Why is there a 1 gig cache limit?

i could easily allocate 10 gigs or more if it would make SL work better.

Why the 1 gig cache limit?

I cannot speak about SL specifically, but with caches generally there becomes a size at which hunting through the cache for the item becomes more expensive than getting the item directly.

Indeed, on my system, SL seems to be more responsive with 0.5G cache than 1.0G (although I haven't measured it).

Hmm i don't believe 10 gb won't make your SL even better.

2 GB.. yeah maybe.

But more then that wouldn't do much i'm afraid.

Hunting through the cache? Why not just use direct access? I wouldn't mind keeping all of SL in my HD if it meant less lag and expecially less bandwidth comsunption. Why, last December my ISP bill was over $125!
Which is one of the reasons why I try not to log in very often these days.

i've fiddled abit and found even at 1g often times theres a noticable 'pause' when flying to a common area such as welcome as the HD scans through the cache and loads in all the stuff from the last time you were there... i actually set it down to .5 normally jus because that pause is so much shorter then

I disagree with the idea that increasing the cache won't increase speeds. There are plenty of search algorithms that can be used to make searching very fast.

The bottleneck is definitely downloading, not hard drive. Hard drive access is in the milliseconds to seek, and crazy fast for continuous reads, in the order of megs per second.

Whereas internet is, at best, hundreds of kilobyters per second, which is 3 orders (1000x) slower.

I think the developers could take a look at other MMOs to see what they do. I came from Everquest, and they put everything on the HD, and that flew unless you had 100 people around you.

-Hiro Pendragon
Varney 200,200

forget 'seek' the real bottleneck is the system pausing while ya load in that 500 megs of crap.. even the best HD's take multiple seconds to load in that much data

Eggy,

Your ISP charges you for bandwidth? I would have thought it would be worth your while to search around a bit and find one which doesn't. I currently pay £23 a month for mine, and that is a standard price unrelated to the bandwidth I use.

Most ISPs have bandwidth caps
There is actually one ISP without bandwidth caps here, but rumor has it that it totally sucks, precisely because its uncapped and all the l33t w4r3z d00ds clog the pipes 24/7 with their pirated crap.
An ISP with unlimited bandwidth will either give you horrible download speeds, institute caps, or bomb. It's an unsustainable business model in most countries. Your ISP also pays for their bandwidth so it's only fair that they should pass on the costs to you.
Of course there are certain countries like sweden where bandwidth is subsidized by the government and so you can have a 10mbit line for $20 /month.

well they'd like you to believe they can't sustain it without chargin by the byte, jus like the phone companies would've liked ya ya believe they have to charge by the minute, even for locals that don really cost the company anything.

theres been more than enough proof that once people get tired of it and stop PAYING the companies that charge like that, the others make more'n enough money.

Indeed, but the EQ world is static and therefore the "cache" - the world on your hard drive - needs no updating or cache consistency mechanism. SL is anything but static; if I retexture my floor your cache entry needs to be invalidated and reloaded.

You may find Philip's article on The Evolution of 3D Streaming informative. Incidentally, I like your first name; alas it seems as if Hiro Protagonist was reserved.

eltee,

I see what you're saying now - the WHOLE cache is being loaded into memory. Okay, gotcha, certainly that's a time-busted.

But still, even with that extra obstacle, I think it'd be fairly simple to organize it. Why not just load the cache for the current sim, and maybe surrounding sims?

-Hiro Pendragon
Serving SL from Varney 200,200

well no the problem with the larger cache as it is currently is the bigger the cache gets, the more hits yer going to get upon entering a new area... the more hits you get all at once (since level of detail calculations generally let you get near then 'snap' a whole pile of things in at the same time) the more elevator action yer hard drive gets to play inside the cache file movin the heads back an forth tryin to load in the data needed for the current system.

basically pausing yer system (since SL is already taking the lions share of cpu/memory/pci bandwidth)

thats the problem with so much cache on such a stressful system environment.. this isn't google where you can have a machine basically idle until it get to fulltime search a relatively cpu intensive optimized database query... loading the cache is jus one small part of a VERY taxed client rendering system

the *only* real savings the cache gives you is in how much you end up downloading, as far as actual system performance, its actually a penalty because pulling too much data all at the same time is a huge strain for the already taxed motherboard pci bus. Streaming pieces of it slowly off the net actually is less taxing to the system, just more taxing to yer bandwidth pipe.. so its a balancing act.

in all the offline games basically you get the luxury of pausing the game, stopping all the pci traffic basically, givin ya lotsa breathin room to 'load' in an entire level then ya never hit the HD again till ya get to the next level... so large multi gigabyte pools of installed data, that works okay.

An active online streaming game like SL has no 'levels' so when you do hit data from the hd, especially large chunks of it, its enormously taxing and does cause significant 'choppyness' on most systems when moving into areas with alot of people and objects that you have cached textures on

{edit}

so basically assuming you have sufficient network bandwidth to stream in the data in a relatively orderly manner, cache ends up causing more system 'lag' in concentrated bursts of disk access hiccups than it eases with download considerations

** deleted **

It'd be faster if LL just let us all live in the same building as the servers.

yeah but would you want to wait a few minutes to read through and sort the cache data out every time you went to a new area?

At this point a cache larger than 1GB won't really help you. That's enough for a good "working set" of stuff you've seen and are likely to see again. We compress this stuff very heavily, so your typical 1GB cache will contain 10,000 textures, 1000 sounds, and half the objects in the entire world. But, no matter how big the cache is, you're still going to do alot of downloading because SL is in such constant flux.

As eltee correctly points out, having a local copy of data can actually hurt performance, but it does save the delay while you wait for stuff to download. Long-term, there are things we could do to take better advantage of bigger caches: right now, SL has to decompress everything when it loads from the cache, just like when it downloads from the net. This isn't so good for your framerate. With a large enough cache (we're generally looking at 8:1 compression so 1GB becomes 8GB) we could store stuff uncompressed... but that introduces the new problem that the actual load from the disk takes alot longer. So, as a prerequisite, we need to improve SL's asynchronous disk loads... and it all adds up to something we won't be doing for a while.

Hey Ian (or whoever),

Where exactly does SL store the cache? I went looking for it a couple times, and couldn't find it...

Another suggestion - why not give prim bonuses for keeping texture / sound / etc size lower?

totally unworkable as to the bonuses... aka someone comes along, drops a couple of 1024 textured cubes on yer land, suddenly yer 80 prims over yer limit n stuff starts gettin returned


the cache is in yer user ocuments and settings\application data\second life folder i believe

Err, eltee, if someone dropped anything on your land that put you over the limit, it would be THEIR stuff that would get returned. Objects belonging to other people or groups are always returned before any of yours are even considered.

it would depend on how the system worked... in its current state if that was implemented you'd be 50 of the limit after one prim drip, and theirs, and 49 of yers would be returned, then you'd have 49 more available.. or mabye 80, or mabye 20

it jus totally defeats ANY chance of actually *PLANNING* based on yer prim allotment... there fore it gets the thumbs down. its totally subjective anyway... you could make a prim with *no* textures that'd take 50x the system resources as a single cube with a high resolution texture on one face, how could you possible actually judge that?

you'd have to abandon prims altogether and go back to polygons, which is somethin LL doesn wanna do

Eltee, you seem more interested in shooting down my ideas than making suggestions to make them more feasable.

Why not just not count other people's (not group, obviously) textures when calculating the prim bonus? The whole point of my idea was to reward land owners for using low-size textures on their prims. And, to a degree, you're right, you can't plan PRECISELY the amount of prims you will get once you place your textures, but the idea is a small bonus, and I don't think that not being able to totally measure the bonus ahead of time is any detractment from implementing a bonus itself. The bonus prims could be listed in a seperate number, this way a person knows what his/her base prims are and won't go overprim.

Is that so "totally unworkable"? No, and even if you come up with another counter-example, I'm sure I could resolve that too, but the point remains - you originally said it was impossible, and then you said it'd require polygon rendering, which is not true.

Please, if you're going to reply to my suggestions, don't tell me something is impossible. At least give me the benefit that it "might not be possible" or "is not likely possible". Just because I have different ideas that you might not have thought about doesn't mean they're instantly wrong. And if you really must post a negative reply, please take extra time to think about your negative feedback, because it's clear from your responses that you haven't totally thought them out.

And as for your criticism of Eggy's idea... why wouldn't it FIRST remove the newer, foreign prims, like it currently does for the prim limit, and THEN check the limit with the texture bonus?

Once again, you seem to me to be more intent on disproving peoples' points than on working with peoples' ideas to make them better, honed, more feasible, and saying things like that our ideas will force us to go back to polygon rendering is just plain exagerration and not thinking through things before you post your negative reply.

heh sorry bout that... jus don think that theres any real way to work prim rewards into the current system...

sides.. tho i don't actually advocate the devil, i do take out mephistopheles' dry cleaning on alternate thursdays