mesh came along and somebody even submitted a fully working higher detailed avatar B2.0 patch. LL rejected it all, presumably because they felt a higher avatar resolution would cause lag issues. The inevitable result was of course that users came up with even laggier workarounds and that's what we're stuck with today.

When it comes to content performance in general, there are three things LL could and should have done. For a start their developers should have made sure they understood the practical implications of the software they were working on. They should have made a far more thorough analyzis of what was needed of the software and what it was likely to be used for. They should have supplied proper user documentation and, just as important, they should have led by example. Their own mesh builds - both the ones they made themselves and the ones purchased from others - should have been at the best standard to show people what is actually possible. Instead we got stuff like the oerly heavy premium gifts.

And we have Meauxle Bureaux. For those not familiar with it, it started as the scene for an old Medieval Fantasy Faire. LL bought the whole thing, added a few details and released it as "the home of the Linden Department of Public Works". It's a train wreck of incompetently made mesh by incompetent mesh makers (although to be fair, some of them have improved a bit since then), a catalog of noob mistakes and an insult to all considerate content creators in SL. This is how it is described in the destination guide:

Meauxle Bureaux is the home of the Linden Department of Public Works, a program focused on improvements related to the experience of living in and visiting the Linden Mainland. This intricate build was lovingly crafted by resident experts for all to enjoy, so come see the ultimate in shared creative spaces!

This total lack of understanding of what mesh is, what you can do with it and how it shouldn't be handled, is - as far as I can see - shared by all Lindens and Moles even today. Faced with such massive ignorance from The Gods Of Second Life a considerate content creator doesn't stand much chance.

---

Bu having said all that, there are a lot of could haves and should haves here. It's all sins of the past. At this stage, transforming SL into the universal virtual reality it once aspired to become would require some drastic changes, changes that would certainly do more harm than good. Nothing big is going to change and all things cosidered, that's for the best.

Second Life is what it is, take it or leave it. Many people do both.

 

[Sid]

But having said all that, there are a lot of could haves and should haves here. It's all sins of the past. At this stage, transforming SL into the universal virtual reality it once aspired to become would require some drastic changes, changes that would certainly do more harm than good. Nothing big is going to change and all things cosidered, that's for the best.

Second Life is what it is, take it or leave it. Many people do both.

Exactly that.
SL is the only income source for LL apart from maybe investors.
Sansar is only a money pit so far and personally I don't see it change in a milk cow like SL any time soon.
The Lab is in no position to implement drastic changes.
If the regulars get annoyed and start walking away en masse from SL, it is over and out for LL within months.

 

[Beebo Brink]

To me, lag is mainly about retention and concurrency rates. Whether we like it or not, most people are not willing and able to shell out money for extra hardware to run Second Life.

Mobile/tablet usage is growing and it's taking a big bite out of desktop computer sales, and that trend is accelerating. Users who buy a "better computer" are a shrinking market. This means that the pool of casual, impulse visitors to SL is also shrinking.

When it comes to bandwidth, it's not even always a question of money, not everybody have access to top notch high speed connections no matter how much they are willing to pay.

This has been a impasse for me with Sansar. It's close to unusable for me, especially with the massive downloads every time I visit.

 

[Khamon]

Who are some of the considerate competent mesh creators in Second Life? Where are some places I can shop inworld and know that the mesh they’re selling will render fairly efficiently? Or, at least, how can I examine items in a shop to determine their render requirements and what can I compare those numbers to?

 

[Knutz Scorpio]

... Or, at least, how can I examine items in a shop to determine their render requirements and what can I compare those numbers to?

You can switch your viewer to Wireframe mode and back again with ctrl-shift-R (in wireframe everything appears as open triangles instead of solid) and observe how dense the triangles are that make up the object. I've seen a stupid amount in shoelace eyelets that appear solid in wireframe till I zoomed in really close.

 

[Chin Rey]

Who are some of the considerate competent mesh creators in Second Life?

Aww, I'm going to get sooo flamed for this!

The three top notch mesh optimizers I know of, are Arton Rotaru, Yetius (Rubber Bunny) and ... well, me (OPQ / ChinRey Houses). I'm sure there are more - I keep discovering good makers all the time - but those are the ones I know of right now. None of us have much mass market items to offer though. Arton and Yetius have mainly focused on rather esoteric vehicles. My stores are mostly filled up with very simple ultra-low lag filler/background items. I have made more elaborate builds too but so far I've only bothered to list a few of them for sale and I can't see that's going to change soon.

A few good mesh makers/merchants in no particular order. I'm sure others here will add to the list:

• Hatris Panacek (Ex Machina)
• Pamela Galli (La Galleria)
• Chic Aeon
• InVerse
• Theresa Matfield (T-Spot)
• Reid Parkin (Mesh Plants)
• Eldowyn Inshan (United InshCon)
• Nadine Reverie (3D Trees)

Some that seem good but I don't really know them well enough to say for sure:

• MeganBlu (Careless)
• IsisKowalsky (Naive Design)
• Cain Maven (Maven Homes)
• Rya Nitely (TUFF Old Boats)
• Morphe Inc.

Uneven quality - ranging from poor to excellent:

• FANATIK
• Alex Bader (Skye)

I've also seen some good content from Two Moon Gardens too. It's a bit up and down. They don't make their own meshes (thankfully) so it all depends on what full perm items they can find on MP.

Charlotte Bartlett (Scarlett Creative) and Lilith Heart (Heart Trees Flowers Plants) were struggling for a long time, Charlotte with overtexturing, Lilith Heart with poor LoD, but they've both been working hard. The more recent builds I've seen by them have been really good and they both seem to be very keen on improving their technical skills so we should expect a lot from them in the future.

 

[Amity Slade]

Every time I upgrade my computer, most of my programs run better. Second Life never does.

Second Life is not like the top-notch computer games. The reward for the gamers who constantly spend money on their gaming machines is that they get to experience better, professional content content. The reward for upgrading a machine for Second Life is to somewhat cope with the inefficiencies of amateur content.

But I really have to laugh when I hear that the solution to the poorly optimized content is to buy a new computer. There are people who honestly believe that the way free market economies work is that consumers should expect to pay more money to have worse stuff.

 

[Dakota Tebaldi]

You can switch your viewer to Wireframe mode and back again with ctrl-shift-R (in wireframe everything appears as open triangles instead of solid) and observe how dense the triangles are that make up the object. I've seen a stupid amount in shoelace eyelets that appear solid in wireframe till I zoomed in really close.

The fact that mesh in SL can use normal maps means there's even less excuse for stuff like this.

 

[Risotto]

I upgraded my GPU recently. It lets me run Black Dragon... in my skybox. It's a slideshow anywhere else.