AW Groupies/Chat Logs/AWGroupies-2009-04-07

From Second Life Wiki
Jump to navigation Jump to search
  • [2009/04/07 09:26] Zha Ewry: Morning?
  • [2009/04/07 09:26] joao Mastroianni is Online
  • [2009/04/07 09:26] Zha Ewry: For who?
  • [2009/04/07 09:26] Saijanai Kuhn yawns. It's always mornign for some of us
  • [2009/04/07 09:26] Morgaine Dinova: For this world
  • [2009/04/07 09:27] Saijanai Kuhn is a f k for a bit. Doing transcript duty as usual
  • [2009/04/07 09:28] Jenna Poliatevska is Offline
  • [2009/04/07 09:29] Zha Ewry: Fair enough
  • [2009/04/07 09:30] Morgaine Dinova: Just poking random vga= numbers into the linux boot line to try get a better text console ... hope it doesn't destroy anything. They're from online info, but still seem to have random effect.
  • [2009/04/07 09:31] Morgaine Dinova: And they don't match up with the info from probing either. It's all a bit sordid.
  • [2009/04/07 09:31] Zha Ewry: I've not a ton of stuff for today
  • [2009/04/07 09:31] Zha Ewry: Wanted to very quickly touch on one thing in Christian's reply to my lates (and it looksl ike the MMOX list is back up, after much pain)
  • [2009/04/07 09:31] Morgaine Dinova: Hiya Aimee
  • [2009/04/07 09:32] Aimee Trescothick: hi :)
  • [2009/04/07 09:32] [[User: Latha Serevi|Latha S[\1] \2: erevi]]: woof
  • [2009/04/07 09:32] Dahlia Trimble: hi :)
  • [2009/04/07 09:33] Zha Ewry: So... I've said this, by Morgaine said it hasn't been terribly clear, and Christian's reply makes it clear it isn't to him as well.
  • [2009/04/07 09:33] Zha Ewry: So...
  • [2009/04/07 09:33] Zha Ewry: Seperation of concerns, security and policy
  • [2009/04/07 09:34] Zha Ewry: Let me try this again, (oh, and Jon Watte totally got this wrong too)
  • [2009/04/07 09:34] Zha Ewry: My take is we make sure the *SLOTS* are in the protocol
  • [2009/04/07 09:34] Zha Ewry: We need that, so peopel can chose to use them
  • [2009/04/07 09:35] Zha Ewry: Given that we're donig nice, extendiable stuff, that is almost zero cost (or can be zero cost, if you don't include them at all)
  • [2009/04/07 09:35] Zha Ewry: If your service doesn't chose to implement (by its policy) a requirement that you offfer up a given token, fine*[2009/04/07 09:35] Joel Savard is Online
  • [2009/04/07 09:35] Zha Ewry: (It can publish, its needs, on well known caps)
  • [2009/04/07 09:36] Zha Ewry: If it does need one, and you don't present it, http, handles this just fine
  • [2009/04/07 09:36] Zha Ewry: (401, Not authorized)
  • [2009/04/07 09:36] Zha Ewry: The policy choice of things like whitelists, blacklists,a nd requiring tokens is utterly seperable from the mechansisms which we define in prototcol*[2009/04/07 09:38] Goldie Katsu is Online
  • [2009/04/07 09:38] Zha Ewry concludes her droning has put everyone to sleep, or Morgaine has stuffed such a bad value in into VGA nobody can speak
  • [2009/04/07 09:39] Latha Serevi: Presumably the service-discovery mechanism can tell me what kind token I need, as part of the protocol?
  • [2009/04/07 09:39] Morgaine Dinova: That's a good way of putting it, re "slots". It makes the point that you have to implement the extension mechanism, even if you then reject the option as a matter of policy. That way clients of a service can use the same code with multiple services, whatever the actual policies implemented.
  • [2009/04/07 09:40] Zha Ewry: right
  • [2009/04/07 09:40] Zha Ewry: The point, is full seperation of concerns
  • [2009/04/07 09:40] Zha Ewry: We need
  • [2009/04/07 09:40] Zha Ewry: to permit people who want them, the spots
  • [2009/04/07 09:40] Zha Ewry: But...
  • [2009/04/07 09:40] Zha Ewry: We make it painless for those who don't need them
  • [2009/04/07 09:41] Zha Ewry: and.. let the handy fact of the web having worked out 99% of this, kick in
  • [2009/04/07 09:41] Zha Ewry: Realistically, both use cases are entirely valid
  • [2009/04/07 09:42] Zha Ewry: Some people will want to run services which don't have strong needs to secure data
  • [2009/04/07 09:42] Zha Ewry: Some people won't be able to use the protrocols at all, if they don't include them*[2009/04/07 09:42] Jonit Ivory is Offline
  • [2009/04/07 09:43] Zha Ewry: Seperating out the policy concern from the mechanism ought to let us andle both
  • [2009/04/07 09:43] Zha Ewry: slots helps with that Morgaine?
  • [2009/04/07 09:43] Zha Ewry: (I'm all for finding language which helps people sort out the insanity which is in my head)
  • [2009/04/07 09:44] Particle Control: Could not find script 'blood'.
  • [2009/04/07 09:44] Particle Control: Could not find script 'cloud'.
  • [2009/04/07 09:45] Morgaine Dinova: Well, "slots" isn't actually meaningful within the implementation (probably), but at least it gives people a mental model for it. We've been using "extensible" too loosely.
  • [2009/04/07 09:45] Zha Ewry nods
  • [2009/04/07 09:45] Zha Ewry: At a formal level, its probably rererving a named key for the item in the protocol
  • [2009/04/07 09:45] Zha Ewry: (reserving)*[2009/04/07 09:46] John Zhaoying is Online
  • [2009/04/07 09:46] [[User: [Inner Strength] 1.00 [script:Physics Control]|[Inner Strength] 1.00 [script:Physics Control]]]: Script run-time error
  • [2009/04/07 09:46] [[User: [Inner Strength] 1.00 [script:Physics Control]|[Inner Strength] 1.00 [script:Physics Control]]]: Stack-Heap Collision
  • [2009/04/07 09:48] Zha Ewry: I'm pretty much assuming, we'll do it fully nested, so we get to say "Here is where you put an "Identity token" and then.. allow people to define those, so "OpenID" becomes one possible one. This leads, inevitabbly towards the hungry dragon that is "We give you so many choices, you need profiles"
  • [2009/04/07 09:48] Morgaine Dinova: However, it still kind of brushes the issue that John brought up about trust agreements under the carpet. I think you need to be more explicit about the fact that a trust agreement isn't needed at all everywhere, only in specific cases, otherwise you're wide open to all the issues that people have raised. And to which Infinity agreed.
  • [2009/04/07 09:49] Zha Ewry: Sure
  • [2009/04/07 09:49] Morgaine Dinova: An "untrusted trust agreement" isn't enough.
  • [2009/04/07 09:49] Shamir Katsu: I think it is more precise to say that there is always a trust agreement, but that there are degenerate cases at the ends
  • [2009/04/07 09:49] Zha Ewry: So.. there are two lurking problems
  • [2009/04/07 09:49] Zha Ewry: One is a scaling issue
  • [2009/04/07 09:50] Dahlia Trimble: I kind of got the impression that if trust agreements were in place that interoperating with a region/grid that had no trust agreement was not possible
  • [2009/04/07 09:50] Zha Ewry: One is the brute fact that many, many endpoints in this dance won't ever know each other in advacne and won't have one in advance
  • [2009/04/07 09:50] Zha Ewry: That's a policy choice
  • [2009/04/07 09:50] Saijanai Kuhn: Dahlia, I think its more along the lines of "are very limited"
  • [2009/04/07 09:50] Saijanai Kuhn: "by policy" ;-)
  • [2009/04/07 09:50] Zha Ewry: and.. yes.. for some regions, the policy will clearly be "If I don't know you, I don't talk to you"
  • [2009/04/07 09:50] Morgaine Dinova: Shamir: lack of a trust agreement is not the same thing as a trust agreement that states nil trust, that's the issue. The first scales, the 2nd doesn't.
  • [2009/04/07 09:51] Zha Ewry: For others, it will be "I'll gladly accept your "open" content" but I won't give you my closed content"
  • [2009/04/07 09:51] Zha Ewry: And.. for a lot of the most common cases it will be
  • [2009/04/07 09:51] Zha Ewry: "Hey, I just don't care"
  • [2009/04/07 09:51] Zha Ewry: The scaling argument says
  • [2009/04/07 09:52] Saijanai Kuhn: that will cause interesting issues with viewers if they are in the "untrusted" camp and dont receive the "trusted" content in the first place
  • [2009/04/07 09:52] Morgaine Dinova: Given thousands or millions of worlds, you simply can't have trust agreements with them all, and if you automate one, it's meaningless trust, totally worthless.
  • [2009/04/07 09:52] Zha Ewry: I don't wan't and can't manage a seperate agreemeent between my regoin and every single Linden named endpoint
  • [2009/04/07 09:52] Zha Ewry: So, I need to lump them together
  • [2009/04/07 09:52] Zha Ewry nods
  • [2009/04/07 09:53] Zha Ewry: Right, end to end, for arbitrary end-points, its wicked hard to have meaningful trust
  • [2009/04/07 09:53] Zha Ewry: In particular, I strongly suspect, people will end up having to make hard choices*[2009/04/07 09:53] Talia Tokugawa is Online
  • [2009/04/07 09:54] Zha Ewry: "Do I want to affiliate with a trust scheme, so people can bring restricted content" or "amd I ok forcing people to wear "open" content on my region"
  • [2009/04/07 09:54] Goldie Katsu: But if you allow access to the world without an agreement there is an implicit trust - a sensible default "automated" one.
  • [2009/04/07 09:54] Zha Ewry: Right, one which doesn't let you get much special rights
  • [2009/04/07 09:54] Goldie Katsu: or unsensible default
  • [2009/04/07 09:55] Goldie Katsu: no requirement to be sensible
  • [2009/04/07 09:55] Zha Ewry laughs
  • [2009/04/07 09:55] Zha Ewry: None at all
  • [2009/04/07 09:56] Shamir Katsu: which is why I say that there is always a trust agreement, and having no agreement (you still need the behind the scenes machinery) just is a "sensible default"
  • [2009/04/07 09:56] Shamir Katsu: and right now we are talking about pointwise agreements between administrative domains, I take it
  • [2009/04/07 09:56] Zha Ewry: I'm also, to be honest, fully expecting people to sell content, which is high quality, and allowed to go everywhere, for a cost, essentially, accepting the risk of theft, for the higher reward. (and theyy won't lose the right to sue over its theft,, just know it will be wicked hard)
  • [2009/04/07 09:57] Zha Ewry: Well
  • [2009/04/07 09:57] Zha Ewry: at the protocol level, I'm arguing for making sure we slap in the places (and name them properly) for the tokens people need to manage that
  • [2009/04/07 09:57] Zha Ewry: Which policies people set, is close to orthogonal
  • [2009/04/07 09:58] Zha Ewry: (As long as we make sure we have use cases which give us some belief we have the right token, and the right ways of expressing what's needed, and failure)
  • [2009/04/07 09:58] Zha Ewry: Ideally, if you try to request an asset, which you don't have policy permission to use
  • [2009/04/07 09:58] Zha Ewry: you'll get a nice crisp error
  • [2009/04/07 09:58] Zha Ewry: like
  • [2009/04/07 09:59] Zha Ewry: "I can't deliver that asset to that sim, because it's not on my whitelist"
  • [2009/04/07 09:59] Morgaine Dinova: Hard to say, Shamir. We haven't really worked out the interop model down at the object end. The AD currently has a pile of services lumped in, so it's confusing matters. Eg. one might have assets stored in an asset service entirely elsewhere, so the AD has no business determining any trust issues for object transfer.
  • [2009/04/07 09:59] Zha Ewry: or
  • [2009/04/07 09:59] Zha Ewry: I'm assuming, until proven otherwise that...
  • [2009/04/07 10:00] Zha Ewry: we do protocol only at the "services endpoint" level
  • [2009/04/07 10:00] Shamir Katsu: I'd want to have a reference monitor sort of model, and the agreements, however they are represented, are rules used by that monitor, so that if you cannot transfer an object you can get that crisp error: cannot rez that object because domain X says you cannot transfer that object to here.
  • [2009/04/07 10:00] Zha Ewry: and that most policy will manage "domains" which are not, in fact mapped onto "agent domani" or "Region domain" but rather define a set of cooperative services (servers and endpoints) which share trust
  • [2009/04/07 10:01] Zha Ewry: The current highly coupled mash of services, is purely an artifact of Linden's evolution
  • [2009/04/07 10:01] Zha Ewry: THAT said
  • [2009/04/07 10:01] Morgaine Dinova: That would be good Zha. But it's a million miles away from Zero's thinking.
  • [2009/04/07 10:01] Zha Ewry: We need to make sure that we accomdate that mash
  • [2009/04/07 10:02] Geo Meek: good luck
  • [2009/04/07 10:02] Morgaine Dinova chuckles
  • [2009/04/07 10:02] Zha Ewry: Actually, Morgaine, I think its more that Zero has on his "I've got to make this work with me existing world" hat most of the time.When you pry that off, and make him think about it more abstractly, he does pretty much seem
  • [2009/04/07 10:02] Zha Ewry: to get it
  • [2009/04/07 10:03] Zha Ewry: And, I'm sympathetic, but only to thep oint where I am willing to say "This has to work with thier current mash as *ONE* deployment" not "The current mash is *the* deployment"
  • [2009/04/07 10:03] Morgaine Dinova: Yeah, hopefully. But at a rate of 2 hours every month, we're not even getting to touch his SL hat, let alone pry it off.
  • [2009/04/07 10:03] Zha Ewry chuckles
  • [2009/04/07 10:03] Zha Ewry: I think, at the spec level, its easier. But, agreed
  • [2009/04/07 10:04] Zha Ewry: Mind you.. the current mash may be annoying, it's also proven workable.
  • [2009/04/07 10:04] Zha Ewry: There are plenty of ways to re-swizzle the services which won't have nice properties. I'm sure we'll see people disocver some of those
  • [2009/04/07 10:05] Morgaine Dinova: I've got a couple of agenda items listed on Zero's page that hint at that problem, both in AWG sessions and in MMOX. Zero's not going to like it, but it's a very real problem: http://wiki.secondlife.com/wiki/User:Zero_Linden
  • [2009/04/07 10:06] Zha Ewry: Those ought to help
  • [2009/04/07 10:06] LathaSerevi: Shamir's use of the phrase "reference monitor" caught my attention. Sounded like a useful bit of theory. Looked up http://en.wikipedia.org/wiki/Reference_monitor . Would be*[2009/04/07 10:06] SignpostMarv Martin is Online
  • [2009/04/07 10:06] LathaSerevi: happy to hear any thoughts on how that might fit into the MMOX world.
  • [2009/04/07 10:07] Zha Ewry: The other bit of pain I've recently been tripping over, is the whole "We do C/M/T inside the "inventory entry" and trust that" stuff inside SL
  • [2009/04/07 10:07] Morgaine Dinova: I rather thought that Zero would be massively present at MMOX to move the design ahead. Instead, we have total absence. It's like during his 4-month AWOL last year, but worse.
  • [2009/04/07 10:07] Zha Ewry: That's hopelessly wonky, in a world where assets are seperately
  • [2009/04/07 10:08] Zha Ewry: held from inventory
  • [2009/04/07 10:08] Morgaine Dinova: Yeah, the SL model of inventories can't work for interop.
  • [2009/04/07 10:09] Zha Ewry: On several levels
  • [2009/04/07 10:09] Zha Ewry: I've been trying to find the time to sum up the "why" of it
  • [2009/04/07 10:09] Zha Ewry: (There are some really good reasons for SL to have done it) and the
  • [2009/04/07 10:09] Zha Ewry: "How could one get away from it"
  • [2009/04/07 10:10] Zha Ewry: And its not just at the easy to code or easy to save space on the asset cloud level
  • [2009/04/07 10:10] Morgaine Dinova: I rather think it's a lost cause Zha, too much legacy to change. Instead, I think it might be more productive to think of putting a layer or gateway in front of it, out of which pop clean objects for interop.
  • [2009/04/07 10:10] Zha Ewry: The niaive "Oh, make everything a seeperate URI" approach loses you a lot of chances to re-use a single texture in tons of places
  • [2009/04/07 10:11] Zha Ewry: (and a single ref, in places as well)
  • [2009/04/07 10:11] Zha Ewry: Well, the trick, is to first sort out some of the reasons it was done
  • [2009/04/07 10:11] Zha Ewry: and what you want to preserve
  • [2009/04/07 10:12] Zha Ewry: and.. then.. get the full model right.
  • [2009/04/07 10:12] Zha Ewry: I did a back of the enevelope doodle and found ways to easily increase by an order of magnitude the apparent unique items to be sent over the wire
  • [2009/04/07 10:12] Zha Ewry: by gettign this wrong
  • [2009/04/07 10:13] Zha Ewry: (You can cook up much better, and much worse cases too, of course)
  • [2009/04/07 10:13] Shamir Katsu: and the load on a reference monitor is higher if you do it wrong too. collapsing things that are the same saves a lot for all.
  • [2009/04/07 10:13] Latha Serevi: Shamir - in an OS, the reference monitor probably lives in the kernel. Where could it possibly live in MMOX, I wonder?
  • [2009/04/07 10:13] Zha Ewry: The current scheme is flawwed on lots of levels, btu it also has some wicked efficiency hiding in it
  • [2009/04/07 10:14] Zha Ewry: (and equally, misses some shots to be much more efficient)
  • [2009/04/07 10:14] Zha Ewry: frex, since all the beanbag chairs are copies of one of my items
  • [2009/04/07 10:15] Zha Ewry: The sim need only send 90% of its content once
  • [2009/04/07 10:15] Zha Ewry: Losing that would be painful
  • [2009/04/07 10:15] Zha Ewry: At the same time...
  • [2009/04/07 10:15] Teravus Ousley notes that the Images also consist of 90% of the network usage
  • [2009/04/07 10:15] Zha Ewry: If I had one which was the same, except for a tiny change, we have no way of leveraging that at all
  • [2009/04/07 10:15] Zha Ewry nods
  • [2009/04/07 10:15] Zha Ewry: Yeps
  • [2009/04/07 10:15] Zha Ewry: and system textures, and ground and so on
  • [2009/04/07 10:16] Zha Ewry: And.. increasingly sculpty aps
  • [2009/04/07 10:16] Zha Ewry: (and down the road, no doubt, meshes)
  • [2009/04/07 10:16] Latha Serevi: I wonder what huge fraction of those image transfers could be eliminated by a really good caching scheme.
  • [2009/04/07 10:17] Zha Ewry: Unclear.. I actually want to do some RL measurements of how many unique GBs of content you go past in a session on SL
  • [2009/04/07 10:17] Goldie Katsu: like how many copies of freebie textures are there that are each their own copy while being precisely identical.
  • [2009/04/07 10:17] Zha Ewry: Its easy to say "cache it all"
  • [2009/04/07 10:17] Zha Ewry: Well, all of those, have the same UUID
  • [2009/04/07 10:17] Zha Ewry: That's what's wicked efficient about Lidnen's scheme
  • [2009/04/07 10:17] Saijanai Kuhn: another thought to add to the mix: will http-texture actually save anything if its really https-texture?
  • [2009/04/07 10:18] Latha Serevi: Remedial SL dept: is there even any client-based control over which textures get streamed to me, or in SL do I get an extra copy of every texture that I have in my cache, once per visit to a sim?
  • [2009/04/07 10:18] Goldie Katsu: Do they? When I get it in a box and unpack it - is it still pointing to the same inode (or equivalent)
  • [2009/04/07 10:18] Zha Ewry: Yep
  • [2009/04/07 10:18] Zha Ewry: The inventory item, changes, but the UUID it points to doesn't.
  • [2009/04/07 10:19] Goldie Katsu: so it is only uploads that cause a problem.
  • [2009/04/07 10:19] Saijanai Kuhn: MichelleZ (I think) was saying that testing the current caching, it doesn't work for smaller textures
  • [2009/04/07 10:19] Shamir Katsu: so now we are trying to solve a meta-caching problem in addition to security during transfers?
  • [2009/04/07 10:19] Zha Ewry: They found a bug, which blocks any caching of tectures which are under 600 bytes
  • [2009/04/07 10:19] Morgaine Dinova: I'm a bit worried about the premature optimization that this seems to imply. How about making "replicable objects" part of the object transfer model instead, so that it's not so much an optimization as a clean design for objects?
  • [2009/04/07 10:19] Latha Serevi: (sorry, Shamir, we got distracted)
  • [2009/04/07 10:19] Zha Ewry: Mind you.. I'm not at all sure there are lots of those
  • [2009/04/07 10:19] Goldie Katsu wonders if boiling lakes is much faster than oceans
  • [2009/04/07 10:20] Zha Ewry: Not trying to optimize it, yet, Morgaine, just grok how it works well enough to not totally bork things by niaiavetee
  • [2009/04/07 10:20] Saijanai Kuhn: ah, did anyone note that Philip LInden and the http-texture team will be talking in Hippo Island at 11AM?
  • [2009/04/07 10:20] Zha Ewry: yes, and I intend to break before then
  • [2009/04/07 10:20] Zha Ewry: well, before so we can get seats
  • [2009/04/07 10:21] Latha Serevi: I need a pointer to remedial SL texture transfers 101. Anyone have a URL pretty please?
  • [2009/04/07 10:21] Morgaine Dinova: Let's all TP out on the dot at 1059.
  • [2009/04/07 10:21] Dahlia Trimble: so is the http-texture transfer over some secure channel?
  • [2009/04/07 10:21] Goldie Katsu: load testing?
  • [2009/04/07 10:21] Morgaine Dinova: Is it the same place as where we do bug triage, on Hippotropolis?
  • [2009/04/07 10:21] Zha Ewry: Yeah
  • [2009/04/07 10:22] Zha Ewry: got a lm?
  • [2009/04/07 10:22] Dahlia Trimble: or is it just map images sent over http?
  • [2009/04/07 10:22] Saijanai Kuhn: Latha, the SLDEV mailing list has discussion the past few days: LLTextureCache::writeToCache() does notcache textures smaller than TEXTURE_CACHE_ENTRY_SIZE
  • [2009/04/07 10:22] Morgaine Dinova: I have LM, will dish out
  • [2009/04/07 10:22] Teravus Ousley thinks they're answering those types of questions in 40 m
  • [2009/04/07 10:22] Zha Ewry: right
  • [2009/04/07 10:22] Teravus Ousley wants it t00 ㋡
  • [2009/04/07 10:22] Zha Ewry: the theory is that they
  • [2009/04/07 10:22] Dahlia Trimble: I wont be here in 40 m :/
  • [2009/04/07 10:22] Zha Ewry: will pass down stuff to the client
  • [2009/04/07 10:22] Zha Ewry: and it will http fetch them
  • [2009/04/07 10:22] Zha Ewry: (or https, not clear)
  • [2009/04/07 10:22] Teravus Ousley: I'll be in two places at once.. but.. meh .. so be it.
  • [2009/04/07 10:23] Saijanai Kuhn: michellz raised a point that https may not allow SQUID caching
  • [2009/04/07 10:23] Morgaine Dinova: Anyone else want LM?
  • [2009/04/07 10:23] Latha Serevi: Morgaine is laboriously passing out http://slurl.com/secondlife/Hippotropolis/242/30/23
  • [2009/04/07 10:23] Zha Ewry: You can.. get about 90% of the security benefits with caps, and just http, as guessing will be hard
  • [2009/04/07 10:23] Morgaine Dinova: Ta Latha
  • [2009/04/07 10:23] Dahlia Trimble: I know a few people who would be quite upset if textures were transferred over open http :(
  • [2009/04/07 10:23] Saijanai Kuhn: was thinking about that. https to pass out cap, then http to actually get texture
  • [2009/04/07 10:23] Zha Ewry: Well, if youc an't guess it
  • [2009/04/07 10:23] Saijanai Kuhn: could be a server-side decision anyway
  • [2009/04/07 10:24] Saijanai Kuhn: mark it "I don't care if my customers are POed at the slow downloads, send this https ALWAYS"
  • [2009/04/07 10:24] Saijanai Kuhn: and then make sure everyone knows who decided to do that
  • [2009/04/07 10:25] Zha Ewry: So..
  • [2009/04/07 10:25] Zha Ewry: The reason to hate open http, is because people could steal it
  • [2009/04/07 10:25] Zha Ewry: (gnoring OpenGL intercept)
  • [2009/04/07 10:25] Dahlia Trimble: sending all textures over open http is fine with me, as long as the people who implemented it dont mind if their bank accounts can be accesed and debited over open http
  • [2009/04/07 10:25] Morgaine Dinova: Horses for courses. Sending unencumbered assets over HTTPS would be a pointless waste of resources. Only use HTTPS to send encumbered ones.
  • [2009/04/07 10:26] Zha Ewry: If you use a secure cap to get a limited one time use http cap
  • [2009/04/07 10:26] Latha Serevi: Can anyone point me to the a page describing the SL mechanism for telling the texture server what to send me? That's the remedial bit I need.
  • [2009/04/07 10:26] Zha Ewry: that's pretty close to a non issue
  • [2009/04/07 10:26] Zha Ewry: That's basic deep sim behavior
  • [2009/04/07 10:26] Zha Ewry: The sim sends you all the tecxtures in your frustrum
  • [2009/04/07 10:26] Teravus Ousley: currently it gets requested via the client.
  • [2009/04/07 10:26] Zha Ewry: Actually
  • [2009/04/07 10:26] Zha Ewry: it does it via the client
  • [2009/04/07 10:26] Zha Ewry: so, it sends a list of the uuids
  • [2009/04/07 10:27] Teravus Ousley: it does it quite oddly also.
  • [2009/04/07 10:27] Teravus Ousley: ㋡
  • [2009/04/07 10:27] Zha Ewry: so, the client can peek the cache
  • [2009/04/07 10:27] Zha Ewry: (in theory)
  • [2009/04/07 10:27] Zha Ewry: and then ask for the ones it needs
  • [2009/04/07 10:27] Zha Ewry: all on UDP
  • [2009/04/07 10:27] Zha Ewry: And.. it all works sort of ok, until there is *any* spot on the pipe which gets to UDP congestion
  • [2009/04/07 10:27] Teravus Ousley: yep, with the discard quality layers and all
  • [2009/04/07 10:27] Zha Ewry: and.. then it all goes REALLY pearshaped
  • [2009/04/07 10:27] Latha Serevi: Thanks, that's the kind of thing I need. Still want a URL if anyone has one.
  • [2009/04/07 10:28] Zha Ewry: right, SL actually takes wicked clever advantage of jpeg2000 and incremental encoding
  • [2009/04/07 10:28] Zha Ewry: and, then there are tons of LOD bits hidingi n there
  • [2009/04/07 10:29] Zha Ewry: From a few sniffs of the stream that actually gets sent, a few of them probably are premature optimzations
  • [2009/04/07 10:29] Teravus Ousley: yep, according to some previous conversation, it uses 'pixel real-estate on screen' to prioritize
  • [2009/04/07 10:29] Zha Ewry: Right, the more of the texture is "visible" the higher the priority
  • [2009/04/07 10:30] Teravus Ousley: among other things.. like.. if your mouse is over it..
  • [2009/04/07 10:30] Zha Ewry: or touch it
  • [2009/04/07 10:30] Morgaine Dinova: All a bit of a red herring for interop. It goes without saying that that kind of sordid mess isn't going to end up in an IETF protocol :-)
  • [2009/04/07 10:30] Teravus Ousley: .. if you've selected it.. or if it's a body texture
  • [2009/04/07 10:30] Zha Ewry: whicSure
  • [2009/04/07 10:30] Teravus Ousley: or some part of the interface
  • [2009/04/07 10:30] Zha Ewry: sure
  • [2009/04/07 10:30] Latha Serevi: I currently use "texture voodoo" of that sort in addition to "draw distance 64m" ... because I still don't have any dsescription of what really happens. :-(
  • [2009/04/07 10:30] Zha Ewry: Heh
  • [2009/04/07 10:31] Teravus Ousley: the client gets a list of texture UUIDs from object updates
  • [2009/04/07 10:31] Zha Ewry: I *hope* over time to get a coherent spec for how a "camera" works in the abstract, but... its messy, and pretty client specififc
  • [2009/04/07 10:31] Teravus Ousley: .. the client then decides to get the texture or not.
  • [2009/04/07 10:31] Teravus Ousley: .. it starts by requesting it with a -1 discard level
  • [2009/04/07 10:31] Zha Ewry: right, at the protocol level, you would like to define 'Here's how you declare you camera, and what you want to hear about"
  • [2009/04/07 10:32] Zha Ewry: and then, let the client sip that stream, and decide what to request
  • [2009/04/07 10:32] Dahlia Trimble: thats a common optimization - using low res textures for distant objects
  • [2009/04/07 10:32] Zha Ewry: (with content negotiation, and all)
  • [2009/04/07 10:32] Teravus Ousley: .. after the first packet.. it reads the jp2 stream.. and uses that to determine what discard levels are available(speculation).
  • [2009/04/07 10:32] Zha Ewry: The current schme isn't all that odd
  • [2009/04/07 10:32] Zha Ewry: I think that that's correct, ter, it skims the header, which in JP2, let you know where the incremental encodings are
  • [2009/04/07 10:33] Teravus Ousley notes that the header is usually under 200 bytes.. and the first packet is 700
  • [2009/04/07 10:34] Teravus Ousley: .. so it contains the header.. and part of the lowest discard level.
  • [2009/04/07 10:34] Morgaine Dinova: Need to keep the two issues separate, agent position and camera. If a world disallows the camera moving away from the av, fine, that's policy. But the mechanism must keep them separate otherwise SL-type mobile cameras won't work.
  • [2009/04/07 10:34] Zha Ewry: http://www.jpeg.org/jpeg2000/CDs15444.html
  • [2009/04/07 10:34] Zha Ewry: Not only that, Morgaine, but there is no reasno to assume, at the "abstract spec" level that cameras always come with Aves at all
  • [2009/04/07 10:35] Morgaine Dinova: Sure, you have to request one. Eg. a text client won't be interested.
  • [2009/04/07 10:35] Zha Ewry: I for one, don't care for worlds where cameras can't be declared
  • [2009/04/07 10:35] Latha Serevi: Is there a separate "chat listener" position too, I wonder, or is that tied to the position of my avatar's hair?
  • [2009/04/07 10:35] Zha Ewry: At the moment, you can hear at your camera or ave
  • [2009/04/07 10:36] Zha Ewry: Actually, that;'s voice
  • [2009/04/07 10:36] Zha Ewry: is that true for text too?
  • [2009/04/07 10:36] Zha Ewry: hmm.
  • [2009/04/07 10:36] Teravus Ousley: no, not for text.. unless you've got a scripted object following your camera
  • [2009/04/07 10:36] Morgaine Dinova: Chat position is an "acoustic camera" :-) Looks like we'd better generalize the concept of "cameras" to include various forms of perception.
  • [2009/04/07 10:36] Zha Ewry: Right
  • [2009/04/07 10:36] Zha Ewry: I can fly a invis object to liste, but not my camera
  • [2009/04/07 10:36] Zha Ewry: and.. yeah, you really want "Stream listener"
  • [2009/04/07 10:36] Zha Ewry: and abstrct out
  • [2009/04/07 10:37] Morgaine Dinova: Yup
  • [2009/04/07 10:37] Zha Ewry: "Text, voice, objects"
  • [2009/04/07 10:37] Zha Ewry: Ditto speakers, btw
  • [2009/04/07 10:37] Zha Ewry: I for one, don't want worlds where those don't have an emodidied viewpoint, but.. that's probabl policy
  • [2009/04/07 10:37] Latha Serevi: Types of presence: 3-d avatar presentation; agent announced to others in sim; point "touched" by agent; camera position/orientation/zoom; text listening/output; voice listening/output...
  • [2009/04/07 10:37] Morgaine Dinova: Good point. A speaking person may actually have her voice coming out of PA speakers on a stage, instead of from her mouth. :-)
  • [2009/04/07 10:38] Zha Ewry nods at Latah
  • [2009/04/07 10:38] Zha Ewry: Or, a audio stream may actually be spaitalized
  • [2009/04/07 10:38] Morgaine Dinova: Ventriloquists are going to love that :P
  • [2009/04/07 10:38] Zha Ewry: The current "Oh, the music comes from the aether" is sort of odd
  • [2009/04/07 10:39] Teravus Ousley: yay for movie trailer type disembodied voices
  • [2009/04/07 10:39] Zha Ewry: That said, inivisbble viewpoints, are definitely not my policy choice
  • [2009/04/07 10:39] Zha Ewry: We'll end up witht hem
  • [2009/04/07 10:39] Zha Ewry: at the protocol level
  • [2009/04/07 10:39] Zha Ewry: (Machinema people will love us)
  • [2009/04/07 10:39] Dahlia Trimble: bye all :)
  • [2009/04/07 10:39] Zha Ewry: But, they have wonky social effects
  • [2009/04/07 10:40] Teravus Ousley: take care
  • [2009/04/07 10:40] Morgaine Dinova: See you Dahlia :-)
  • [2009/04/07 10:40] Zha Ewry: ta, Dahlia
  • [2009/04/07 10:40] Zha Ewry: Peeps don't like evesdroping objects, evesdroping cameras, would be even less liked
  • [2009/04/07 10:41] Latha Serevi: we love and hate 'em
  • [2009/04/07 10:41] Zha Ewry nods
  • [2009/04/07 10:41] Teravus Ousley wonders if it would be legal in some states if it made a tone every few minutes
  • [2009/04/07 10:41] Zha Ewry: OK, I'm going to hop to hippo land now
  • [2009/04/07 10:41] Zha Ewry: Heh.
  • [2009/04/07 10:41] Zha Ewry: Or Japan
  • [2009/04/07 10:41] Zha Ewry: Cameras have to click in Japan
  • [2009/04/07 10:41] Teravus Ousley: ㋡
  • [2009/04/07 10:41] Shamir Katsu: heh
  • [2009/04/07 10:42] Shamir Katsu: I should see what the texture discussion produces
  • [2009/04/07 10:42] Zha Ewry: okies
  • [2009/04/07 10:42] Shamir Katsu: the security issue at the protocol level is more interesting to me right now though
  • [2009/04/07 10:42] Zha Ewry: See you all at the land of the hippos
  • [2009/04/07 10:42] Shamir Katsu: but the caching problem will need to be solved someday
  • [2009/04/07 10:42] Zha Ewry: Oh, me too
  • [2009/04/07 10:42] Zha Ewry: and.. yeps
  • [2009/04/07 10:42] Morgaine Dinova: See ya there
  • [2009/04/07 10:42] Zha Ewry: Even with security pain
  • [2009/04/07 10:42] Teravus Ousley needs to keep track of it to ensure that OpenSimulator uses it as soon as possible! ㋡
  • [2009/04/07 10:42] Shamir Katsu: yup
  • [2009/04/07 10:43] Saijanai Kuhn: cutting off transcript
Retrieved from "https://wiki.secondlife.com/w/index.php?title=AW_Groupies/Chat_Logs/AWGroupies-2009-04-07&oldid=318472"