Megaprims and performance

I had a message earlier from someone complaining that their neighbour was using megaprims and it was causing performance issues. I checked the stats and didn't see anything bad but I was wondering what the performance issues are with megaprims and how many is too many?

I don't have an answer to your question, Ciaran, but I just read this today:-



Scroll down to the "Region Optimizations for Improved Viewer Performance" part. It doesn't answer your question either, but it's interesting that big prims are mentioned there.

Cheers Phil, this link works better

https://support.secondlife.com/ics/support/default.asp?deptID=4417&task=knowledge&questionID=4780

never had any problems with them yet..as far as i know it's if you have them hanging over sim borders si when the problem happens..
i've hear something that they are no different than a regular prim from many threads in this forum..
that the older prims had problems but with havok 4 thats all gone now..

i use them in builds all the time and everything is smooth..maybe it's the scripts in the prims that are the problem..but using less prims i would think would be less of a problem than using a bunch of small prims..

Cheers Ceka, I did notice that two different creators prims had been used so maybe one is more efficient than the other. I've purchased builds using megaprims and not noticed anything unusual.

i don't even think there is a problem with the older prims either..i believe Havok 4 fixed those issues..
it could just be scripts in the prims or coming from anyone in the sim..
if someone moves in they are gonna have scripts that come with them..performance is gonna be affected..sometimes it's worse than others..

If they're not physical, they're probably not causing problems.

If they're physical, and if they're not world axis aligned and/or have lots of stuff interacting with them, they can cause performance problems..

Well recently i redesigned my TARDIS house into a quarter sim sized station.

The outer shell of the 3 main "blocks" are 7 massive cut megaprim cylinders. The floor dividers are 16 50x50x0.5 cylinders as well with megaprim 50x50x40 cones for the roof.

Sim performance, fps, dilation and all that didnt change one bit.

Gotta love the salt hud

I'd suspect that if they're used in a way that doesn't trigger any of the physics problems mentioned, megas might even improve sim and/or viewer performance?

Consider a 100mx100m floor. As a single mega-prim, there are 6 faces to render, and one object for the sim to keep track of. Make it out of regular prim and both figures rise by a factor of 100. Even if you discount hidden faces, there are still 240. Of which 120 could be in-view at once, as opposed to 3 for the mega-prim.

Or am I making canine vocalisations at an inappropriate arboreal herbage?

edit: Just read the link on region optimisation, and it says: "Big prims greater than 10 m x 10 m can slow down viewer FPS. These kind of objects cannot be created under normal circumstances." I wonder why that happens?

The only big issue I had was with megaprim transparency. Client draw lag. I get another computer with better video card so is not an issue like before.

If anyone has a relatively empty sim for a little while, I'd be happy to test this. I figure that rezzing a few hundred medium sized megaprims and comparing it to the equivalent of a few thousand normal prims would be revealing (checking both sim frame rate and client frame rate).

I doubt the statement in the link above is accurate -- I suspect that at best it's overgeneralized. Furthermore, we know that very large physical or nonphantom prims can cause *server* issues, whereas the link only mentions *client* problems, and for prims over 10x10. I suspect that whoever wrote that didn't really understand this particular issue. It's hard to collect all the disparate facts for that kind of article, so it's understandable.

I'd love the hear the results from that test.

I know many say they create no problems, but how many are they talking about?

One sim had over 400 megaprims on in, many/most physical and apparently they were causing issues.

I'm sure that Megaprims CAN cause problems, but some testing needs to happen to clarify WHEN they do.

I build a city 1000m up over my sim to race cars around, as I didnt realy care about the buildings I used mega prim cubes for them.... must have 40 or 50 floating up there and have noticed zero change in the sims performance.

400 physical prims of any size would be a load on the physics engine. Do you mean physical or do you mean non-phantom?

This is an abandoned sim, almost empty. I'm using it to check the impact of some scripts. Maybe is a good place to see the impact of megaprims.

http://slurl.com/secondlife/Woolsan/128/128/2
___________________

[EDIT] You can drop objects there for 10 minutes. Must be quickly, but enough to see the sim's performance.

This is a question that would be good for Andrew's office hour. What Sindy says above is based on her asking Andrew about the performance cost of volume detect'ing megaprims.

See http://wiki.secondlife.com/wiki/User:Andrew_Linden/Office_Hours/2008_09_11 starting at 17:43. The short answer is that the sim keeps a std::map of collision pairs and the map can grow really huge (aka: slow) when you throw megaprims into the mix and have lots of stuff interacting with them.

Objects with volume detect are treated a bit differently than physical ones - from what Andrew says, they don't appear to just be 'physical/phantom' prims - so there may be other differences that they didn't talk about..

Of course, this says nothing about viewer-side lag. Dunno anything about that..

In all things moderation. Last time Andrew Linden came to the forums to talk about megaprim performance, he implied that one reason megaprims should never be fully supported is because of the difficulty rendering them when they're near sim boundaries. When an avatar is looking across a sim boundary, the client "loads" 10m worth of visuals into that other sim. If that includes a 50m sphere, it can get tricky - even deciding what SHOULD be done would be tricky. When that discussion happened, before the short "booboo" that led to much greater availability of megaprims, LL seemed happy to let megaprims of reasonable size (50m or so and less) exist, so long as residents don't have an expectation that the rendering engine handles them perfectly when viewed across sim boundaries. I haven't noticed lately, but it used to be that megaprims would sort of "blink" out of existence as you cammed across sim boundaries. But, if the megaprim is of reasonable size, I've never noticed any performance impact, and LL doesn't seem greatly concerned.

On the other hand, there is a lot of anecdotal evidence of immense megaprims causing performance problems. Here's my personal one: I was living on an estate of about 20 islands. The estate owner decided it would be cool to make a "ceiling" that folks could use as a shared build area over the entire estate at 770m or so. He did this with one nonphysical megaprim per sim. The performance on these generally empty class 5 sims began fluctuating, with inexplicable downspikes in fps and little rushes of packet loss. Puzzling.

It'd be great to see some systematic tests.
.

Apologies for not making myself clearer

I just mean 400 ish Non Phantom ones.

Making as many as possible (eg not floors) Phantom seemed to have made all the difference.

Thanks -- 10 minutes should be plenty long enough. It'll take me a couple days to prepare, and i never know when I'll have time ingame, but I'll see what I can do. Hopefully I can script an object to rez the prims without hitting the gray goo fence, which should make it simpler.

Hmm, though -- I'll need a bunch of avs tramping around as well, to help, in order to test server load. I don't have a bot army, so I'd need volunteers. But maybe I can find enough impromptu friends to come help.

I plan to focus on what I think is reasonable use of megaprims, meaning ones well under sim-sized, including both phantom and non-phantom. Non-phantom is important for floors and walls.

I may not be able to test sim border issues, unfortunately, so results probably won't be totally conclusive. If anyone has any important considerations to include, please pipe up, and I'll ask a few smart folks I know who might have some ideas.

Thanks for the feedback guys, I'll be interested to see the results of Lear's investigation.

It's interesting that when I searched the JIRA issues, there are only a few open ones dealing with prim size and not much interest in those at that.

I made my last skystore out of 20x20x.5 megaprims which cut the total prims by a factor of 4 and gave it a nice seamless look. I really think reasonably sized ones need to be officially supported while I'm all in favor of nuking the huge sim spanning ones. My limited understanding is that SL can deal with physical prims up to a certain volume.

>BUMP!<

I'd be interested to know the results as well. I'm currently building a store on a parcel I've just purchased and it seems I'd be able to save a number of prims by using Megaprims for floors, trim and such. none of them would be larger than 20 x 20 x 2, but still it makes one wonder. It would be great to know for sure.

Sigh ... as much as I want to investigate, I don't know when I'll have time -- pesky RL. Hopefully within a week or two.

I agree that those 20x20's are terribly useful!

One would think LL has already tested these things. I wonder why they just do not come out and say what they have found. If they were seriously detrimental to the functioning of the grid or the clients then more than likely they would be abolished from the grid.

I don't know how much difference it makes under havok 4 but before that it was recommendable to place any cuts on megaprims on the sides(s) away from traffic to cut down on physics handling (and associated bugs back then)