User:Andrew Linden/Office Hours/2008 09 04

Transcript of Andrew Linden's office hours:

[17:02]	Arawn Spitteler:	Using Vista for the Servers?
[17:02]	Sindy Tsure:	1.21.. i'd be happy to have it run long enough to run outta memory..
[17:02]	Sindy Tsure:	windows server 2008, actually..
[17:02]	Simon Linden:	Arawn - no, I'm out of the loop for viewer issues. Is that on an RC viewer?
[17:02]	Sindy Tsure:	not as bad as vista but in the family
[17:03]	Sindy Tsure:	1.21 is at RC0
[17:03]	Sindy Tsure:	heya andrew
[17:03]	Arawn Spitteler:	I can run it to memory leak crash, but using the one from the Beta Grid
[17:03]	Andrew Linden:	hello
[17:03]	Sindy Tsure finishes dinner and afks for a minute
[17:03]	Simon Linden:	Is there a jira? I can check our internal one that's linked to it
[17:03]	Sindy Tsure:	you can start without me, andrew
[17:03]	Sindy Tsure laughs :)
[17:04]	Arawn Spitteler:	I kind of figured, the stability of 1.21 would be broadly obvious.
[17:04]	Arawn Spitteler:	There's also some question, whether SL's been laggier, lately. Is that new with 1.24?
[17:04]	Andrew Linden:	not necessarily... I haven't been paying attention to 1.21-Viewer stuff either
[17:05]	Arawn Spitteler:	My firefox has taken to dissappearing
[17:05]	Andrew Linden:	There could very well be some 1.24-Server related lagginess. I've heard rumors of certain stuff causing problems but don't know any details
[17:06]	Andrew Linden:	and the updates for 1.24 are coming along quite quickly
[17:06]	Arawn Spitteler:	CG on vacation?
[17:06]	Andrew Linden:	Yes, I think CG is out
[17:06]	Simon Linden:	CG's down at SLCC
[17:06]	Andrew Linden:	Ah
[17:06]	Sindy Tsure has noticed a definite increase in the time it takes to rez stuff and in how long textures stay gray.. we sorta abuse our sim, though..
[17:07]	Sindy Tsure:	there's a bunch of forums talk about it, too
[17:07]	Sindy Tsure:	heya rex
[17:07]	Arawn Spitteler:	A thouroughly abused sim brings the lag into better relief. Hi Rexz
[17:07]	Rex Cronon:	hi sindy
[17:07]	Sindy Tsure nods
[17:07]	Rex Cronon:	hi arawn, andrew, simon
[17:07]	Andrew Linden:	hiya
[17:07]	Simon Linden:	Hi Rex
[17:07]	Arawn Spitteler:	Oh, TP issue. Is it too late, to get the unfix into 1.25?
[17:08]	Andrew Linden:	What TP issue would that be?
[17:08]	Arawn Spitteler:	SVC-1931, where the link distance was reduced to H1 dimensions.
[17:08]	Sindy Tsure:	though i do have to say that even with the (apparent) increased lag of late, we're still waaaay better than last year
[17:09]	Andrew Linden:	ah, the SLPP issue
[17:09]	Andrew Linden:	where last year is ... last September?
[17:09]	Sindy Tsure:	uh.. pre-h4, probably
[17:09]	Sindy Tsure:	which does include last september :)
[17:10]	Andrew Linden:	for the record, I think Arawn meant SVC-2931
[17:10]	Simon Linden:	We may have some people looking at the whole rez issue soon. It's beyond a few bug fixes and optimizations, it needs more of an overhaul
[17:10]	Arawn Spitteler:	Eep, yeah, what's 1931?
[17:11]	Sindy Tsure nods. you've mentioned before that you were looking at the slowdown when high-prim people tp in
[17:11]	Andrew Linden:	Well, before we get into that, and to allow for any late comers to participate... I've got a few small announcements...
[17:12]	Andrew Linden:	I successfully got our SL server code building on gcc-4 + 64bit last night... havent' actually tested it yet, but it compiled
[17:12]	Sindy Tsure blinks
[17:12]	Arawn Spitteler:	Gnu-C?
[17:12]	Andrew Linden:	I'm currently cleaning stuff up now, hopefully I'll have a branch others can play with by the end of tomorrow
[17:13]	Arawn Spitteler:	Will it be open to user widgets, or Gods Only?
[17:13]	Andrew Linden:	yeah, GNU Compiler Collection - 4.1.2 I think
[17:13]	Sindy Tsure:	oh.. you're already running a 64-bit OS on sim hosts?
[17:14]	Andrew Linden:	it won't be up for testing for a while, but there are hopes that the 64bit versions will provide a modest (5% - 10%) performance increase all around -- yet to be proven
[17:14]	Andrew Linden:	we're setting up 64-bits hosts now
[17:14]	Andrew Linden:	the tweaking of the 64bit configuration is in progress
[17:14]	Arawn Spitteler:	64 bits being hardware difference?
[17:14]	Andrew Linden:	and I was helping by getting our server code ported
[17:15]	Andrew Linden:	all of our hardware is currently 64bit
[17:15]	Andrew Linden:	however the vast majority of it (all of the SL simulator hosts) run in 32bit mod3e
[17:16]	Andrew Linden:	The other small announcement is that I've got a hard stop at 18:00 tonight... my net is going down at that time
[17:16]	Rex Cronon:	?
[17:16]	Andrew Linden:	so I'm going to try to grab the transcripts right before taht
[17:16]	Arawn Spitteler:	We need a special net-crash poofer
[17:16]	Andrew Linden:	that is, there is maintenance being done on my local net at 18:00
[17:16]	Andrew Linden:	the main switch is being swapped out
[17:17]	Sindy Tsure has 2 questions related to the 64-bit stuff
[17:17]	Andrew Linden:	ask Sindy
[17:17]	Sindy Tsure:	this stuff's a bit below where i usually live...
[17:17]	Sindy Tsure:	will moving the sim code to 64-bit eat more memory?
[17:17]	Sindy Tsure lives on an abused class-4 host..
[17:18]	Sindy Tsure:	which has 1/2 the memory of a class 5.. between that and the increase in memory for mono scripts, i've been wondering about memory lately
[17:18]	Andrew Linden:	Sindy, I think it might use up a bit more memory, but not much
[17:19]	Sindy Tsure:	fair enough.. was wondering more about it doubling the footprint.. :)
[17:19]	Andrew Linden:	the memory will go up by however many pointers we're carrying around...
[17:19]	Andrew Linden:	4 bytes per pointer...
[17:20]	Rex Cronon:	i guess with two words u can reach anything:)
[17:20]	Andrew Linden:	so assuming a million pointers that would be 4MB... actually pretty small compared to the typical size of a simulator process
[17:20]	Sindy Tsure:	other quetion i think i asked a while ago but i either missed your answer or you missed the question
[17:20]	Sindy Tsure:	from the forums..
[17:20]	Andrew Linden:	Meanwhile... 32bit machines are limited to 4GB of memory space
[17:21]	Andrew Linden:	64bit machines can hold more RAM
[17:21]	Sindy Tsure:	couple years ago, ian linden announced class 5 hardware and gave some specs..
[17:21]	Sindy Tsure:	the question is how closely does new hardware follow that recipe?
[17:21]	Sindy Tsure:	like.. will some class 5 machines be twin, dual-core and some uni quad-core?
[17:22]	Sindy Tsure:	http://blog.secondlife.com/2006/10/16/looking-forward-to-class-5/ is the old blog post
[17:22]	Andrew Linden:	I think class 5 hardware are all dual CPU dual core... so 4 cores
[17:22]	Andrew Linden:	however, the newer class 5's may have faster CPU and bus clocks... don't know for sure
[17:22]	Sindy Tsure:	ok.. ty!
[17:23]	Andrew Linden:	but since class 5 has been purchased for years now... I'm guessing that the clock speeds have changed
[17:23]	Les White:	hiyas
[17:23]	Sindy Tsure:	heya les
[17:23]	Andrew Linden:	meanwhile... we are buying class 6 hardware, which we are using to test the 64bit install, I think
[17:23]	Arawn Spitteler:	How many lindens for a used server?
[17:23]	Andrew Linden:	however I don't know any of the stats on class 6
[17:23]	Andrew Linden:	actually, we're running 64bits on all the hardware we have
[17:23]	Sindy Tsure:	this is the first 'official' mention of class 6, andrew
[17:24]	Sindy Tsure calls slherald
[17:24]	Andrew Linden:	which is part of the effort to get 64bits to work across the board
[17:24]	Sindy Tsure:	:P
[17:24]	Sindy Tsure:	you can edit it from the transcript.. i won't tell
[17:24]	Andrew Linden:	sure go ahead. I've heard "class 6" mentioned internally, but don't know any of the details
[17:25]	Sindy Tsure:	:)
[17:25]	Andrew Linden:	Arwan, dunno when we will be liquidating old hardware\
[17:25]	Andrew Linden:	however, when we do it will probably be on the order of hundreds of servers at a time, so bulk buyers only
[17:25]	Sindy Tsure:	hopefully, they'll deploy it to old mainland first.. get everything up to class 5/6
[17:26]	Arawn Spitteler:	Are they just typical motherboards? I'm on a 4 year old Dell
[17:26]	Simon Linden:	They're all rack-mounted servers
[17:26]	Les White:	i doubt it Sindy
[17:26]	Andrew Linden:	Arwan, they are not good machines for desktop use
[17:26]	Sindy Tsure:	they probably don't even have graphics cards in them
[17:26]	Simon Linden:	very, very noisy fans
[17:26]	Andrew Linden:	they are 1U (1 unit high) pizza boxen
[17:26]	Andrew Linden:	they are very loud, and don't have AGP slots
[17:27]	Andrew Linden:	however, they might have a PCI-E slot. Some do I'm pretty sure
[17:27]	Andrew Linden:	but you couldn't fit a video card in the case
[17:27]	Sindy Tsure:	think i heard a linden say that one of the most expensive things a sim does is make it's square on the map, since it has to do it all in software
[17:28]	Sindy Tsure:	maybe that's an urban legand, tho
[17:28]	Andrew Linden:	it is done in software, however I think the copying and parsing of the data takes much longer than the rendering
[17:29]	Andrew Linden:	it only needs to render a few frames, all at fairly low resolution
[17:29]	Sindy Tsure:	and not the whole volume
[17:29]	Simon Linden:	We've been running some profiling the past week or two on some high-traffic sims. Rezzing stuff, passing around textures, and updates are the biggest time sinks
[17:29]	Les White:	yes. rezzing is a killer
[17:29]	Les White:	and physical shapes too
[17:29]	Andrew Linden:	maybe we should send the textures via HTTP...
[17:29]	Sindy Tsure:	updates? like scripted prim movement stuff?
[17:29]	Les White:	10ms 30 prim cars...erm wtf
[17:29]	Andrew Linden:	(that was an inside joke)
[17:30]	Simon Linden:	There are some one-time events that are a slowdown, like arrival of a complicated AV with lots of attachments
[17:30]	Les White:	yeah. TPs in make a crunch too. cant do much about that
[17:30]	Simon Linden:	Ah, cars ... hopefully they will get faster soon
[17:30]	Sindy Tsure:	just add more gears?
[17:31]	Simon Linden:	Fewer and simpler shapes in physics
[17:31]	Les White:	he means faster math. some cars use half a CPU just to be alive
[17:31]	Simon Linden:	If you all made your cars look like a shoebox, it would be much faster :)
[17:31]	Arawn Spitteler has a three prim carriage, great for Lag Orbitting
[17:31]	Les White:	indeed. some of us are
[17:31]	Les White:	but the whole attachment thing is just silly
[17:31]	Les White:	set link parts phantom for the win!
[17:31]	Simon Linden:	Actually, it's more 'phantom-like'
[17:32]	Arawn Spitteler:	Attachments are phantom already.
[17:32]	Simon Linden:	Since phantom objects cause collision events, they have to be included in the physics engine
[17:32]	Simon Linden:	I'm thinking of a 'pure phantom' where they aren't even added to Havok
[17:32]	Les White:	even better
[17:32]	Andrew Linden:	"phantom" actually means "collides only with the terrain"
[17:33]	Arawn Spitteler:	My original desire for linked phantom was to add the mass, without the collison shape.
[17:33]	Simon Linden:	So there isn't much or any gain in performance when things are phantom - there still are calculations going on for the terrain collision and events with other bodies
[17:33]	Les White:	i dont know.. when someone wheres an attachment car the sim use is low. when the car is solid..not so low :)
[17:34]	Sindy Tsure:	if you enclose a complicated physical object in a box, it's faster?
[17:34]	Les White:	no
[17:34]	Andrew Linden:	Now that we compute the mass properties ourselves it will be easier than ever to NOT include certain prims in the physics engine
[17:34]	Les White:	that would be a beautifull thing
[17:34]	Les White:	it's really overdefining the physical shape on some things
[17:34]	Simon Linden:	Andrew - you're right. In my tests, when I omitted a child-prim's shape from the physics engine, the mass doesn't change :)
[17:35]	Les White:	simon, are you sayign these pants make my ass look fat?
[17:35]	Andrew Linden:	excellent. I'm particularly proud of the mass properties computation code
[17:35]	Sindy Tsure is still confused..
[17:35]	Les White:	too random. ignore me :)
[17:36]	Sindy Tsure:	if you put a transparent shoe box around a complicated shape, it's less laggy?
[17:36]	Les White:	no
[17:36]	Sindy Tsure:	should it be?
[17:36]	Simon Linden:	No, we're thinking of a feature where builders could have more control over the shapes used in the physics engine
[17:36]	Sindy Tsure:	if not, why not?
[17:37]	Andrew Linden:	actually... if the complex shape is much smaller, and centered deep inside the simple box... such that no collisions ever get close you might save some CPU cycles
[17:37]	Les White:	we need less math there or better servers (go c6!)
[17:37]	Simon Linden:	So let's say you have a car ... now, the wheels may be a torus or squashed ball, and that full complex shape is used in the physics engine.
[17:37]	Andrew Linden:	however, if the complex shape is inside the box, just under the surface then you save little to none
[17:37]	Simon Linden:	If, as a builder, you could say "use a box" or "skip this shape" you could tweak the object to have less load on the physics engine
[17:37]	Les White:	for sure
[17:38]	Les White:	definable hull would solve it all
[17:38]	Les White:	cept the really odd shaped things, but the trade for lighter use would be good i think
[17:38]	Andrew Linden:	only if you defined a less complicated hull
[17:38]	Sindy Tsure:	hm..
[17:38]	Rex Cronon:	could u use a sculptie to define the hull?
[17:38]	Les White:	sculpts i find to be not too bad on the phsyics
[17:39]	Andrew Linden:	most odd shaped things could be broken down into groups of simple shapes
[17:39]	Andrew Linden:	sculpts are a special case for collisions... they typically don't have a collision shape that is anything like their visible shape.
[17:40]	Simon Linden:	That's one of our debates - how to put this into the viewer, and how complicated to make it. A definable hull would be powerful, but more work in the viewer to make it easy to use
[17:40]	Sindy Tsure:	could start it as a prim property with only script access
[17:40]	Andrew Linden is in favor is some simple collision hint/features
[17:40]	Arawn Spitteler:	Different order of Viewer?
[17:40]	Les White:	maybe just allowing linked parts to goto some form of noclip will do it
[17:40]	Rex Cronon:	i mean. that could it be possible to specify a mesh generated from the sculpt image to be used as a collision shape for a linked set?
[17:41]	Andrew Linden:	proper colliding sculpt shapes are something that needs to be done
[17:41]	Les White:	they are not too bad i've found
[17:42]	Les White:	though with h4 things like to "touch" when they are a mile apart still
[17:42]	Simon Linden:	Probably the first step with sculpties is using something like their viewable mesh in the physics engine - perhaps with some triangle reduction applied
[17:42]	Andrew Linden:	however... defining the collision hull using a sculpt that is independent of the actual shape of the collection... that would be a bit more complicated, but not outrageously so
[17:42]	Les White:	yeah
[17:42]	Les White:	not user friendly
[17:42]	Les White:	a simple two vecotor convex kinda rectangle would do it maybe?
[17:42]	Arawn Spitteler:	Under the Objects Tab, Precision of collision?
[17:43]	Les White:	the trade would need to be clear else all will put percision to 100
[17:43]	Les White:	like how they use 1024 tx for vendors. cause it looks better!
[17:43]	Arawn Spitteler:	Simple per-prim slider. We also need to be able to see collision spaces, as with Sculpties, and that four inch buffer.
[17:44]	Andrew Linden:	content creators would want to be able to visualize the true collision shape
[17:44]	Les White:	or just some back end to reduce the amount of triangulation when it gets out of hand
[17:44]	Simon Linden:	Well, a slider might work. At a basic level, I was thinking of being able to pick "normal", 'use a box", "use a sphere", or "no shape" for the physics attribute
[17:44]	Les White:	it's only some shapes it over defines
[17:44]	Les White:	sounds clean simon
[17:46]	Andrew Linden:	I've been wanting to code up a little "smart" collision proxy calculator that would hunt for the "best" convex shape approximation for various concave shapes
[17:46]	Andrew Linden:	sphere vs box vs cylinder vs convex_hull
[17:46]	Rex Cronon:	vs torus?
[17:46]	Les White:	thar be no torii mate!
[17:47]	Simon Linden:	Torii turn into a complex mesh - they cost a lot
[17:47]	Andrew Linden:	the idea being... the content creator wouln't have to specify "box" or "sphere"... just specify "simplify" and the calculator would find the best fits
[17:47]	Les White:	donut holes are expensive :)
[17:47]	Arawn Spitteler:	Isn't that what Sculpties staret as?
[17:47]	Les White:	spheres
[17:48]	Andrew Linden:	no, as of Havok4 sculpties collide as doughnuts with the holes filled (convex hull of torus)
[17:48]	Les White:	o rly
[17:48]	Les White:	i should have a joke for that
[17:48]	Sindy Tsure:	who knew?
[17:48]	Arawn Spitteler:	Jelly Doinut Collision shape
[17:48]	Les White:	it can get messy :)
[17:49]	Arawn Spitteler:	"Ich bin ein Berliner
[17:49]	Rex Cronon:	what do u know. all those sculptie bullets people are using, are actually doughnuts:)
[17:49]	Andrew Linden:	as to SVC-2931... what super cool content was using the large bounds of llSetLinkPrimitiveParams() of avatars?
[17:49]	Arawn Spitteler:	Teleporters, mostly
[17:49]	Rex Cronon:	yes
[17:49]	Les White:	TPers?
[17:49]	Sindy Tsure:	replacement for warppos..
[17:49]	Arawn Spitteler:	Someone sold U$500 worth
[17:50]	Rex Cronon:	and NPVs
[17:50]	Sindy Tsure:	really? i'd have written that for only.. er.. US$250..
[17:50]	Les White:	wonder what ever happened to llTPAgent
[17:50]	Andrew Linden:	Well, the feature only worked when the avatar was sitting...
[17:51]	Andrew Linden:	would the avatar then be unseated to achieve the avatar relocation?
[17:51]	Arawn Spitteler:	I can figure how to get WarpPos across a sim border, but not back
[17:51]	Sindy Tsure still votes that llTeleport would be the best way to fix that..
[17:51]	Rex Cronon:	it can be done
[17:51]	Les White:	yes. it warps, unsits. ding! you are there
[17:51]	Arawn Spitteler:	SPP and SLPP both carry the .2 second delay, that keeps a person from flying. I'll be doing a lecture on that, tommorrow, if anyone shows up
[17:52]	Andrew Linden:	hrm...
[17:52]	Sindy Tsure:	my object to those is that the avatar movement is a side-effect - i don't think LL intended for them to do that stuff
[17:53]	Arawn Spitteler:	I should Jira a very slight problem with Teleporters, is that drag copy leaves a left click touch, rather than left lcick sit
[17:53]	Sindy Tsure:	which makes them more likely to break..
[17:53]	Sindy Tsure:	llTeleport would be something that's _supposed_ to do that - less likely to break
[17:53]	Andrew Linden:	I'm more inclined to try to push for llTeleport() instead of unfixing the SLPP-avatar misfeature
[17:53]	Sindy Tsure:	*objection
[17:53]	Les White:	i dont think that is a bug Arawn
[17:53]	Les White:	just prim param
[17:53]	Les White:	needs to be defined
[17:53]	Andrew Linden:	however, I just know that Estate owners are going to ask to be able to optionally disable llTeleport()
[17:53]	Les White:	they drop settext and other params when you clone too
[17:53]	Arawn Spitteler:	It was functioning fine, until our last upgrade.
[17:54]	Les White:	hmm
[17:54]	Arawn Spitteler:	llTeleport should be an estate owner function, just as G-Team on the mainland can summon
[17:54]	Sindy Tsure:	didn't it also break earlier this year?
[17:54]	Les White:	also...mega prims have issues again
[17:54]	Les White:	cant sit inside the bounding box. as 2 or 3 updates ago
[17:54]	Arawn Spitteler:	Megaprim issues?
[17:54]	Andrew Linden:	ack, my net is going down in 5 minutes
[17:54]	Les White:	the "terraframed" sims
[17:55]	Arawn Spitteler:	I'll have to go back and try my bridge
[17:55]	Les White:	a mountian range outside the sim on a big prim dissallows sitting inside the prims square area
[17:55]	Les White:	on script sits work
[17:55]	Les White:	this was broke then fixed a few server codes ago. broke again with 1.22 i think
[17:55]	Les White:	no killer
[17:55]	Sindy Tsure:	IS guys are always late, andrew.. plenty of time..
[17:55]	Arawn Spitteler:	The railings on my bridge used to be fine seats, but I'll have to look
[17:56]	Les White:	script seats will work
[17:56]	Les White:	regular sit. not
[17:56]	Les White:	i think it was simon or kelly fixed it last time
[17:56]	Andrew Linden:	broke in 1.22?
[17:56]	Les White:	yeah...maybe even 1.21
[17:56]	Simon Linden:	I remember fixing a "cant
[17:56]	Les White:	it broke with the first havok deploy. then fixed 2 or 3 later. now broke again
[17:56]	Simon Linden:	... can't sit inside a megaprim bug"
[17:57]	Les White:	right
[17:57]	Les White:	if you move outside of the prims box area all works well again
[17:57]	Les White:	we are using sculpted 1024X1024X100 prims for outside hills
[17:58]	Les White:	so no sitting below 100m :)
[17:58]	Simon Linden:	And this worked before?
[17:58]	Les White:	yes it did
[17:58]	Les White:	worked fine
[17:58]	Les White:	h4 killed it
[17:58]	Andrew Linden:	ok sounds like a perfect bug for Simon's bug-fixing-sprint which is going on this week
[17:58]	Les White:	then it was fixed. now it's dead again
[17:58]	Sindy Tsure:	go simon!
[17:59]	Andrew Linden:	I'm going to have to go now.
[17:59]	Simon Linden:	Can you file a jira, and put in the region and position where the sit fails? If I can get that info to reproduce it, it should be easier to fix
[17:59]	Les White:	see you Andrew
[17:59]	Rex Cronon:	bye andrew
[17:59]	Les White:	thanks for the time
[17:59]	Sindy Tsure:	cya, andrew! ty!
[17:59]	Les White:	ok Simon. i'll do that
[17:59]	Les White:	thanks guys