User:Andrew Linden/Office Hours/2008 10 16

Transcript of Andrew Linden's office hours:

[17:03]	Kitto Flora:	Hello
[17:03]	Ellla McMahon:	hello :))
[17:03]	Andrew Linden:	Hi Kitto
[17:04]	Kitto Flora:	Hi Simon
[17:04]	Simon Linden:	Hi Ella, Kitto, Andrew
[17:04]	Ellla McMahon:	hello Simon :)
[17:04]	Andrew Linden:	Hi
[17:05]	Andrew Linden:	Wow, today really flew by.
[17:05]	Andrew Linden:	Busy busy busy
[17:05]	Kitto Flora:	Were you having fun?
[17:05]	Andrew Linden:	But I was hear on time!
[17:06]	Andrew Linden:	Hrm... not really all that fun.
[17:06]	Kitto Flora:	Will you here what we have to say?
[17:06]	Andrew Linden:	Sure, I'm paying attention.
[17:06]	Kitto Flora:	:)
[17:06]	Ellla McMahon:	LOL
[17:07]	Andrew Linden:	They have the 1.25-Server branch made, but they are still trying to deploy it for internal tests.
[17:07]	Kitto Flora:	Will it appear on Preview first?
[17:07]	Andrew Linden:	1.25-Server coincides with some 64bit operating systems that will be going out around the same time
[17:07]	Andrew Linden:	so there are some config issues being resolved.
[17:08]	Andrew Linden:	Yeah, it will show up on some preview, however not until after QA has a chance to look at it for a few days.
[17:08]	Kitto Flora:	OK
[17:08]	Kitto Flora:	Are some servers 64-bit CPUs?
[17:08]	Andrew Linden:	All of our hardware is 64bit.
[17:08]	Rex Cronon:	hello everybody
[17:09]	Kitto Flora:	Hi Rex
[17:09]	Kitto Flora:	What cpu is in latest server, intel or amd?
[17:09]	Andrew Linden:	I think it is AMD
[17:10]	Kitto Flora:	I have an ongoing problem with lag, LSL and physics thats annoying, I'm wondering if there is a good way around this problem
[17:11]	Kitto Flora:	I slow vehicles to a stop by running at a slower speed for a while
[17:11]	Rex Cronon:	hi
[17:11]	Andrew Linden:	hrm... maybe I take that back... the one I looked at is Intel Xeoon 2.66 GHz... sim8670.agni
[17:11]	Kitto Flora:	I change the liner motor speed to say 2M / sec for 2 secs, then to 0
[17:11]	Kitto Flora:	So the vehicle should move 4M and then stop
[17:12]	Kitto Flora:	But when theres lag it moves much further
[17:12]	Kitto Flora:	I'm suspecting that the LSL timer() lags but the linear motor does not. Is this likely?
[17:12]	Andrew Linden:	Yes, that is probably what is happening
[17:13]	Andrew Linden:	the physics engine runs asynchronously with the script engine
[17:13]	Andrew Linden:	and the vehicle is actually a physics engine feature, with script hooks for changing its properties
[17:13]	Kitto Flora:	Thats what I though :/
[17:13]	Kitto Flora:	t
[17:13]	Andrew Linden:	so... hrm... there might be a different way to do it...
[17:14]	Andrew Linden:	There is probably a magic set of vehicle params that would get you a more reliable stop on a few rapid sets
[17:14]	Andrew Linden:	however there will still be some jitter in the script calls
[17:14]	Kitto Flora:	turn off the drive, leave physical and play with the linear friction?
[17:15]	Andrew Linden:	That is... if you set the linear_motor to zero velocity, then also set the friction and motor_vehicle_timescale correctly you could get more accuracy.
[17:15]	Kitto Flora:	OK
[17:15]	Andrew Linden:	yeah
[17:16]	Kitto Flora:	harder to set up as the vehicle params are non-obvious
[17:16]	Andrew Linden:	however, I dunno what the settings would be... it would be very much an effort of tweaking the numbers until you got it right
[17:16]	Kitto Flora:	But I have an ongoing problem with vehicles running off end of track
[17:16]	Kitto Flora:	in lag
[17:16]	Andrew Linden:	yeah, and it requires a sequence of script calls to do it
[17:16]	Andrew Linden:	and they won't all trigger at once, in general
[17:17]	Andrew Linden:	expecially under script lag
[17:17]	Kitto Flora:	Yeah but likely faster than laggy timer() event
[17:18]	Andrew Linden:	and the order in which they are called can matter... best not to tighten the linear_motor_timescale before setting the speed to what you want
[17:18]	Andrew Linden:	set the linear motor speed first when tightening the timescale, and the timescale first when relaxing it
[17:19]	Andrew Linden:	but that matters most when making big changes
[17:19]	Kitto Flora:	tighter linear_motor_timescale means vehicle will achieve the new speed faster?
[17:19]	Andrew Linden:	I debugged someone's hover vehicle where they were setting the global hover mode before they were changing the hover height
[17:20]	Andrew Linden:	and script aliasing with the physics engine caused it to break in an inconsistent way
[17:20]	Andrew Linden:	yes Kitto, by "tighter" I meant "shorter"
[17:20]	Kitto Flora:	Ok
[17:20]	Andrew Linden:	"faster" ... in less time
[17:20]	Kitto Flora:	Its what linear_motor_timescale actually does that I wondered about
[17:21]	Andrew Linden:	the timescales represent the period of time it takes for the corresponding effects to kick in
[17:21]	Kitto Flora:	Good
[17:22]	Kitto Flora:	That I can understand
[17:22]	Andrew Linden:	sorta... most of the effects actually kick to within (1 - 1/e) of their goal within the timescale
[17:22]	Andrew Linden:	er...
[17:23]	Andrew Linden:	1/e closer than they were at the beginning of the timescale
[17:24]	Kitto Flora:	For my vehicles , the riders never notice acceleration effects.
[17:24]	Kitto Flora:	If vehicles atrt & stop instantly no one complains :)
[17:24]	Kitto Flora:	start & stop
[17:24]	Kitto Flora:	But making them slow down at stations before stopping does look better
[17:25]	Andrew Linden:	Yeah, most of your trains and guided vehicles putter along at a fixed speed
[17:25]	Kitto Flora:	Sort of fixed
[17:25]	Andrew Linden:	so only the start and stop matter... and the stop is harder when aiming for a particular spot
[17:25]	Kitto Flora:	The analyse and average LSL sspeed and set speed accordingly
[17:26]	Andrew Linden:	really? You measure the speed of the script engine and try to workaround the script lag?
[17:26]	Kitto Flora:	Also the monorail slows on curves as that takes more computing
[17:26]	Kitto Flora:	Yes
[17:26]	Kitto Flora:	As they use target()
[17:26]	Andrew Linden:	llMoveToTarget() ?
[17:26]	Kitto Flora:	the script measures how fast the event shows up
[17:26]	Kitto Flora:	No
[17:27]	Kitto Flora:	llSetTarget()
[17:27]	Andrew Linden:	oh, they adjust their velocity constantly to try to achieve a target position? or target velocity() ?
[17:27]	Kitto Flora:	No.
[17:28]	Andrew Linden:	llSetTarget()... hrm, my mind is blanking on that feature
[17:28]	Kitto Flora:	As they go along the track they use the time it takes for the not_at_target() event to show up to figure out how fast LSL is running
[17:29]	Kitto Flora:	That is smoothed - averaged - and the result used to adjust the vehicle speed
[17:29]	Andrew Linden:	ah ok
[17:29]	Rex Cronon:	llsettarget an event when object gets to that posi
[17:29]	Kitto Flora:	So that theres likely enough LSL computing power to run the navigation calculations
[17:29]	Rex Cronon:	generates an*
[17:29]	Andrew Linden:	so the vehicles just go faster in less laggy regions because they can expect better steering response
[17:29]	Kitto Flora:	Yes
[17:30]	Kitto Flora:	You got it
[17:30]	Andrew Linden:	gotcha
[17:30]	Kitto Flora:	The steam train is a 4spd fwd, one speed reverse
[17:30]	Kitto Flora:	in a fast sim it goes fast
[17:30]	Kitto Flora:	in a slow sim it slows down
[17:30]	Kitto Flora:	If the TD gets bad or the fram rate, it stops
[17:31]	Kitto Flora:	That way it tends to stay on the track :)
[17:31]	Andrew Linden:	yup. We need to optimize the script engine some more.
[17:31]	Andrew Linden:	Kitto, are you using MONO yet?
[17:31]	Kitto Flora:	In some places yes
[17:31]	Andrew Linden:	does it help at all?
[17:31]	Kitto Flora:	I have an All MONO steamn engine on trial
[17:31]	Kitto Flora:	Its virtually no diffeernt
[17:31]	Arawn Spitteler wonders if Disable Camera Contraints is working as it should.
[17:32]	Kitto Flora:	MONO engine works same. The per fram script time might be 10% less at best
[17:32]	Kitto Flora:	But variations make it hard to tell
[17:33]	Andrew Linden:	right
[17:33]	Kitto Flora:	Its all events()
[17:33]	Kitto Flora:	Not really that much calc
[17:33]	Andrew Linden:	I suspect that a cleanup pass could gain as much script performance gain
[17:33]	Andrew Linden:	what do you think Simon?
[17:33]	Kitto Flora:	If events were faster - some events - it would be better
[17:34]	Andrew Linden:	You think we could profile the script engine and knock down a few hot spots?
[17:34]	Andrew Linden:	Or should we wait for Babbage?
[17:34]	Andrew Linden:	Simon is probably multi-tasking ;-)
[17:34]	Kitto Flora:	llSensor() is probably the biggest problem
[17:35]	Simon Linden:	I'M AWAKE. YES, I AM
[17:35]	Andrew Linden:	how is llSensor() the biggest problem?
[17:35]	Kitto Flora:	llSensor() with regex for the object name would sure help
[17:35]	Simon Linden:	What were you talking about cleaning up?
[17:35]	Rex Cronon:	if only volume detect would work on attachements:)
[17:35]	Andrew Linden:	I was wondering if you thought there was fertile ground in the script engine for optimizations.
[17:36]	Simon Linden:	Hmm, probably some stuff in the LSL engine. I haven't profiled that code specifically, but I'd be there are some bottlenecks there
[17:37]	Simon Linden:	I think the Mono people did some profiling and performance tuning in the new code, as they did have some problems early on in the project
[17:37]	Andrew Linden:	Oh right.
[17:37]	Andrew Linden:	They would have fixed the easy stuff if they found it.
[17:38]	Simon Linden:	I suspect they only looked at Mono, however
[17:38]	Kitto Flora:	That .. exclueds events() ?
[17:38]	Andrew Linden:	yes, and the LSL callback system is largly unchanged by the MONO work
[17:38]	Kitto Flora:	It apears to be unchanged
[17:38]	Andrew Linden:	which means... the cost of any llFoo() call, and events and sensors
[17:39]	Andrew Linden:	MONO primarily sped up regular logic calls... if, else and arithmetic,
[17:40]	Kitto Flora:	I would think wo
[17:40]	Kitto Flora:	so
[17:40]	Andrew Linden:	so if you were doing recursion, or doing a lot of math then you'd see very noticable improvements
[17:40]	Arawn Spitteler:	Does the event caller even see a difference?
[17:40]	Kitto Flora:	I do some of that but its mostly llSensor() and llTarget() for real-time vehicles
[17:40]	Andrew Linden:	perhaps the event caller is faster, but the cost of each event is largely unchanged
[17:41]	Arawn Spitteler:	I'd expect the event response to be faster
[17:41]	Kitto Flora:	I think it is about the same
[17:42]	Kitto Flora:	I think I get around the same rate of no_at_target
[17:42]	Kitto Flora:	The speed conrtrol seems unchanged
[17:42]	Andrew Linden:	well, it doesn't really matter if it takes you 1 minute, or 1 hour to decide to travel to the other side of the world... it still takes you a long time to get there once you leave
[17:42]	Kitto Flora:	So I recon the time to get that event is unchanged
[17:43]	Andrew Linden:	the scheduler is like the decision making process... the travel time is like the cost of the llFoo() call
[17:43]	Kitto Flora:	Oh - I am navigating on 2M to 10M intervals
[17:43]	Andrew Linden:	I flipped the order of the comparisons there
[17:43]	Kitto Flora:	2M at 6m/sec = 3 mav point cals / sec
[17:44]	Kitto Flora:	3 nav point
[17:44]	Arawn Spitteler:	What do you even use Sensor for?
[17:44]	Kitto Flora:	Finding the track ahead
[17:44]	Kitto Flora:	find track
[17:44]	Kitto Flora:	find location and type
[17:44]	Kitto Flora:	analyse where you have to go
[17:44]	Kitto Flora:	Then go there
[17:44]	Kitto Flora:	when you get there look for next track
[17:44]	Rex Cronon:	crashed:(
[17:45]	Kitto Flora:	WB
[17:45]	Rex Cronon:	thanks
[17:45]	Kitto Flora:	One of the bigger problems is sorting out what llSensor() finds
[17:45]	Kitto Flora:	Have to pick thru list looking for track
[17:46]	Kitto Flora:	Thats why regex would be good
[17:46]	Arawn Spitteler:	What's Regex?
[17:46]	Andrew Linden:	regex on object name?
[17:46]	Kitto Flora:	Wildcards
[17:46]	Kitto Flora:	Yes, andrew
[17:46]	Andrew Linden:	regex = regular expressions... wild cards
[17:47]	Kitto Flora:	llSensor("KFtrack*", ....)
[17:47]	Rex Cronon:	that would be nice
[17:47]	Rex Cronon:	:)
[17:47]	Andrew Linden:	for example, using . and * to specify any single character, or any sequency of characters
[17:48]	Kitto Flora:	I really dont know how much it would sped things up
[17:48]	Andrew Linden:	so Kitto, you want to search for multiple things and then sort through them?
[17:48]	Kitto Flora:	But the Monorail ,which uses only one name for the track can go much faster
[17:48]	Kitto Flora:	Yes, Andres
[17:48]	Kitto Flora:	w
[17:49]	Andrew Linden:	or your track all starts with "KFtrack"... so "KFtrack001" and "KFtrack002" ?
[17:49]	Kitto Flora:	Yes
[17:49]	Andrew Linden:	so... you have to run multiple scans and then figure out which one hit?
[17:49]	Kitto Flora:	Well one Sensor
[17:50]	Andrew Linden:	if you did one wild card scan you'd get several potential hits
[17:50]	Kitto Flora:	Then go thru the list of returns looking for object that starts 'KFtrack'
[17:50]	Object: Hello, Avatar!
[17:50]	Kitto Flora:	Youd only get the track section ahead, within range
[17:50]	Kitto Flora:	As it is I have to scan for *anything*
[17:51]	Kitto Flora:	I've had people build ssuspention bridges with 100 prims
[17:51]	Kitto Flora:	and the train stops on them cause the prims flood out the sensor
[17:51]	Andrew Linden:	you can restrict the sensor on an exact name...
[17:51]	Kitto Flora:	The track has no exact name
[17:52]	Kitto Flora:	Theres around 50 track types so far
[17:52]	Andrew Linden:	right, I get it. You can specify exact name, exact key, or exact owner
[17:52]	Arawn Spitteler:	My Rail Follower HUD follows objects that are named as described in Object Description
[17:52]	Andrew Linden:	or any of those three together
[17:52]	Kitto Flora:	Right Andrew
[17:53]	Andrew Linden:	but a name filter could reduce the positive hits drastically
[17:53]	Kitto Flora:	Right
[17:53]	Arawn Spitteler:	Listeners could call for responses
[17:53]	Kitto Flora:	It should reduce the sim load?
[17:53]	Andrew Linden:	Hrm... I wonder how much content that would break to introduce regex in the llSensor()
[17:53]	Kitto Flora:	Is * valis on an object name?
[17:53]	Kitto Flora:	* or ?
[17:53]	Arawn Spitteler:	Is there an available character, not used in names?
[17:53]	Andrew Linden:	it would break for anyone who was searching for their special object named "Object*"
[17:54]	Kitto Flora:	Yeah
[17:54]	Kitto Flora:	Looks like * is valid char in a name
[17:54]	Andrew Linden:	so that there is breakage is unquestionable... the real question is how much
[17:54]	Kitto Flora:	I never saw * used in an object name
[17:54]	Andrew Linden:	alternatively... llRegexSensor() could be added
[17:55]	Kitto Flora:	Searc Inv for *
[17:55]	Kitto Flora:	*1337~Tutenkhamun-18th-Dynasty.jpg
[17:55]	Andrew Linden:	That you haven't seen it does not mean it does not exist Kitto.
[17:55]	Kitto Flora:	One hit form 11,218 items
[17:55]	Kitto Flora:	Right
[17:55]	Arawn Spitteler:	\\
[17:55]	Kitto Flora:	But its pretty low
[17:55]	Andrew Linden:	I'm seaking from years of experience at breaking people's content.
[17:55]	Simon Linden:	Ha - just tried to re-name "Object" to "Object*?" and the viewer crashed
[17:56]	Kitto Flora:	Oh?
[17:56]	Kitto Flora:	I named on Object* and it was ok
[17:56]	Andrew Linden:	I would try to repro how, but don't want to crash.
[17:56]	Simon Linden:	Well, try with the ? after the office hours :)
[17:56]	Andrew Linden:	Yeah, maybe I can exit with style.
[17:56]	Rex Cronon:	with a bang:)
[17:57]	Kitto Flora:	Exit witha whizzle-pop
[17:57]	Andrew Linden:	No, that would be !
[17:57]	Kitto Flora:	Thats a whiz-pop
[17:57]	Arawn Spitteler:	\? ?
[17:58]	Andrew Linden:	Sigh... much of this feature and optimization speculation is pure whimsy...
[17:58]	Kitto Flora:	Indeed :(
[17:58]	Andrew Linden:	I've already got sooooo much work to do.
[17:58]	Arawn Spitteler:	Chat works across boundaries, doesn't it?
[17:58]	Kitto Flora:	More stuff - it would be nice to be able to allow next owner to mod texture, but nothing else
[17:58]	Andrew Linden:	yes, I think so Arawn.
[17:59]	Kitto Flora:	All chat but llRegionSay()
[17:59]	Andrew Linden:	well... if you could mod texture maybe you'd also like to mode color
[17:59]	Kitto Flora:	Yes
[17:59]	Kitto Flora:	Texture color
[17:59]	Andrew Linden:	maybe those two whould be grouped
[17:59]	Andrew Linden:	yup
[17:59]	Kitto Flora:	People wanna change textur/color on my vehicles
[17:59]	Arawn Spitteler:	So,m in stead of Sensor, we could simply ask the rails to answer, and those named correctly can be filtered after. Is there a native lag, on the listen event5?
[17:59]	Kitto Flora:	But if I set them 'mod' they'd be wrecked
[18:00]	Andrew Linden:	Hrm... the permissions bits are currently full. We'd have to use a longer bit register and mask
[18:00]	Kitto Flora:	The rails dont answer - they are passive
[18:00]	Andrew Linden:	Arawn, that would require script shouting, and a script with listener on every rail
[18:01]	Andrew Linden:	listeners are not all that cheap
[18:01]	Andrew Linden:	and the train would have to listen back, and potentially sort through a lot of chat noise
[18:01]	Arawn Spitteler:	Actually, it would require llwhisper, on a long nose
[18:01]	Kitto Flora:	Well - chat noise is low
[18:02]	Kitto Flora:	But that kinda comms is going to be slower than llSensor
[18:02]	Andrew Linden:	ah yes, whisper, and I guess the response noise wouldn't be any worse than positive sensor hit on all valid objects
[18:02]	Arawn Spitteler:	They'd be the only listeners on that channel, for the most part
[18:03]	Kitto Flora:	But the idea of putting a script in every track prim is .. not good.
[18:03]	Andrew Linden:	this hour is about up. Anyone have any last minute issues or questions that are other than what we've been talking about?
[18:03]	Arawn Spitteler:	It would be more localized that Sensor, and a question of Geometry, as to how far ahead would be good to search. SVC-2931?
[18:05]	Andrew Linden:	Sigh... I guess I should actually comment on SVC-2931.
[18:05]	Simon Linden:	I'm going to try and crash again, so I'll say 'bye' in case it's a repeatable bug...
[18:05]	Andrew Linden:	I'll make a note and try to comment tomorrow.
[18:05]	Rex Cronon:	bye:) succesfull crashing:)
[18:05]	Andrew Linden:	change the name on a temp-on-rez box Simon
[18:07]	Arawn Spitteler:	Anyone familiar with recent builds of Omega? I'm trying to compile on Glorious Haskel Compiler, but not getting anywhere.
[18:07]	Kitto Flora:	Is he dead, Jim?
[18:07]	Kitto Flora:	Ah
[18:07]	Kitto Flora:	he gone
[18:07]	Andrew Linden:	hrm... I changed the name on this one.
[18:07]	Object*?: Hello, Avatar!
[18:08]	Andrew Linden:	no crashy
[18:08]	Kitto Flora:	Maybe its a client version issue
[18:08]	Andrew Linden:	yeah, could be
[18:08]	Rex Cronon:	must have been something else
[18:09]	Kitto Flora:	Safe for me too
[18:09]	Kitto Flora:	On RC viewer