My findings on Second Life RC Magnum 11.05.23.230743 - Mesh prep

[Ash Qin]

So today I have been heavily testing my region running Second Life RC Magnum 11.05.23.230743 (Mesh prep).

Before I get into specifics, I need to summarize a few things about my region. There have been significant changes to a lot of things that the content my region (Deshima) relies on. To give a bit of background information, my sim has a rather large space station build and relies heavily on physics, trickery and things that make Falcon cringe like a volume detect megaprim. There are also a few bots that do weird things on the region that I sometimes suspect may sometimes harm region performance, one example is Subnova Hax which sits and downloads the Second life map data and stores the map texture UUIDs in a database for others to access via the Subnova map API (example script using it).

The region has a lot of physical elements to it. The primary way to get up and down the station is elevators that travel through tubes, there are roughly about 600 scripted asteroids which dynamicly generate their shape, size, sculpt map, can be knocked around, blown up, form from dust etc. Vehicles that transverse the outside of the space station. There is a single large volume detect megaprim that is used to pick up vehicles, avatars and display them on a 3d representation and act on events depending where these are on the station. There are megaprim walls surrounding the station to prevent objects from flying off world as well as preventing people from accidentally flying their vehicles out of the region unless they intended to (there is a hole in one of the megaprim walls where a sort of vortex gateway sits for people who want to leave).

 

Going ahead with my findings. I am pleasantly surprised to say that the amount of issues I have encountered is minimal, I have thrown everything I could at both region physics, volume detect. The new physics engine appears to be handling very heavy loads well, which is great news for those of you in the SL combat scene.

Here are some pretty graphs to show how much the sim was overloaded and the continued performance. The white bits are due to sim restarts and testing different settings in the sim console. First white bit is due to the rolling restart.

 

Things that were not broken which concern the Second life Military Community (combat sims using LL damage) was 'armor' systems, where by someone sitting on a volume detect object cannot be killed by Linden damage (it's considered a feature by many, since they can build tanks and such where the avatar doesn't die, the tank needing it's hit points taken out and then scripted to explode or such). I admit that if I had been on the beta grid more often, I would have been able to confirm this sooner, however, I've had a lot of real life work to deal with thus preventing me from spending much time on things.

 

Issues I did find. Steep terrain, particularly newly created terrain can cause an avatar to jump continuously when standing on the side or even suddenly shoot them up into the sky. This does not seem to be the case when accurate terrain is enabled. Over time, the violent reaction of jumping or tossing goes away or reduces it self (like after an hour). This could be a big concern for sims that rely heavily on land mass for walking and such.

 

The next issue I found was when the sim physics was getting abused, the edges of prims had a very powerful rebounding effect, enough so that while there is heavy physics usage, my elevators are unable to transverse their tubes as they keep bouncing off the corners of prims that didn't cause an issue before the update. In this case, a 95% hollow cylinder sized at 4.300x4.300x2.875 would cause issues for a 3.500x3.500x0.090 cyliner traveling inside it. I was able to reduce the issue by flipping these mysterious collision_tolerance value in the sim console to 0.01 - I'd love to know exactly what this does.

 

When changing collision_tolerance between 0.1 and 0.01, I did some testing to see if volume detect prims, both thin and thick had any particular issues detecting prims I was firing at them. To my surprise, neither settings made any significant difference in results, even when overloading the sims physics while testing this. I did however notice that prims tended to do less bouncing off things at 0.01 and dying on contact like they're scripted to do. I feel that for these sim console functions that deal with physics, it would be a good idea to have more values for llGetEnv to return information on including collision_tolerance and accurate_terrain. Particularly in accurate_terrain, there are instances where vehicle creators can use some alternative prim configuration that handles better in that mode. It's entirely possible though that the sim host at the time may have been the cause for either reaction mentioned, as you have to restart the region to flip these values.

 

The sim is extremely fast at recovering after heavy physics usage has stopped, it's almost instantaneous, something I have not observed on previous server versions. I have not observed any noticeable negative effects in either way when the volume detect megaprim is rezzed or not. I did not find any issues with accuracy of the megaprim's detection either. I also have not noticed any major increase in events being fired either - Such changes would require changes to how certain functionality is throttled to work optimally.

 

I was disappointed to find that llCastRay is still disabled, which would definitely be of use for one of my recent projects involving turret systems that have predictive logic to determine where a target is going to be and fire off projectiles to hit them. llCastRay would have been super useful to determine if there was something obscuring the trajectory of the projectiles. I really look forward to the day it's enabled.

I have truly had minimal problems with this framework update, my fears of things breaking horribly didn't come true. I did notice the new sim console value 'max_prim_scale', which sets the maximum prim scale, sadly read only. I guess I'll get my 64meter prims later on. :3

 

I'll be comparing the sim performance over time and comparing it against past data too, I don't expect to see a major difference with what I've experienced so far. If others wish to do so as well, some of the statistics for Deshima are publicly accessible on:

http://www.subnova.com/secondlife/deshima/html/week.htm

 

In all,

Great work on the server!

[Acheron Gloom]

Great topic, I realy enjoyed reading it, however...



Doesn't that kinda show that theres been a huge increase in lag spikes since the update ? Or was that from you messing around and testing it? I'd assume the latter, since if it was the former you would've made a note.

 

I'm also curious about the collision tolerance setting.

[Ash Qin]

Yep, as mentioned in my post I was doing some heavy testing, which is why you see those spikes. What is interesting is when you look at the other stats and don't see them as adversely effected as sims are normally known to be when physics is heavily abused.

 

Sorry, I guess I should have clarified that further in my post.

[Acheron Gloom]

Ah, alright thanks. I was just confirming.

Thanks again for posting your results.

[Bavid Dailey]

Ash,

 

I'm astonished to find you have a tool that can collect this data; what is its provenance?

[Ash Qin]

---

Bavid Dailey wrote:

Ash,

 

I'm astonished to find you have a tool that can collect this data; what is its provenance?

---

I have two bots for monitoring statistics in Deshima. One of them is called "Deshima", which collects a lot of data and dumps it to CSV files (it also does a lot of other things), where I can open it up later in Microsoft Excel and make pretty graphs on just about anything. I wrote said bot using libomv.

 

Second one is ran by a friend of mine (lex.mars), which runs under the username "Seeker". I believe it's running on some bot software called "JVA" to collect the data and then later uses MRTG to draw the graphs you see and then uploaded to a website. I find myself relying more on the latter just because it's easy to just open a webpage and look as opposed to downloading the .csv files, importing it with Excel and then setting up the graphs I want to draw.

 

[Bavid Dailey]

Thanks for the information,  Both libomv and JVA look interesting, pity I don't run windows though :-(

[Ash Qin]

---

Bavid Dailey wrote:

Thanks for the information,  Both libomv and JVA look interesting, pity I don't run windows though :-(

---

My libomv bot is running on Linux under Mono.

[Aries Westland]

did you test avatar response to falling, i mean, the sim i am on is magnum RC since around 4 weeks i have found that i can no longer fall from around 30m up and keep walking by tapping the arrow keys, sounds weird i know but the line of gaming i'm into relies upon being able to fall from heighs and still be walking and not splatting in a heap on the floor. I have tried this method on bluesteel and letigre and secondlife simulator and they don't have this issue so i am guessing it's something to do with mesh physics changing somehow, any imput would be great