9 minutes in one spot ... textures STILL not loaded.

You must learn to embrace the grey you see in the world.

Well, all I can tell you is that unless you got your own T3 line over there you didn't download no 500Mb in anything less than an hour.

Good experiment. We should have a number of people try this, quantify the loading times, and put error bars on the data.

The only explanation I could think of, if that really did happen and nothing else was accessing that file when you cleared the cache is that maybe SL uses hard-coded sizes? Like all those files have to be that 500-something-MB size and it uses some kind of bit-stuffing to fill the file up with dead data until it can be filled with real textures...but, that is just a stupid thing to do.

Wouldn't this be basically the Interest List problem described here?

This makes sense to me. It may be likely that it's a placeholder to assure someone doesn't run out of that amount of actual cache space on their hard drive. Or some other technical convenience.

Well, that answers the texture problem then. Seriously...think about it, this Interest list is probably an dynamic array of objects, textures, sounds, etc. that you are seeing alot. So in order for this list to be somewhat accurate you have to use averages, lots of averages, which traslate into CPU clicks. Then you are also sorting that array by constantly, adding things, removing things as you walk around god forbid you drive a car or sail a boat.

I still like the idea of using P2P style texture/sound/etc. downloading...if you got five people around you rather than use this interest list crap I would have it download textures from people around me rather than from their servers so that bandwidth can be used for other stuff...like FPS

Yeah, I'm under the impression that they were pretty certain the interest list, not the cache, was almost totally responsible for texture loading problem. I posted (today) to the answers forum asking for an update on that, but my query has not yet been answered.

Hahaha! Good luck with that

I actually expect some kind of answer (unlike some, I've never been left hanging for an answer). But I also expect it to be a "still working on it" type answer that says very little. This sounds like a non-trivial problem to solve.

Oh no, that is a major thing. Because a component like that cannot be broken! If they are going to completely re-arrange the way textures are loaded and cached, and also accessed? I think LL just said that they are trying to get away from those "BIG" updates and start releasing small incremental updates.

They are having some serious growing pains, but it's silly to think they would design an online world that can only support about 5 grand worth of people before it becomes un-playable.

Hey LL! Want a programmer? heh

I totally missed when you said this, but I just wanted to award an attaboy for innovation here. a P2P solution would be great, but is it implementable at this point, or are we locked into something like an interest list?

Well, in detail I don't think it would be to hard to implement. I don't know if P2P is the exact way to go...thinking in terms of Limewire where you see a file you want and for example 10 other people have it you speed up your download pulling pieces from all of them rather than all of it from just one. There is a reason it puts them on the top of the list

I think SL would be more something along the lines of BitTorrent inspired. Let's say you are in a club or event with alot of people. The way it works right now everyone is there and pulling from SL servers, not complicated, not dynamic.

I would say it works like this. You goto the club and everything loads from SL like normal, but other people are there when you show up and you are downloading textures and the music loops from them and SL too, and they are downloading your AV from your IP and SL.

More and more people show up, and now you got a party going on! People dancing, trading, talking, changing clothes, models, etc. My question is why do all these peoples have to pull from ONE place when they all could be pulling from each other (here comes the bit torrent part).

After you are in an area for a certain amount of time, and knowing that SL uses a grid system of some kind that groups clusters of computers to power that square only, and nothing else and all the people on it. I say make it like a torrent! If you are in an area for more than say...10 minutes you "seed" yourself on those computers and become apart of a array that people can download textures from. This way when alot of people get together it won't stress the server, it would make things easier! The more people that showed up the less the real server would have to constantly be "sending" and could focus more on physics and FPS.

EDIT: All of this is assuming that those servers are load balanced, meaning if textures need to be sent MORE than physics need to be calculated imagine a pie chart shifting power to one over the other and back and fourth to level things off easier.

Sorry for post length!!

I don't know windows internals, but under Unixes it sometimes makes sense to preallocate file space as it doesn't have to search for new free space when needed.

geek note: yes, I know lseek past end of file doesn't actually allocate space but a tight write loop will

This jibes with my memory of things. People have been mocked for suggesting going back to a previous version to the point where I have been afraid to say that. Are you saying it wrong with 1.6 or that 1.6 the last one that worked ok? (Note that I am not suggesting reverting to a previous version)

The big problem with SL scalability has been the asset server which handles textures, inventory, and such. Somewhere around the time where the active population regularly exceeded 1000, the asset server began to become overloaded and LL added secondary caches to each sim to try to distribute the hits on the central server. This was a smart move especially considering the locality of data needed by agents (AVs) in a sim. Unfortunately, one-layer caches are hard to keep consistent and two-layer caches are more than twice so. As there is a third-level cache on your client, the problem is compunded.

The processing power is certainly there, the places where I expect that performance could be enhanced is throwing more iron at the central asset server (which I believe they continue to do). I expect that the most troubling faults are in handling the three-layer cache consistency which is a coding task that is tricky and error prone. As an example of how something so conceptually simple can be so difficult there is a very common sorting routine called quicksort which - as the link shows - takes very little code to write. However, even experienced coders will often make errors when asked to implement it because the devil is in the details. Three-tier caches are considerably harder to get right.

Peer-to-peer assets distribution is not what is needed here, reviewing the hell out of the cache code is. There also appear to be faults in recent changes to the interest list handling which is almost totally unrelated to caching.

Well I don't know much technological and you know that.

But I do know this much:

1. Earlier than 1.6 - or 1.6 and earlier, I forget which - the textures were okay.

2. With 1.7 all that was history, and we've never been able to see the world the same way since.

3. And we're supposed to be okay with that, and never expect them to be the way they were again.

I believe it was you, Flipper, who tried to explain to me at the time how this was all going to be for our own good, and how we had been "spoiled" by the way things used to be, and how eventually it was all going to be for some very wonderful cause. Or something like that - as I said, I don't understand technology very well, but something like this was what you were trying to explain to me at the time.

Well, I figure it isn't really for our own good. I'm still not understanding why it had to be this way.

But I figure - it doesn't matter if we never see things as well as we used to in 1.6. This just won't be a game known for people's ability to actually see it in any short amount of time. I

It would be good, though, if it would get back at least to the best it has been subsequent to 1.7, cause it seems like it has been at times better than it has been lately.

I'm actually pretty cool with the fact that overall, it's always going to be worse than it was in 1.6. Though still I don't understand why they would want to purposely make something worse than it was before.

coco

Well, if the interest list is determining what is being add/removed from the cache then it would be an important piece I figure. I agree with everything you said other than what I quoted there. You make very good point indeed sir!

I am thinking also along the lines of the future, and when you have so many users roaming around that some kind of sharing is going to have to happen sooner or later.

The synching issues you talk about are very real indeed, but that is why I spoke in terms of "seeding" and the benefits of doing so would only start to happen if you stayed in an area long enough for it to really mean something in terms of you serving textures and content to the people around you.

This interest list would still be an important part because I imagine that list, if people are together long enough (like a club, events, etc.) you could take all those lists and combine them to create a large "packet" that everyone in the area is "seeding" to eachother and that would help take the load off the servers which would be processing the physics of that badass gladiator battle that you bet 500 L's on!

Do you really think going this route is a bad idea? Sooner or later I would hope large events can be held, not this 20 - 30 crap, I mean real events like SL's own Nascar, or sports events, gladiator style matches (and betting on them too!).

EDIT: If you don't like the idea of pulling textures from random users, what if every person you added to your friends list would only be used to pull textures from? But this might be a really bad idea, because you would only share data when you where around the people on that list.

Swarming asset data peer-to-peer could be easily made safe by having the asset server sign assets with a asymmetric key cryptography and have the client only accept authenticated assets; 10 zillion SSL users have proven this feasible. However, I really don't think asset data egress from LL is the bottleneck.

There was an announcement from the development manager today (sorry, too tired to search and link) acknowlegding that they broke something in the interest list in 1.7.x and that they think they've unbroken it on the preview grid already. This is good news.

well it would seem to me that its time to raise the cache size. maybe make it so you can set the cache to anysize you want. i have the space and wouldn't mind a huge cache if it makes SL run smoother.

*Drones, zombie-like: Embrace the gray . . . embrace the gray . . . impeach Bush . . . embrace the gray . . .*

coco

Anybody find the quote?

When will the new preview version be available? What is at the preview download site is older than the current version and doesn't work at all . Gee, perhaps links to old non-functional preview clients should be removed?

Here.

They may be pre-allocating the entire cache file size to prevent disk fragmentation as it fills. It's probably just zeros with a lot of reserved space.