Mono Beta reports

Anyone have any interesting results to report from the Mono Beta Grid?

Speed Tests?

Still too buggy or did they close it?

At one time a mono test would have produced some comments in the forums here. I suppose the comments are all somewhere else like the wiki or the jira or some such.

I only tried one script and didn't compile, I got an error, I don't remember now exactly but it was about CIL...and it ended saying "Try again later".

At the moment, next to me there was someone using a fractal generator script that rezzed lot of little balls that together generated a fractal shape and it was very fast.

So yes, MONO can be good, but it depends on how it is implemented.
SL philosophy is keeping stuff buggy and go on with new things like Voice and Windlight, let them buggy and go on with other stuff and so on. In this case they CAN'T do this, this have to work well otherwise they will break SL more than it is now.

Shoot, mono is looking great to me. If a script does anything with http, it won't compile, so that's an area we can't explore yet. There are a couple of sim crashers to fix, and some behaviors to emulate. The speed boost on some scripts is incredible.

They fixed a bug that was killing most of my scripts, thereafter everything mostly seems to be working, apart from a couple of cases when something is now running too fast and some other script can't keep up!

I was informed there was something neat in "a corner" of Cordova in the beta grid. Cordova being one of the Mono sims in the Beta grid.

I flew to each corner several times and didn't see anything neat in the corners, anyone know what they were talking about?

Here are the results of sometesting I did recently:

Testing done with objects in SL and the MONO test grid.

Identical scripts were tested in the same sim, one set compiled for mono and the other for LSL.

Scripts were of two types, 1) the typical 5 line physical property setting script that is never run again and is responsible for about half of all scrips in SL by my guess. 2) A complex bit of machinery with listens, link listens, says, menus etc that is representative of a lot of things like jacuzzis, vehicles, games and vendors and the like.

MONO sim was a class 5 and had a 0.1 resting script load when testing LSL vs MONO script loads.

All times are in milliseconds

I did not have acess to the estate tools so better numbers could not be gotten, these numbers were tested 4-5 times each with a large multiple of objects to make the differences clear.

-----------------------------------------------
10 Shelly's Test

Test......................MONO.............LSL ...........Difference

Resting.................0.2..................0.2............No difference
Random Moving.....0.2 to 0.5........0.3 to 0.7....30% faster (2)
Calling..................0.3.................1.7 to 2.5....5 - 8 times faster (2)
Feeding................7.6 to 23..........1.8 to 3.1...4 to 8 times SLOWER (Feeding required rezzing an object with a script in it) (4)

--------------------------------------------------------
100 prims with scripted parameter settings scripts. 200 simple 5 line prim setup scripts total (One particle, one texture rotation per prim)

Test......................MONO.............LSL ...........Difference

Rezzing.................54 to 234........< 0.1..........Astronomical (4)
After Rez...............0.4.................0.4.............No difference (3)


-------------------------------------------------
For fun, 10 Shellys test on class 5 vs class 4 sim (LSL only)

Test......................Class 4...........Class 5 ...........Difference

Resting..................1.0.................0.2..................5 times faster
Random Moving......1.2 to1.7.........0.3 to 0.7........3 - 4 times faster
Calling...................4.4 to 5.5........1.7 to 2.5....... 2 - 3 times faster
Feeding..................7.6 to 23.........1.8 to 3.1.......4 to 8 times faster (1)

-------------------------------------------------
Preliminary Conclusions

1) This reaffirms my intial impressions that Class 5 sims seem to be about 3x faster than class 4 on script processing.

2) MONO will be a little bit faster overall given an average sim script load of 1k or more scripts, but... maybe only 50% more? Hard to predict but ceratainly not 50x or "up to 220x" or even 10x like I hoped. (Well in VERY specialized single math intensive instances yes, but in the total sim script load? Not so much).

3) The problem of "dead scripts" taking up CPU time is still there. Scripts that have no message processing are still being polled for no reason. Given the astronomical number of these scripts in an average sim and the complete waste of time they use for doing nothing (Unless of course you scour your entire sim for them and disable all you can), I was really hoping this would be fixed.

4) MONO seems to have a HUGE setup time penalty for rezzing scripts int he sim I mean HUGE. If a 1500 script dragon pops into your sim it'll take a big hit. And the "normal" script load of 100-200 scripts that most people seem to carry when they TP around will add even more lag to your sim when they rez up than you get already..

Bugs
---------------------------------------
* Floating Text not working
* Sounds not working
* Had to manually start some scripts, like the llSetScriptSTate is not working or something?

[Edited to restore spacing]

Excellent report, Farallon.

I wonder if it turns out that there is only a fifty percent speed improvement, would that mean that allocating resources towards mono development was an error, since the decision was based on an estimate of a tenfold increase in speed? It might be too early to draw conclusions.

That there are things it is slower at is depressing. There's nothing in SL that needs to be any slower.

I've lost almost all interest in these forums, which is why I didn't post much about my experiences here, but my experience testing this was just absolutely horrible.

I wrote about it in more detail here: http://www.daikonforge.com/webtab/2008/01/29/second-life-mono-vm-not-yet-useable/

Basically, many of my scripts wouldn't compile, and some of those that did failed to operate properly. Many scripts I tested ran extremely fast, a select few ran *extremely* slow, and some had errors I wasn't able to track down.

Of those few scripts that ran slow, I was unable to determine whether the behavior was because of the Mono VM alone or was a result of running both Mono and pre-Mono LSL scripts together in the same object. As the scripted object I was testing was my main product, and those scripts are all intended to work together, it was not worthwhile to me to try to isolate which of those possibilities was the correct one.

I'm extremely pleased that LL is working on Mono, and I have very high hopes for it in the future, but at the moment it's simply not usable for the kinds of things I can already do in pre-Mono LSL.