How do we improve SL Performanc for the long term future

The best way to improve SL performance for the long term future, is to have LL hurry the hell up and release the server-side code to the open source community.

The faster that code gets in the hands of people as talented as Nicholaz, and the Electric Sheep Co., then the faster we'll see improved service. Look at what folks like that have done with the client code.

Imagine...your own sim...on your own server...with the ability to connect it to the main grid...or use it as your own off-grid resource. I drool over the day when I can back up my account inventory on my own hardware...on my own asset server...which could very conceivably be nothing more than a bargain-basement desktop.

Let LL move away from trying to tweak its own code, and simply move into the realm of providing grid connectivity.

The idea of charging per-pixel is interesting but it doesn't help with all the textures that are already out there - even if they implemented this today, it'd take a long time for it to make a difference.

Since bandwidth seems to be one of the big problems, I wonder if the sim and/or viewer could be smarter about looking at the size of the face that a texture is going to.. Maybe have the asset server scale down big textures into smaller versions and send the version that's appropriate for the size of the prim it's going to. I don't see LL doing this but like the idea anyway..

burn it all down and start over with software thats not based from 10 years ago

the entire sl experience, no matter how much they patch and hack at it is all based on already proven in 1999 software, which could range from 95 to 99 depending on its application, this is also why when new hardware comes out theres always a performance issue

p4's issues
dual cores issues
64 bit issues

whereas my old athalon xp stomps sl's butt, since the core hardware design of it was based in (guess ) 1999

a SL 2.0 is long overdue, and the longer they put it off the less efferently this system will become on newer hardware, both theirs and ours, thus giving us the users worse and worse performance from all angles

Sell out to Google?

A 512x512 texture contains 262,144 pixels, or 256 "1024 pixel" units, so should cost L$2560 by these calculations, whilst a 1024x1024 texture is four times as large, so presumably should cost four times as much. Unless for some reason we're summing the multiplicands (?!) to compute our "per-pixel units" to get only L$10 for the 512x512; but then, the 1024x1024 should cost L$20. So... this is some fancy arithmetic.

In fact, I suspect that there's a kind of superstitious fear of large textures that has built up around some "general guidelines" or something. I mean, again, a 1024^2 texture is just four times as large as a 512^2 (assuming no additional benefit in lossless compression, which is a bit pessimistic), and is much more efficient than four separate 512^2s. Of course, if one doesn't need the resolution, one shouldn't waste the pixels, and perhaps there are too many high-res textures in use, but it's not like there's never a reason for them.

Now, on the other hand, if the upload fees were based on bits per pixel, such that alpha-channel textures cost, oh, say, a million times more than opaque, there might be a real benefit to that.

The biggest advantage of SL over regular PC games is the browser style client. SL is not a game but am entirely new part of the internet - it is the 3d version of the world wide web. The point of the WWW is streaming content. SL does exactly the same. Streamed content wiht no local files makes up the core of what SL is. There is just a small browser software to install and you are ready to go wherever you are! You can surf SL on any PC within minutes. The need to download gigabytes of local files every time you want to enter SL would render all the project meaningless.

There is only one way to improve virtual worlds: Increase bandwidth standards and CPU power. Anything else cannot be called progress - it means walking backwards.

The biggest benefit of reducing texture sizes isn't to keep the size of them on disk down, or to get them to download faster, but to make better use of the limited amount of texture memory on a graphics card and increase FPS.

You can increase the bandwidth and the only gain you get is a faster rez, which won't really matter if you spend 2 hours in the same place. Using more efficient textures on the other hand will help download and could give an actual FPS improvement as well.

What I'm talking about is improving the efficiency of the existing engine, to not do so means we are shutting out most gaming or high FPS applications to become a building and shopping platform. We need to get the best of both worlds.

SL is not a game, and IMHO it is not here to compete with any game. SL is a pioneer project, or you can even call it experiment. It's about adding a 3d platform to existing internet applications (WWW, IM, social platforms, media content hosting, web phones and so on). If you want perfect graphics for gaming buy one of those games that come on 5 DVDs or get a PS2. I prefer the still limited world of SL because it is not a game but an entirely streamed, permament, and thus flexible and dynamic social environment that exists OUTSIDE of our PCs. I can live with its limitations because this is just the beginning of something bigger.

Better caching
The current cache for SL is pants, I can store a lot more than 1gb of data. And I don't how understand how it fills up so quickly, I know that I've not used 1gb of data on my home-area, as I mostly re-use textures, and never use a texture larger than 512 x 512. My home is in the sky and nowhere near any other builds.

Really we need much bigger caches, or several caches, one for each region we've visited recently, with a limit to how many are kept. It surely should not be hard to do either. Here's a proposal for that:
http://jira.secondlife.com/browse/VWR-435

Better multi-threading
The main thing that kills SL's performance is that it tries to do most of its work in a single thread! The result is that in order for SL to do things it has to wait for others (such as downloading some texture data) to finish. The result is a significant performance kick as multiple textures download at the same time, but have to fight with things like avatar movement, animation etc. in order to get a chance. The result is that texture downloads, occlusion culling, level-of-detail and other features can actually slow SL down and cause freezes and lock-ups, instead of improving performance in the background like they're supposed to.
Here's the multi-threading proposal:
http://jira.secondlife.com/browse/VWR-1135

Better use of video-memory
Currently SL seems to put whole textures at full-detail onto client graphics cards, the result is that 500mb of texture data may end fighting for a place on a 256mb graphics card. This means benefit of fast VRAM is lost since data has to be swapped in and out to regular RAM. And that's if all the texture data fits into the user's RAM!
One of the benefits of JPEG-2000 is that we can download texture progressively, first as a low-detail layer, with detail added on top of that, and again, and again till we have the full-resolution texture. Really what we need is for SL to only use the lowest detail layer it needs, with higher detail layers shunted to hard-drive and only passed in (or downloaded) if the user moves closer to the texture and thus requires more detail.
For the Texture-LOD proposal please see:
http://jira.secondlife.com/browse/VWR-1119

More proposals
For other proposals to consider, and to cast your vote for performance improvements in general (whether you vote for a specific solution or not) please refer to the performance meta-issue:
http://jira.secondlife.com/browse/VWR-864

Other than a good understanding of how networks, graphics & a multitude of computers deal with SL....

What about plain ol' efficient coding?

So .... essential SL would run quicker if there was no content and we were spoon fed content ala WoW. Hmmmmm Cant see LL going for that as Land would be worthless as you couldnt build, no upload charges etc. I think technically Haravikk has it pretty much spot on.

We are expecting so much from a system that is still the only one in its class or market demagraphic. Im not one for singing LL praises these days but clustered computing isnt exactly easy nor will i think open source will put the world to rights.

SL was ok while it was run on a smaller scale, not its got more users than it can "afford" to run. I know ive said this before but if you want to make SL run better ....... cut the dead wood..... trim back on waisted resources and scrap the Tao.

Focus on customer needs and service and less on basement coding.... we know SL is hard to do after all .......we are the beta testers

If they allowed people to give a URL to a texture rather than actually uploading textures to SL, it would save LL a huge amount of bandwidth and storage space, since every time that texture is viewed, our client would download it from the textures server rather than from LL's server.

Active worlds does that.

Maybe have the 10L$ or more price if you want LL to host it and free if you use a URL.

Text on a prim would also drastically reduce the need for custom textures as at the moment we have to use a 512 x 512 texture or larger just to display a few lines of simple text

Two words...

Open Source.

I actually dont think open source is the answer, there are too many variables with regards to security etc. ... the open sourcing of the client pretty much invented every bot known to man in SL.

i.e. Assett servers, how can they trace your assets , inventories, money etc Distributed computing works very well in a closed network... it works very badly in an open network. Not saying that open source wont make things better as it will allow other coders to fix things that LL just cant be arsed to fix BUT it will also allow a multitude of security problems. I for one would not connect to an open source server when my personal information and linden balance was open, the best you could hope for is that assett retention stays within LL and game rezzing is kept open source.... BUT seeing that every item in world is also on the assett server stating its UUID and XYZ that pretty much means having access to the assett server aswell. This may be partial to the reason why we havnt seen open source as yet.

I think our expectation of this little world goes beyond LL, we can hope that LL gets things together... i still have a LOT of hope even IF its bloody too expensive for the average person and a free system is taking the piss.

Marty

I dont know if the 3D web is going to truly become universal and LL wants to be the blue print then Open Sourcing the server would be a must.

The asset server will probably disappear in favor of everything living on the server and on people's own hard drives.

Your entire look would be saved on your hard drive, for example.

While the buildings and furniture in a SIM would be on that server's hard drive.

Collette, I don't think it's practical for av appearance (for example) to be stored locally. That data has to be where everyone can get it. I don't think you want to TP into a club with 50 people, and each of their clients then connects with your computer to retrieve the image data.

Frankly, I think the biggest hopes for better performance are:
* faster computers & video cards, with more memory
* faster networks
* less flexi hair

OK, just joking (partly) about the flexi hair.

Restraint on the part of content creators is called for, but hard to expect because there are just too many. It's a GREAT aspect of SL that nontechnically minded people are encouraged to create content. I'd rather live with the disadvantages of that, than raise the bar so that only highly tech-savvy folks can create content. (Even then, there's no guarantee they'd actually USE their savvy to make efficient content. From what I know of us tech types -- we tend to take the easy way.)

What I really WOULD like to see, and which might help a bit, is an easy way to make uploaded textures public domain, with a browser for the public domain textures built into the client. (Heck, I'd even like it if included pay-per-customer textures.)

In other words, when uploading a texture, there's a "share it!" check box, which is ticked by default. (And perhaps, a "cost" box, to be payed by anyone who takes a xfer/copy copy of the texture.)

The problem with this is sorting this huge collection of sharable textures. I suggest we use the public to help us with this, allowing anyone to post keywords to textures which are then associated with the photo, using cacheing techiques to use the most commonly associated keywords when displaying and ranking (as well as "use counts" for rankings).

BUT, that's a whole 'nuther kettle of fish. It would lead to a lot more texture sharing, which might help, but I'm not convinced that's the major hurdle.

you don't think that is how it is going to eventually be?

Keeping in mind the "3D web of the future" idea. There simply is not going to be a central asset server for EVERYBODY.

As soon as you teleport into a Sim, your computer will send all your personal appearance stuff to that Sims computer and then that computer will be where everyone else gets the information from.

Kind of like your "name" when going to an old school chat room. You show up , log in tell it your "name" then the computer running the chatroom "tells" everyone you are there.

Say good bye to the clothing and skin industry then.

I think that's sorta how it already works, except for prims your wearing..

Every year the speed of graphics cards and the amount of memory on them increases dramatically. SL is a bit ahead of its time in terms of the load it places on people's computers, but dumbing it down doesn't seem like a way to future-proof SL. Quite the opposite in fact. SL's biggest selling point is that it's all user created and a great platform for creativity. Hobble that and you've hobbled the whole idea behind the platform.

People who currently have computers that barely meet the minimum specs aren't going to have those computers for much longer because platform and game developers aren't going to cater to them. Even operating systems are leaving them behind. At some point people have to accept that if they want to play they have to invest in hardware capable of allowing them to do it with acceptable performance.

SL will never have the performance that a static 3d world that isn't streamed has, and that's just fine, because the fact that SL is completely dynamic and streaming is what makes it SL. If this was just "chat world" with limited ability for users to build, texture, and script, I sure wouldn't be here any more, nor do I think SL would have ever grown to what it is today.

The future of better SL performance depends on ISPs, graphics card makers, and LL finding ways to optimize viewer performance and asset delivery, not on limiting what people are allowed to create.

We don't. Its hopeless and its just going to get worse and worse.

Sooz

Note that asset servers are distributed. There's no need for a "central" one. We already do something like this with DNS, the world-wide database that keeps and distributes the hostname-to-IPaddress correspondences. Completely decentralized, and surprisingly robust (at least, when your service provider technicians don't goof things up).

In the future, your information won't have a specific location. Would you want it to be in your home, which could burn down? Or in some building, where the power could fail? OK OK, yeah, i know YOU do periodic backups, but enough people don't and won't and that's not how the future will be.

You don't keep your voice mail at home for your cell phone. Similar thing. We'll have information service providers, who will keep it for us better than we'd be able to do for yourself. And some of us (nerds) will choose to be our own info service providers, just as some of us (not me) host our own web pages from home.

When you TP into a sim, the sim gets the info from wherever it's available closest, which might be nearby, might be across the continent, or might be your home computer -- but my guess is usually not. For example, the textures in the clothing you're wearing -- need those be stored on the home computer of every person who has the same clothing?

And where should your inventory be?

And don't you want to be able to log into SL (or TL) when you're visiting your favorite 96-yr-old Auntie, who's gone to bed, and your computer is still at home, while she just got a fast new one, with 16GPix display glasses and complete VR Activity Suit in YOUR SIZE? (How do you think she stays so young anyway? She SAYS she only uses it for areobics ...)

On the other hand, whenever I predict how "it's going to be" ... I'm sure to be wrong about some of it!

Bingo, Chip. Well put.

Nope. Note that you can log into SL from any computer and be YOU. When it says "Still downloading clothing", that is from asset servers to your computer, not the other way around.