Jump to content
NevaehHeartstrings

Demise of High Fidelity, how that might impact Sansar

Recommended Posts

In sad news it would seem development is ending  with apps discontinued in January

Sansar has always been a more polished idea of what a next gen virtual world might be, I wonder of parts of High Fidelity could be integrated into Sansar now since its open source.  while its disappointing to see that project die. this could be a opportunity for linden lab to strengthen and re-position its own projects.

 

 

 

 

Edited by NevaehHeartstrings
  • Thanks 2
  • Haha 1

Share this post


Link to post
Share on other sites

Since Linden Lab has -- from the beginning -- only wanted to do their OWN THING (hence building the platform from the ground up) and since they are now almost totally focused on events -- AND since I suspect they are very different platforms (  @animats could likely comment on that --- I don't think it will make much difference. They really didn't even have the same audience base (aside from the VR component).    

But we'll see. 

 

I believe HiFi began work (at least publicly) about a year before Sansar (also publicly) :D.  But I don't keep up on all of that and others likely know better. 

 

 

  • Like 1
  • Thanks 1

Share this post


Link to post
Share on other sites
1 hour ago, Chic Aeon said:

 AND since I suspect they are very different platforms

 

 

I agree were managed with very different expectations.

Edited by NevaehHeartstrings

Share this post


Link to post
Share on other sites

High Fidelity shut down as a public virtual world back in May. Virtual Worlds Startup High Fidelity Lays Off 25% of Staff, Pivots to Enterprise Communication - Variety. This is the "pivot to enterprise" failing.

“If you had asked me when we started the company in 2014, I’d have said that by now there would be several million people using (VR headsets) daily, and we’d be competing with both big and small companies to provide the best platform—but I was wrong. Daily headset use is only in the tens of thousands, almost all for entertainment and media consumption, with very little in the way of general communication, work, or education." - Philip Rosedale.

The sad thing about High Fidelity, technically, was that it wasn't a "big world" system like SL. It was a "game level loader", like Sansar and SineSpace, where you can load an instance of a canned world and have some of your friends in. That niche, in VR mode, doesn't seem to interest many people. All of those systems have user counts in 2 digits. All that work on a dead-end concept.

(I tend to think that SL has the right concept - the big shared world. The tech just needs to be brought up to about 2015 game quality. This isn't impossible, just hard and expensive. Probably cheaper than Sansar was, though.)

  • Like 5
  • Thanks 1

Share this post


Link to post
Share on other sites
11 minutes ago, animats said:

I tend to think that SL has the right concept - the big shared world. The tech just needs to be brought up to about 2015 game quality. This isn't impossible, just hard and expensive. Probably cheaper than Sansar was, though.

I have heard something like that in the past, does it have to do with being able to use modern video graphics cards?

Share this post


Link to post
Share on other sites
3 minutes ago, NevaehHeartstrings said:

I have heard something like that in the past, does it have to do with being able to use modern video graphics cards?

No, that's not it. Long story, probably belongs in a different topic.

  • Thanks 1
  • Confused 1

Share this post


Link to post
Share on other sites
17 minutes ago, animats said:

The sad thing about High Fidelity, technically, was [...] niche [...]

All of those systems have user counts in 2 digits. All that work on a dead-end concept.

How much it must suck to have worked in startup mode for years developing technology and now nobody can value it enough to make it worth integrating.

How cool would it be if somebody could graft thin wedges of High Fidelity onto OpenSim and make the next virtual world? Y'know, just for fun.

Must be some High Fidelity (even Sansar) graduates with money from earlier ventures and some time on their hands.

Wouldn't ya think?

  • Like 2
  • Haha 1

Share this post


Link to post
Share on other sites
6 hours ago, Qie Niangao said:

How cool would it be if somebody could graft thin wedges of High Fidelity onto OpenSim and make the next virtual world? Y'know, just for fun.

it could be possible that somebody will have a go seeing as how the source code is released Apache 2.0. There could be some upside for a developer to make a single canned world from it

probably the key would be methods to upload and surface SL-compatible assets. Which would I think mean the SL avatar also, in addition to the HiFi avatar 

 

  • Like 1
  • Thanks 1

Share this post


Link to post
Share on other sites
7 hours ago, Qie Niangao said:

How cool would it be if somebody could graft thin wedges of High Fidelity onto OpenSim and make the next virtual world? Y'know, just for fun.

Must be some High Fidelity (even Sansar) graduates with money from earlier ventures and some time on their hands.

Wouldn't ya think?

Can something on the asset-prepping side be taken from Sansar or High Fidelity and used for SL? SL just takes what the creator puts in as mesh, does some checks, makes an attempt to create lower level of detail models, and pushes it out to an asset server to be downloaded by the viewers. In a game development shop, there's an intermediate step, where, with junior artists and automated tools, meshes and textures are cleaned up and simplified, and impostors are generated. Sansar has some of that. I'm not sure about High Fidelity.

  • Like 1
  • Thanks 1

Share this post


Link to post
Share on other sites
5 hours ago, animats said:

In a game development shop, there's an intermediate step, where, with junior artists and automated tools, meshes and textures are cleaned up and simplified, and impostors are generated

That's not exactly entirely true, animats. First, impostoring is not that widespreadly used anymore. The few situations where this technique is being used, it's done manually. It is a so specific need related technique that it requires manual intervention in order to level the impostor to the exact location and possible angles of view to not break the illusion of a solid model. The example of windows does not cut it, as very rarely a indoor scene is located on the main level and, in the few cases that this happens, it's usually a shack used for loot or hideout purposes and a simple plane with the texture is sufficient enough to avoid making LoDs for it. We are talking about a different context than the average SL region, for which a building external door that teleports you into a skybox to see the indoor space is frowned upon. 

Second, the automated tools generate LoDs in a similar fashion as the decimation tools in 3d apps do, the percentage of reduction is just less prohibitive than what the SL uploader claims for a low land impact. Plus, the distance thresholds for each LoD in a Lod group can be set manually if needed, with the exception of Unreal Engine which bases that on the screen space of an object. Still, even UE LoD generator is decimation based, give it a try. 

As for what textures go, that is mip mapping, and that's a feature that is embedded already in most engines natively, including the substance engine, because it's a quite old and by now consolidated working model and there's no need to purposely run anything on textures to achieve that.

What I agree with, though, is the need for a better upload policy, including a decimator that doesn't suck so badly and, since impostors are needed, that would allow the low and lowest lod to accept a different model with an additional material even if not included in the original high Lod. This way one can create a bunch of possible impostor shapes with predefined UV mapping that can be mapped to a texture of our choice at a later time, saving a lot of time from avoiding the need to create a dedicated impostor for each model, put a hidden reference in the high lod and have its texture already mapped somewhere. 

  • Like 2

Share this post


Link to post
Share on other sites

HI Fidelity's social aspect died because the early group of users there were toxic, not only to new users, but even to each other. Hi Fidelity got tired of dealing with this drama, especially when it spilled over into their general area, with a 'Money Tree' at the center, and inevitably after that was removed, moved the drama and bickering and fighting to the Help Desk area, driving staff there crazy.


Of course, with the social aspect removed, and creators pretty much hamstringed, along with the technical and poorly documented methods of even getting any content to work there, it wouldn't be long before it would die.

The irony is, that same group who used to bash Sansar and SL, moved to Sansar because they had no other choice, and brought that crap attitude along with them.

Sansar is headed the same way. Has a similar toxic early community who wants to defend its power structure and keep the platform small, except with the added detriment of Staff actually supporting the toxicity - which includes vetting, ostracizing and favouritism.

Meanwhile VRChat is soaring, because it allows users to be users, and there are plenty of creative opportunities there.

There is a place to integrate Corporate events and such, but the lifeblood, much like it was in Second Life, will be what imaginative creators bring into existence.

Edited by entity0x
  • Like 2

Share this post


Link to post
Share on other sites
7 hours ago, OptimoMaximo said:

First, impostoring is not that widespreadly used anymore. The few situations where this technique is being used, it's done manually. It is a so specific need related technique that it requires manual intervention in order to level the impostor to the exact location and possible angles of view to not break the illusion of a solid model....

On impostors: for those who don't know what we're talking about, here's an explanation.

For a few months, I had an "impostor garden" in SL, where I'd created impostors for a half dozen objects and set them beside the real object for comparison.

impostorgarden1.thumb.png.6da181cdef76c7cd8c7d112622a63339.png

Impostors vs low LOD-models. For each pair, the one on the right is the impostor. The impostors are a bit blurrier but have no pieces missing. This is closer to an impostor than you'd usually get; it's for background objects off in the distance.

Many creators and developers came to see this. The impostors were simple 8-plane images, with an LSL script to show the one facing the nearest avatar. Beyond 25 meters, the impostor usually looked better than the lower-LOD model. The impostors were made with a simple rig in world that put the object on a turntable and took pictures against a green screen.

glow_005.thumb.jpg.937615a58f36bd962ad0afef73560202.jpg

Impostor making in world. The table is sitting on an invisible turntable with a green screen behind it and a red frame in front of it. The avatar making the impostor images sits in a chair attached to the rig with their viewpoint locked like this. The turntable is started, and rotates the object in 45 degree increments. Each time the rig beeps, the user takes a picture like this one.

 

glow_005-clipped.png.afc542e61caa275f0ff1b71912229473.png

Final impostor image. A program clipped to the red frame, removed the green screen, and resized to SL texture dimensions. Then all 8 views were put into one image so that texture offset could switch to the proper texture. All automatic.

This LSL demo worked only for one avatar at a time, because it was done in-world without viewer mods. In a real system, you'd do this in the viewer, so each user sees the one of the 8 images appropriate to their viewpoint. This was a proof of concept to show that impostors in SL looked better than bad LOD models.

To do this for real would take viewer mods. You'd have an option in the uploader, in the LOD menu, for "use impostor", and you could select, say, 8 or 16 planes, and whether you needed top and bottom views too. The uploader would then create images like the above (it can render the object, so it can do that)  and upload the low level of detail marked as an impostor. When the viewer displayed an impostor, it would rotate the impostor plane to face the viewpoint and apply the texture for that view direction. This is something that could be retrofitted to SL. It doesn't require a new design.

The impostor illusion breaks down in some situations. Mostly when impostors overlap. A distant avatar in an impostor chair or vehicle is the worst case. Probably need to turn off impostoring for objects being sat on or with an avatar inside them. If you get close to an impostor, the illusion breaks down, of course, but before that point, the viewer would switch to a higher LOD model.

This isn't magic. But it's better than what we have now.

Impostors are cheap to render. Much less draw overhead at distance, leading to higher frame rates. That's what this is all about; getting frames per second up to acceptable levels. Gamers think 60 FPS is slow. SL has trouble making it to 20 at times. That needs to be fixed to retain new users with game-level expectations. 30 FPS (TV scan rate) is probably the minimum acceptable.

So, that's what impostors are all about. I don't think High Fidelity did much in that direction. Every High Fidelity world I ever tried was so simple it didn't need them. SineSpace, being based on Unity, does have impostors. Does Sansar?

 

 

  • Like 1
  • Thanks 3

Share this post


Link to post
Share on other sites

I was in Hi Fidelity VERY briefly when they first opened to the masses. I am pretty sure that was still an  early alpha time. A lot of SL and Opensim creators went over. We ran into each other as most kept their avatar names.   It only took a few days to discover HOW insecure the platform was. There were some posts on the forums in the "this can't really be how it is -- is it?" vein and pretty much everyone left in mass --- even the Opensim folks and we KNOW how insecure that is LOL.

 

I disagree (a bunch) about the old guard  "Has a similar toxic early community who wants to defend its power structure and keep the platform small".

When I went over at the beginning of the open beta there were still plenty of alpha guys around and they did their best to answer questions (well most anyway).  It was very much -- back then --- a boy's club.  There were only a couple of gals (gal avatars anyway) that were posting on Discord.  THAT was a little tricky and of course  "I" had my point of view which differed from the norm ^^ and also of course I did not keep quiet about it.  Eventually most of us learned to get along and appreciate that we were all trying to support the platform.  

 

Now, most of the old-timers are gone and some are just absent from Discord. Some are still working it seems as we see their final projects now and then.  There are some nasty people on Discord. There are some nasty people on our SL forums (most I have blocked by now LOL). This is oh so typical of any  on-line chat system.  But  --- plenty of the old-timers still help the newcomers.   What those of us that have been around there for awhile DO not like is they way the rules change and regulations get "lowered". Some of these changes are very much in line with hurting the people that have been around longer. That really isn't fair.   The people that were pre-alpha have definitely paid their dues. They don't need slaps in the face.  That is hurting the platform and is easy to see by watching the Discord channel over time.

 

A whole group of people recently helped one gal get her rigging working and in the end Medhue fix the last bit of problematic stuff. Those legwarmers were indeed a "community" project. You really don't see that much here lately on the forums. It USED to be more like that here. 

  • Like 2

Share this post


Link to post
Share on other sites

Yes there were  a few very helpful people there, as there is on any platform. I apologize for not including that disclaimer that NOT ALL were toxic.

  • Like 2

Share this post


Link to post
Share on other sites
On 12/13/2019 at 4:47 PM, animats said:

The tech just needs to be brought up to about 2015 game quality. This isn't impossible, just hard and expensive.

This comes up so often that it's become a SL trope almost as old as SL itself.

Games are single seat 'sports cars', SL is a dump truck.

Games and consumer graphics cards are designed with each other in mind.

SL happens to be able to use hardware designed for games, just like F1 cars and dump trucks both drive on roads.

Games also have stuff made by professional artists to strict guidelines specifically for engines built entirely around performance. The SL dump truck carries all our hobby made rando junk where the only guidelines are 'can I sell this for a buck' and 'does it look pretty'.

This is not "hard and expensive", it's not possible because that's not how any of this works. Why do you think no one has managed to make a better SL in all these years? Why has no one managed to take the published source code and port the login and asset handling to a game engine (of which there are several os licence compatible ones).

Stuff like adding impostors is not bringing anything up to game quality, it's not seeing the forest for the trees. Play some triple A games released in the last 12 months (no, really). What's different, visually (as this is what most people think when they talk about updating SL), it's certainly not the resolution of the models or textures (or the various optimization tricks), it's the lighting. It's all lighting.

SL's lighting model is stuck in the past and not only are the majority of modern lighting methods unavailable (and can't be made available, because dump truck), most folk with builds/homes don't ever rez a single light source to leverage what little we do have. Maybe we can coax a vulkan render pipeline out of LL, maybe we can get a little RTX, maybe then people will rez some lights ... or not, and SL won't look any different.

A new render pipeline that worked with SL would be hard and expensive, but it's all for nothing if we don't use it (ignoring how we wont ever have the depth of control a developer with a game engine would have, because griefers), and with the best will in the world it probably wont be faster than what we have now .. because dump truck, remember?

If you want to make SL run like a game you have to start removing things, the very things that make SL unique .... the end result is Sansar and it's pretty obvious to everyone here how well that's going even if it does have a fancy lighting engine .. maybe Sansar's just not trashy enough for us.

 

On 12/14/2019 at 12:46 PM, entity0x said:

HI Fidelity's social aspect died because the early group of users there were toxic......

Sansar is headed the same way. Has a similar toxic early community ..... vetting, ostracizing and favouritism.

When a platform fails to get decent traction there isn't the opportunity to dilute toxicity that inevitably forms around small cliques, and the devs have little choice but to look after what little user base they do have. Early SL was just as bad and had to go though a couple of solid growth spurts to even things out.

On 12/14/2019 at 12:46 PM, entity0x said:

Meanwhile VRChat is soaring, because it allows users to be users, and there are plenty of creative opportunities there.

I wouldn't say 'soaring' .. it's managing about 10K concurrency, enough to give everyone who uses enough space.

On 12/14/2019 at 12:46 PM, entity0x said:

There is a place to integrate Corporate events and such, but the lifeblood, much like it was in Second Life, will be what imaginative creators bring into existence.

That's a myth a whole host of dead virtual worlds offering creative freedom can attest to. It's not about attracting creators, they will go where ever they find an appealing opportunity (they will go even without that). If anything it's mechanics and structure. What can you do, how you do it, and how you get directed to it. Leaving all this up to 'creators' is not good enough, that places an unrealistic expectation that always fails to deliver.

WoW is a virtual world, has been immensely popular for a very long time (still has millions of concurrent users), and offers almost nothing in situ for creators. The amount of time spent building that experience and world must run into hundreds of years in man hours, we're doing exceptionally well if a lone 'creator' in a virtual world is able to string together a simple cohesive narrative. 

SL's real growth was centered about making money. People came to exploit an opportunity (or heard about it though that hype), some of them were creative. This is why LL still lead with 'make money' as a primary feature.

Sansar tried to lead with making experiences (and money), and aside from actual events (which they tried to pivot to), there is not a single experience on the entire platform that can hold more than 10 minutes of anyone's attention (and that's all anyone could ever realistically expect). In that respect SL is the same, nice place you have ... but what happens here? Nothing .. oh.

 

  • Like 3

Share this post


Link to post
Share on other sites
58 minutes ago, CoffeeDujour said:

SL's lighting model is stuck in the past and not only are the majority of modern lighting methods unavailable (and can't be made available, because dump truck), most folk with builds/homes don't ever rez a single light source to leverage what little we do have. Maybe we can coax a vulkan render pipeline out of LL, maybe we can get a little RTX, maybe then people will rez some lights ... or not, and SL won't look any different.

A new render pipeline that worked with SL would be hard and expensive, but it's all for nothing if we don't use it (ignoring how we wont ever have the depth of control a developer with a game engine would have, because griefers), and with the best will in the world it probably wont be faster than what we have now .. because dump truck, remember?

[Emphasis mine] I know nothing about games, nor about rendering, but I've certainly noticed this effect in SL: almost nobody uses the lighting effects we have, even for the obvious purposes for which they were designed. And I'm not sure why.

I can't stand SL without full lighting and shadows, such as they are, but honestly that's mostly to be able to see all the work I put into lighting some of my own locations. The vast, vast majority of builds I see have little or no lighting at all. Why?

Part of it may be creators (I guess mostly pre-fab buildings) eschew anything that relies on ALM because not everybody has it enabled. Is that the whole problem? Is it all down to the (mostly superstitious) belief that ALM reduces viewer FPS? Is that really it? If so, can we "nudge" folks away from non-ALM settings by making them even slower, or uglier, or... I dunno, tax them somehow? (I can't imagine a way to incentivize well-lit content with some Land Impact bonus, but ARCTAN would be good timing if there were.)

Edited by Qie Niangao
  • Like 1

Share this post


Link to post
Share on other sites
3 hours ago, Qie Niangao said:

[Part of it may be creators (I guess mostly pre-fab buildings) eschew anything that relies on ALM because not everybody has it enabled. Is that the whole problem? Is it all down to the (mostly superstitious) belief that ALM reduces viewer FPS? Is that really it? If so, can we "nudge" folks away from non-ALM settings by making them even slower, or uglier, or... I dunno, tax them somehow? (I can't imagine a way to incentivize well-lit content with some Land Impact bonus, but ARCTAN would be good timing if there were.)

ALM absolutely will tank FPS on older video cards. Often to little or no visible benefit. 

  • Like 1

Share this post


Link to post
Share on other sites
1 hour ago, Lyssa Greymoon said:

ALM absolutely will tank FPS on older video cards. Often to little or no visible benefit. 

I have one friend inworld who uses about eight year old laptops to create their environment. They do not use ALM at all, citing Second Life is so laggy when they turn it on. Because they do not use ALM, they do not see how their Lara mesh body they decided to use also has full gloss on their skin making them look like a porcelain doll. They do not see that when the region's default windlight goes to midnight, their build becomes painful to walk around in (mostly because they believe that fullbright textures = lighting). I have even offered to set up some basic lighting so ALM users are not totally blinded by all of the fullbright, but they decline as they want the LI to be used for "useful" things.

Some do not see lighting as being all that important, I guess, and if they can move around without ALM, then so should others.

[sighs]

  • Like 1

Share this post


Link to post
Share on other sites
3 hours ago, Lyssa Greymoon said:

ALM absolutely will tank FPS on older video cards. Often to little or no visible benefit. 

The "little or no visible benefit" is what I was complaining about: there's so little content that takes even rudimentary advantage of ALM that, unless one does it oneself, there's not much to look at. And that would be just fine as long as we could never again hear that SL needs to update its technology because reasons.

As for lower FPS, those must be really older cards. Mine is vintage 2012 (a GTX 660) and I don't notice any framerate difference at all, with or without ALM. Admittedly full shadows are a different matter, but I can't live without those anymore either.

  • Like 2

Share this post


Link to post
Share on other sites
8 hours ago, Qie Niangao said:

[Emphasis mine] I know nothing about games, nor about rendering, but I've certainly noticed this effect in SL: almost nobody uses the lighting effects we have, even for the obvious purposes for which they were designed. And I'm not sure why.

I can't stand SL without full lighting and shadows, such as they are, but honestly that's mostly to be able to see all the work I put into lighting some of my own locations. The vast, vast majority of builds I see have little or no lighting at all. Why?

It's not hard to get lighting right. Here are some spots at midnight.

 

lightinganimats1.thumb.jpg.471f9d52b832a2b43bed93ad210ad00c.jpg

My place in Vallone, at midnight. Light fixtures where they would be needed in real life.

 

lightinghub.thumb.jpg.9b0df926c6c93cabfa570158349b9541.jpg

Waiting for the truck. Dock at GTFO hub, midnight, with the lights on. This is a well lit, simple place. The lights are what this place would have in real life.

 

lightingcorsica.thumb.jpg.82f54972d5c6778bc9b345f91ff12eb9.jpg

Linden road in Corsica, part of Circuit de Corse. I've always liked this spot. Much of Circuit de Corse is well-lit, and pretty to drive at night.

 

lightingfullbrighttrees.thumb.jpg.6685e159b686c3c4b3ffe0a4c6c6e5bd.jpg

Full-bright trees. Good trees, but for some strange reason, full bright. Don't do that.

 

lightingbigcube.thumb.jpg.9eca88b8a16850e3b6d2cab084320723.jpg

Don't do this. Please.

 

  • Like 2

Share this post


Link to post
Share on other sites
6 hours ago, Lyssa Greymoon said:

ALM absolutely will tank FPS on older video cards. Often to little or no visible benefit. 

The whole 'older cards' thing is a bit of a misnomer,  more accurately it's ALM tanks performance for people who don't have graphics cards (and no I wouldn't count any discrete GPU from nvidia that ended in 10 as a graphics card either).

It's not even older hardware in general (although there's a dedicated few who insist on spending hours in  SL with machines that can't YouTube), it's budget cheap laptops.

3 hours ago, Qie Niangao said:

as long as we could never again hear that SL needs to update its technology because reasons.

We can but dream ... just don't ever ask for specifics.

42 minutes ago, animats said:

The lights are what this place would have in real life.

Don't do this (and don't light a scene at night unless that location's primary setting and use is actually night time). You will burn all your available lights creating a handful of point sources that look right out of 2004.

You have a very limited brush to paint with, use overlapping projectors and broad strokes. Create a projector with a blank white texture, align your camera with the light (so you see the projection as rectangular, open ms paint, plain white image and roughly paint black where the light from your source is excessive, no need for detail (if this takes 2 minutes, you're overdoing it). Swap in your image and then adjust the lights RGB. Pure white light is almost never seen IRL.

Fill and soft ambient lighting that cast shadows adds far more to a scene, if you must have a hot spot (such as an actual lamp) can more often get away with textures. Don't be afraid to use primary almost surreal hues.

(if your stuff is mod) Add a tint to everything to create a more uniform brightness, at the very least drop the default white down 20% as it allows the lights to  show more tonal depth. Tinting objects off white can go a long way towards colour correcting aligning dozens of different things made by different people.

As this is SL and abusing alpha is standard fare, make a circular radial fade texture (white to transparent), place and tint to pull specific areas up or down, such as a soft darkness under objects or in corners. (yes this is technically bad for performance, so don't go crazy and don't stack alphas for a 'fog' effect).

 

 

Edited by CoffeeDujour
  • Like 2

Share this post


Link to post
Share on other sites
10 hours ago, CoffeeDujour said:

This comes up so often that it's become a SL trope almost as old as SL itself.

....

If you want to make SL run like a game you have to start removing things, the very things that make SL unique .... the end result is Sansar and it's pretty obvious to everyone here how well that's going even if it does have a fancy lighting engine .. maybe Sansar's just not trashy enough for us.

This is a common remark in SL. But, having dug into some of the machinery, here's a serious answer.

SL is slow for a number of rather specific reasons, most of which are fixable performance bugs.

  • Viewer side, most of the issues involve level of detail problems. As a rule of thumb, you can put about a million triangles on screen with a midrange computer today. Which ones to show is up to the viewer. It could be better at this. That's part of what Project Arctan is about. For this to work really well, we need content with better lower levels of detail, LI charging which encourages that, better tools for creating lower levels of detail, and, probably, impostors, which I've discussed before.
  • Texture loading was designed back when textures were loaded from the sim servers. Its speed limits are intended to protect them from overload. Textures now come from AWS/Akamai, which SL is not going to overload. And users have much more network bandwidth than a decade ago. So texture loading should be more aggressive. Also, GPUs now come with much more VRAM, computers come with bigger disks, and both cache space and VRAM use can be expanded. Fixing this is hard only because the code in the viewer for that is touchy, badly documented, and a mess. (I fixed a bug in there once, and had to puzzle out the code. Do not want to do that again.) Basically, someone has to sort out the policy part from the mechanism part, collect the policy code into one place, and then work on the policy for which textures load in what order and at what resolution.
  • Server side, there seem to be two main bottlenecks: scripts, and avatars entering and leaving sims. The first is partly a performance bug - idle scripts use about 2 to 3 microseconds per frame, each, and with 5000 script in a sim, most of the script time is gone. Fix that and script overload will mostly go away. The script system has good defenses against single scripts using too much time. It is undefended against too many idle scripts.
  • Avatars entering and leaving sims cause a huge overhead transient as state is copied over. This is a major architectural problem with SL. In the early days, the sim processes did everything. Gradually, some functions, mostly asset serving, were moved out. But there's still a lot of stuff done per-sim that could be done elsewhere. High Fidelity had more different types of servers. A system where the viewer talked to a session server, which managed per-user inventory and such, would reduce the impact of region crossings and teleports. Inventory belongs to the user, not the sim; the sim only needs to know about what's being worn. That's a big fix. An easier approach is more inter-sim bandwidth. On AWS, you can order 10Gb/s network links within the data center, which might help.
  • SL is underprovisioned. There are not enough CPUs behind the sims. This becomes clear when moving a sim to a different server improves performance. Oz Linden insists that more sims are not being put on fewer servers, but it's clear that overload on a server can impact a sim that is not, by itself, overloaded. It's not clear that a move to AWS will help this. Right now, the server farm is a fixed cost. On AWS, it varies with usage, and there will be financial pressure to keep the AWS bill down. Many companies that converted to "the cloud" have found their bills much higher than expected. 
  • SL's development staff is sized to be able to do about two major projects a year, the size of animesh or EEP, plus maintain the system. That's not enough.

Overall, SL isn't a bad system. It just needs more resources devoted to improving it. Probably less than the resources that went into Sansar.

  • Like 3

Share this post


Link to post
Share on other sites
3 hours ago, Qie Niangao said:

The "little or no visible benefit" is what I was complaining about: there's so little content that takes even rudimentary advantage of ALM that, unless one does it oneself, there's not much to look at. And that would be just fine as long as we could never again hear that SL needs to update its technology because reasons.

As for lower FPS, those must be really older cards. Mine is vintage 2012 (a GTX 660) and I don't notice any framerate difference at all, with or without ALM. Admittedly full shadows are a different matter, but I can't live without those anymore either.

A Quadro K600 is no older than your GTX 660, and ALM will wreck it. It's not a great gaming card, but it is capable of running SL, and I'd be willing to bet it's probably above average compared to GPUs running SL, way above the average taking laptops into consideration. Another issue is the Linden Lab Museum of Obsolete Computer Technology aka, recommended system requirements. None of the cards on the list are up to running ALM. They really need to stop recommending cards that aren't anywhere close to being able to run ALM, and really really need to stop turning shadows on by default with them.

Edited by Lyssa Greymoon
  • Like 2
  • Thanks 1

Share this post


Link to post
Share on other sites
12 minutes ago, Lyssa Greymoon said:

A Quadro K600 is no older than your GTX 660, and ALM will wreck it. It's not a great gaming card, but it is capable of running SL, and I'd be willing to bet it's probably above average compared to GPUs running SL, way above the average taking laptops into consideration. Another issue is the Linden Lab Museum of Obsolete Computer Technology aka, recommended system requirements. None of the cards on the list are up to running ALM. They really need to stop recommending cards that aren't anywhere close to being able to run ALM, and really really need to stop turning shadows on by default with them.

Valve has statistics on their users hardware. Yes, this shows gamer PCs. That's not a small community; 14 million users are logged into Steam right now. SL might shoot for a recommended hardware level that covers the top 80% of Steam users, and a minimum hardware level at the 95% point.  That would be below the level Sansar or High Fidelity required.

  • Like 1

Share this post


Link to post
Share on other sites
2 minutes ago, animats said:

This is a common remark in SL. But, having dug into some of the machinery, here's a serious answer.

SL is slow for a number of rather specific reasons, most of which are fixable performance bugs.

  • Viewer side, most of the issues involve level of detail problems. As a rule of thumb, you can put about a million triangles on screen with a midrange computer today. Which ones to show is up to the viewer. It could be better at this. That's part of what Project Arctan is about. For this to work really well, we need content with better lower levels of detail, LI charging which encourages that, better tools for creating lower levels of detail, and, probably, impostors, which I've discussed before.

No. That is just wrong. SL is slow because it is CPU bound. This is why disabling ALM is faster if your GPU is overwhelmed by object detail .. and why that only really impacts speed on less capable GPU's.

Obscene object detail is a problem, but this is not a background problem. It's foreground objects that will (and should) be rendered full detail. Impostor objects would not dramatically help as most of the render weight is right in your face .. probably attached to your own avatar!

But that's not why SL is CPU bound. Turns out asset processing is CRAZY expensive. Downloading stuff is not free. Decoding stuff is not free. File IO is not free. Do all of these things in excess and SL is tanked before it's rendered anything (and guess what SL does .. by design!). This is why SL chugs after you TP to complex environment or event, and seems to recover a little if you're patient.

2 minutes ago, animats said:
  • Texture loading was designed back when textures were loaded from the sim servers. Its speed limits are intended to protect them from overload. Textures now come from AWS/Akamai, which SL is not going to overload. And users have much more network bandwidth than a decade ago. So texture loading should be more aggressive. Also, GPUs now come with much more VRAM, computers come with bigger disks, and both cache space and VRAM use can be expanded. Fixing this is hard only because the code in the viewer for that is touchy, badly documented, and a mess. (I fixed a bug in there once, and had to puzzle out the code. Do not want to do that again.) Basically, someone has to sort out the policy part from the mechanism part, collect the policy code into one place, and then work on the policy for which textures load in what order and at what resolution.

HTTP textures in the viewer are downloaded just as fast as your CPU/Network allows. There is no left over legacy throttling. The download speed might not be the absolute max your connection allows because your CPU is busy doing a lot of this and that comes with significant overhead ... just like copying 1gb of tiny files is way slower than a 1gb single file.

Texture resolution loading is fine. There is a bottleneck with downgrading texture resolution as needed connected to how the cache works. Right now all the extra file IO is too expensive on an already bogged down CPU.

Loading and manipulating huge numbers of individual files is not free. It's arguably the main thing the viewer spends all it's time doing and is fundamental to how core 'SL unique features' work. There is no way to make this problem go away and the end result still be SL.

The max viewer VRAM that TPV's have unlocked is 2GB. It is rare to find a location in SL (with sane draw distance) that actually uses all of this. Unlocking more VRAM is not a panacea, it's probably not going to give any tangible benefits for the amount of work required to do it correctly. (TPV's unlock of 2GB is hacky, LL lock their viewer to 512 as there is no defacto right way to do more as integrated graphics often lie about how much vram they have!)

2 minutes ago, animats said:
  • Server side, there seem to be two main bottlenecks: scripts, and avatars entering and leaving sims. The first is partly a performance bug - idle scripts use about 2 to 3 microseconds per frame, each, and with 5000 script in a sim, most of the script time is gone. Fix that and script overload will mostly go away. The script system has good defenses against single scripts using too much time. It is undefended against too many idle scripts.

SL server side runs to a very different drum beat than it does client side. This is by design. Lots of scripts looks bad in the numbers, but if you're not looking at the numbers and trying to continually test microsecond performance then it's not perceptible and not related to the rate of server/client updates - which is very low by modern standards. This is how SL was designed to operate. It will never perform like a twitch based FPS, it can't.

2 minutes ago, animats said:
  • Avatars entering and leaving sims cause a huge overhead transient as state is copied over. This is a major architectural problem with SL. In the early days, the sim processes did everything. Gradually, some functions, mostly asset serving, were moved out. But there's still a lot of stuff done per-sim that could be done elsewhere. High Fidelity had more different types of servers. A system where the viewer talked to a session server, which managed per-user inventory and such, would reduce the impact of region crossings and teleports. Inventory belongs to the user, not the sim; the sim only needs to know about what's being worn. That's a big fix. An easier approach is more inter-sim bandwidth. On AWS, you can order 10Gb/s network links within the data center, which might help.

It's not the data transfer .. it's handling that data, unpacking & unserializing that data, inserting avatar state into the running region and then updating all the clients. 

The fix is trickery to make the impact less perceptible (even if that comes at the cost of numerical performance). The better fix would be to remove teleportation entirely .. but again, we're back to fundamental SL features.

Notice Sansar's architecture doesn't allow for border based region crossing, on purpose.

2 minutes ago, animats said:
  • SL is underprovisioned. There are not enough CPUs behind the sims. This becomes clear when moving a sim to a different server improves performance. Oz Linden insists that more sims are not being put on fewer servers, but it's clear that overload on a server can impact a sim that is not, by itself, overloaded. It's not clear that a move to AWS will help this. Right now, the server farm is a fixed cost. On AWS, it varies with usage, and there will be financial pressure to keep the AWS bill down. Many companies that converted to "the cloud" have found their bills much higher than expected. 
  • SL's development staff is sized to be able to do about two major projects a year, the size of animesh or EEP, plus maintain the system. That's not enough.

Overall, SL isn't a bad system. It just needs more resources devoted to improving it. Probably less than the resources that went into Sansar.

SL is a terrible system. There're dozens of nested interdependent reasons they don't make em like this, why no one makes anything even remotely like this, why when making a new virtual world from scratch even LL didn't make it like this !

Every single better designed virtual world has tanked, probably because the all the magic is in the janky interaction between all these systems.

  • Like 2
  • Thanks 1
  • Sad 1

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.


×
×
  • Create New...