AW Groupies/Chat Logs/AWGroupies-2008-10-28

• [9:36] Saijanai Kuhn: Hey Zha, hey Terav
• [9:36] Zha Ewry: Morning all
• [9:37] lyndell Aleixandre: its 2:36 am where i am lol
• [9:37] Wrapp Seiling: gm
• [9:37] Zha Ewry takes the TP point hint, and moves the beanbag chair
• [9:38] Aimee Trescothick: lol
• [9:39] tx Oh is Online
• [9:39] Zha Ewry: So, I have a small topic today, and its something of a though experiment
• [9:40] Zha Ewry: and a quer from those who know better
• [9:40] Zha Ewry: *query
• [9:41] Zha Ewry: Right now, as far as I can tell, (its a wee bit hard to be sure, as its sort of behind the curtain)
• [9:41] Saijanai Kuhn: quer eye for the straight avatar?
• [9:41] Bartholomew Kleiber is Online
• [9:41] Zha Ewry: Linden keeps the permissions not on the assets at all, but in the inventory tree
• [9:41] Elric Ember: lol Sai
• [9:42] Zha Ewry: I, personally, think this is pretty odd, and very odd for a long term story, and also, pretty non restful
• [9:42] Force Prophecies 1.23 [script:Listen Handler]: Script run-time error
• [9:42] Force Prophecies 1.23 [script:Listen Handler]: Stack-Heap Collision
• [9:42] Saijanai Kuhn: right, there's an inventoyr DB and an asset DB sorta kinda
• [9:43] Zha Ewry nods
• [9:43] Zha Ewry: which means that, among othert things to properly use and item
• [9:43] Zha Ewry: you need TWO resources
• [9:44] Zha Ewry: which,, is certainly not in the spirit of REST
• [9:44] Zha Ewry: it also scales badly, and leads to funny behavior when you want to compose things diffeently
• [9:44] Latha Serevi: The two-part thing seems vaguely sensible to me; object and access-to-object. See your point about two resources tho.
• [9:44] Multi Gadget v2.0.3b by Timeless Prototype, '/44 info'
• [9:45] Tao Takashi is Online
• [9:45] Zha Ewry: Well, I think, three part you could argue
• [9:45] Zha Ewry: these two, tho, seem like an odd conflation
• [9:45] Zha Ewry: and.. I'd actually argue that, in most cases
• [9:45] Zha Ewry: the best place to enforce such behavior is at the asset server, or adhanced, not in the inventory server
• [9:45] Tao Takashi: Hi
• [9:46] Mircea Lobo: hello
• [9:46] Bartholomew Kleiber: Hi
• [9:46] Zha Ewry: Hey tao..
• [9:46] Canon Dench is Online
• [9:46] Zha Ewry: I am guessing that this split is pretty deeply baked into Linden's impl, but I'd like to think about whether OGP should enshrine it
• [9:46] Saijanai Kuhn: probably goes back to the direct connections between avatar inventory GUI and the relevant server
• [9:47] lyndell Aleixandre is Offline
• [9:47] Saijanai Kuhn: a folder signals an asset update event whenever its changed
• [9:47] Kopilo Hallard: seems odd to have two dbs...
• [9:47] Saijanai Kuhn: so that tree is defined server-side, not client-side
• [9:47] Zha Ewry: Well
• [9:48] Zha Ewry: Inventory, and asset reasonably need to be seperate
• [9:48] Zha Ewry: What's much less obvious is whichs hould hold the permissions info, and, in particular, enforce it
• [9:48] Zha Ewry: And. the client side is non-trusted
• [9:48] Zha Ewry: for which, small thanks
• [9:48] Latha Serevi: Hmm, if I'm an asset server, what do I want to see, in order to give someone an asset? Perhaps a very specific capability?
• [9:49] You decline The Mill at Muse Isle NorthWest, Muse Isle NorthWest (208, 206, from A group member named JenzZa Misfit.
• [9:49] Kopilo Hallard: user key?
• [9:49] Latha Serevi: no-o-o.
• [9:49] Tao Takashi: driver license ;-)
• [9:49] Zha Ewry: Well, to hand out the asset,probably depends
• [9:49] Geo Meek is Offline
• [9:49] Zha Ewry: If its copy for all
• [9:49] Zha Ewry: just hand it out
• [9:50] Zha Ewry: The questions chaneg as the policy for it changes
• [9:50] Mircea Lobo: I think permissions like copy and mod are not stored on the asset but on individual inventory entries, if im right
• [9:50] Zha Ewry: They are now
• [9:50] Geo Meek is Online
• [9:50] Zha Ewry: but that seems awfullly odd for an interop world
• [9:50] Saijanai Kuhn: Things might change on the LInden side if the share assets between all avatasr in an account thing (hyphons understood) goes into effect
• [9:51] Mircea Lobo: Also, son't forget there are backups like save-oar which must save a simulator with its in-use assets. It would be hard for it to know which it has permissions for or not
• [9:51] Zha Ewry: It totally breaks if for example, you want to allow multiple grids to share an asset, because you entangle some insanely deep trust in the wronfg place
• [9:51] Latha Serevi: Sure; what I'm getting at is, if we think about it from the capabiilty point of view, the asset server wants to see a "entity X may access object Y today" capability. And the question is, who sensibly manages giving out these capabilities. If I own a copy of the asset, I want to be able to create & hand out to my region the capabilty to access & rez one of my objects, for instance.
• [9:51] Zha Ewry nods
• [9:51] Zha Ewry: That totally breaks the current model, tho
• [9:52] Zha Ewry: afkish one minuite, phone
• [9:52] Kopilo Hallard: if I make an object no mod on one server, should it be no mod on all servers etc
• [9:52] Latha Serevi: I'm in the "first principles" mode, I thought that was Zha's gist today.
• [9:53] Mircea Lobo: As I said, currently I believe that no-copy no-mod no-trans is stored in invididual inventory entries towards an object
• [9:53] Saijanai Kuhn: Kopilo seems reasonable, but there may be situations where that might not apply.
• [9:53] Squirrel Wood is Online
• [9:53] Kopilo Hallard: *nods*
• [9:53] Kopilo Hallard: flexibility vs convience
• [9:54] Saijanai Kuhn: yeah, the OGP thing for inventory is suppoort the current LL model, support whatever LL decides to do int eh open grid AND support as manyther models as possible
• [9:54] Kopilo Hallard: ick...
• [9:54] Latha Serevi: That's a deep and basic question that Kopilo raises; can an object I own ever be changed-under-me by its creator? Usually not, but what if I wanted that to happen? Copy by reference or copy by value, can we really pick one right off the top? (maybe yes, copy by value, with some other way of achieving updates)
• [9:55] Saijanai Kuhn: well, one instance where this might make sense: change permissions to allow transfer to other grids
• [9:55] Zha Ewry: /]back
• [9:55] Kopilo Hallard: wb
• [9:55] Zha Ewry: And yes, first princples for a moment
• [9:56] Zha Ewry: And.. one more thing I am pretty sure about
• [9:56] Saijanai Kuhn: by default, its "no transfer," but if a content dcreator wanted to give permission to all current customers... could she do it?
• [9:56] Latha Serevi: So, if a creator "loosens" restrictions, it seems OK to promiscuously propagate that... maybe...
• [9:56] Zha Ewry: I think Linden share back end assets, between shared items, with only different inevtnory pointers, whic.. ick
• [9:56] Kopilo Hallard: *nods*
• [9:56] Zha Ewry: So...
• [9:56] Zha Ewry: Put asside, for am oment, the permissions details
• [9:56] Kopilo Hallard: everytime an object rezes in world it has a new uuid
• [9:56] Zha Ewry: assume we'll have a ton of variations over time
• [9:57] Latha Serevi: Not "ick" at all, that's basic "shared library" stuff, right? As long as it's constant, I love sharing back end assets.
• [9:57] Zha Ewry: Well.. Yes and no
• [9:57] Zha Ewry: because, it makes it very hard to so something properly
• [9:57] Zha Ewry: *some things
• [9:57] Zha Ewry: In particular, if its only "qausi" shared
• [9:58] Zha Ewry: which much of this is
• [9:58] Latha Serevi: Right, so your point is, a lot of things we want to do, are hard to do using efficient shared guaranteed-constant assets. But we don't want to shoot ourselves in the foot by requiring promiscuous copying of library objects.
• [9:58] Zha Ewry: you end up losing things like ratoinal use counts, and such, and as you move from UUIDs to URLs, the meanig of the URL gets messy
• [9:58] Zha Ewry: Well, thus the lets discuss and think
• [9:58] Latha Serevi: I guess copy-on-write is an option.
• [9:59] Zha Ewry: becuase, yes.. if you don't pay attention, ina world where, in fact 95% of the content is overlapping
• [9:59] Zha Ewry: you lose a lot of efficiency
• [9:59] Zha Ewry: At the same time
• [9:59] Zha Ewry: do note that the web
• [9:59] Zha Ewry: doe snot treat http://zha.com/pics/jpeg1 and http://dwl.com/pics/jpeg1
• [9:59] Zha Ewry: as the same, even if the underlying picture is the same
• [9:59] Latha Serevi: Excellent point about the web. Hmm.
• [10:00] Latha Serevi: Advantage of p2p vs sharing?
• [10:00] Kopilo Hallard: heh
• [10:00] Zha Ewry: I *think* I take the position that storage, is dirt cheap, programmign complexity costs money, but I tyake it weakly
• [10:00] Saijanai Kuhn: can't you set things on the website to refer to other image though?
• [10:00] Elric Ember: Does that also introduce the need to resolve permission conflicts between asset servers?
• [10:01] Zha Ewry: welll
• [10:01] Zha Ewry: Chuckle
• [10:01] Aric Linden is Online
• [10:01] Zha Ewry: which NxM problem do you want to solve?
• [10:01] Zha Ewry: right now
• [10:01] Simon Sugita is Offline
• [10:01] Latha Serevi: Hmm, the web isn't so great at permissions, is it. How would I pass along the capability to access a given resource, web wise.
• [10:01] Zha Ewry: We're deeply tangling up permissions schemes and inventory schemes
• [10:01] Kopilo Hallard: a many to many problem has a simple solution :p
• [10:01] Simon Sugita is Online
• [10:02] Zha Ewry: which, by the way, means, its incredibly hard for OpenSim to have a different strtagey
• [10:02] Zha Ewry: So..
• [10:02] Zha Ewry: In the web?
• [10:02] Zha Ewry: The http get may require a token, or a session, or such like
• [10:02] Zha Ewry: password to get at the asset, and so on
• [10:02] Latha Serevi: Yes, I like the idea of separating out inventory from permissions. It's the network of trus/permissions that needs managing; inventory feels like just an implementation issue.
• [10:02] Zha Ewry: its not exported at all, tho, in general
• [10:03] Zha Ewry: no geenral scheme for finding out what's needed to look at the page
• [10:03] Zha Ewry: (and.. we can add for grins, the firewall issues, but ick)
• [10:03] Latha Serevi: What if we made a scheme that's super-efficient for full-perms assets?
• [10:03] Latha Serevi: I like that it'd encourage folks to make stuff full perms.
• [10:04] Latha Serevi: So, for isntance, instead of always having to contact two resources, an asset can be labeled "no capability required".
• [10:04] Kopilo Hallard: and exclude a good preportion of pre-existing content?
• [10:04] Latha Serevi: Pre-existing content is already totally excluded, sorry. LL will never let it go.
• [10:04] Zha Ewry: I think thats' right
• [10:04] Zha Ewry: If you happily make it easy to just fetch open assets that's a ood thing
• [10:05] Zha Ewry: Not the only way to exist
• [10:05] Zha Ewry: but
• [10:05] Saijanai Kuhn once considered buying an ood, but the differences between that and a lute aren't apparent
• [10:05] Zha Ewry: The "no permssions on this object" case ought to not require the full permissiosn sustem
• [10:05] Tao Takashi: sorry, was sidetracked here in the office
• [10:05] Latha Serevi: wow, Sai, that was very zen.
• [10:06] Zha Ewry flips the switch and shunts Tao back onto the mainlien
• [10:06] Tao Takashi: as for can an object be changed when I own it: I would definitely say: never
• [10:06] Zha Ewry: By someone else?
• [10:06] Tao Takashi: this will break trust
• [10:06] Zha Ewry: Well. own, vs. license
• [10:06] Tao Takashi: also by the creator
• [10:06] Saijanai Kuhn owns a 18 string theorbo lute but its been broken for years :-(
• [10:06] Tao Takashi: people always think they own it
• [10:06] Zha Ewry: If I fully own, it tho, yeps, nobody else gets to touch it
• [10:06] Latha Serevi: I'm lost.
• [10:06] Tess Linden is Online
• [10:06] Zha Ewry: well, I routinely use stuff that's only licensed, IRL. but its an interesting point
• [10:07] Saijanai Kuhn: sorry, was channeling fword for a bit
• [10:07] Latha Serevi: Zha, what does this mean? --> The "no permssions on this object" case ought to not require the full permissiosn sustem
• [10:07] Latha Serevi: Oh, I guess you were agreeing with me.
• [10:07] Zha Ewry: I think so. Latah
• [10:07] Zha Ewry: if an item has "no permissinos on it"
• [10:07] Zha Ewry: it would be nice
• [10:07] Zha Ewry: if I didn't have to go thought a dance to find that out
• [10:07] Tao Takashi: I think one of the things I want to be sure of is that I know exactly what permissions I have with an object when I acquire it
• [10:07] Saijanai Kuhn: so full permissions implies "creative comons" tpe license?
• [10:08] Latha Serevi: Tao, you'd like the option to receive software updates, though, right? How will you do those?
• [10:08] Kopilo Hallard: full permissions implies full permissions xD
• [10:08] Tess Linden is Offline
• [10:08] Latha Serevi: (btw, I agree with you )
• [10:08] Tao Takashi: I get a new object and I signup to some list?
• [10:08] Latha Serevi: Tao, you see no use for copy-by-reference w.r.t. assets?
• [10:08] Tao Takashi: but that's up to the creator I think what he offers additionally
• [10:09] Tao Takashi: I think this makes it more difficulty
• [10:09] Tao Takashi: it's like a DRM server then
• [10:09] Tao Takashi: like "plays for sure" ;-)
• [10:09] Tao Takashi: plays for sure until we shut the server down
• [10:09] Latha Serevi: Not a DRM server, just allowing for variable contents rather than constant. It's a big deal, though, for instance, what about caching.
• [10:10] Zha Ewry: very much so
• [10:10] Tao Takashi: well but like the drama around DRM servers
• [10:10] joao Mastroianni is Online
• [10:10] Tao Takashi: I also wouldn't like updates just happening
• [10:10] Tao Takashi: I want to be informed about them and agree to them
• [10:10] Zha Ewry: +1 tao
• [10:10] Tao Takashi: maybe it's not an upgrade but a downgrade (maybe just in my opinion)
• [10:10] Tao Takashi: so I basically want a new object
• [10:11] Latha Serevi: Except if the creator added an "open grid OK" permission, right? ;-)
• [10:11] Tao Takashi: what does that mean?
• [10:11] Zha Ewry: yeah, I hate when someone breaks a proghram with an update and I don't have a working copy anymore
• [10:11] Tao Takashi: or even revises the UI to make it "better" ;-)
• [10:11] Kopilo Hallard: >_> most object actually update by rezing a new object
• [10:12] Kopilo Hallard: scripts*
• [10:12] Latha Serevi: So, strong consensus for keeping objects constant/immutable. I like it pretty well.
• [10:12] Dizzy Banjo is Online
• [10:12] Latha Serevi: But we'll have to think about how to manage change in a non-hacked way.
• [10:12] Bartholomew Kleiber: the copy by reference would basically having a rererence in the invenotory not the object
• [10:12] Bartholomew Kleiber: so that would still be possible
• [10:12] Zha Ewry: if we change that, the objects change --> new UUDI/URL we're breaking a ton o ftuss including REST
• [10:12] Bartholomew Kleiber: it gets the object while rezzing
• [10:13] Keex Rexroth is Offline
• [10:13] Latha Serevi: So, how do we define a pointer to "the latest blog entry"? Can I pass it to my friend? Do I need a new capability each time a blog entry is posted?
• [10:13] Kopilo Hallard: for preformance: transfer once, rez many times ?
• [10:13] Bartholomew Kleiber: ('invenotory' ... what a typo ...)
• [10:13] Saijanai Kuhn: latest blog entry is permanent cap, I'd think
• [10:13] Zha Ewry thinks
• [10:13] Zha Ewry: So.. you're poking at two issues
• [10:14] Zha Ewry: one is the ocntsant ref to a
• [10:14] Zha Ewry: Hmm
• [10:14] Zha Ewry: OK
• [10:14] Zha Ewry: See. Asset? or place?
• [10:14] Zha Ewry: A blog to me feels like place, not an asset
• [10:14] Canon Dench is Offline
• [10:14] Tao Takashi: lates blog entry can be a redirect ;-)
• [10:14] Zha Ewry nods
• [10:14] KK Jewell is Offline
• [10:15] Teravus Ousley: Would you say that images referenced in web pages are both an asset and a place?
• [10:15] Latha Serevi: The primary issues you're discussing, Zha, management and access to permissions (c apabilities to access), we can probably consider assuming the constant asset case as the primary one.
• [10:15] Zha Ewry: Good, and hard question
• [10:15] Zha Ewry: I think we havr to Latah
• [10:15] Tao Takashi: well, latest blog entry is simply an indirection
• [10:15] Latha Serevi: But yes, I was bring up the secondary issue that the world is not constant, and we need to manage capabilities that generalize to changed objects.
• [10:15] Zha Ewry: So.. one challange, in the VW space is making sure poeple see changes
• [10:16] Zha Ewry: we don't have that issue in web pages
• [10:16] Zha Ewry: nobody cares if they miss an udpate
• [10:16] Zha Ewry: here we do, quite a it
• [10:16] Tao Takashi: what changes for instance?
• [10:16] Saijanai Kuhn: E.G. official Linden Lab Logo texture...
• [10:16] Latha Serevi: Events listing.
• [10:16] Teravus Ousley: haha, newsgroup servers ftw
• [10:16] Bartholomew Kleiber: yes, you have: see google outdated links
• [10:16] Plato Pizzicato is Online
• [10:16] Tao Takashi: events listing is a web page ;-)
• [10:16] Zha Ewry: LL does that when you have scripted changign obejct
• [10:16] Zha Ewry: New UUIDs all the time
• [10:16] Zha Ewry: so every client sees them all, an fetched them all
• [10:16] Kopilo Hallard: x_X
• [10:17] Tao Takashi: ok, but that seems to me an implementation issues with the multiple UUIDs.
• [10:17] Teravus Ousley: we'll have the Alt.Binaries.AssetHoarders or something :D
• [10:17] Zha Ewry: Well
• [10:17] Zha Ewry: Its interesting
• [10:17] Tao Takashi: when a client comes to a sim it gets a list of objects with their URLs which it fetches
• [10:17] Zha Ewry: the "every changed version" matters approach goes oddly with a REST and URLs
• [10:17] Tao Takashi: then some cache header might define how often it should check back
• [10:18] Bartholomew Kleiber: around 2000 there was a concept called things ... which was the notion that certain objects in the web exist only once and cannot be copied. That concept went down the drain ....
• [10:18] Tao Takashi: of course if it changes while being on the sim some additional mechanism needs to be used to inform the client about it
• [10:18] Zha Ewry: That's the real challange Tao
• [10:18] Tao Takashi: some invalidation service
• [10:18] Zha Ewry: if you look at some prim ave parts
• [10:18] Zha Ewry: they change at a pace that is scary
• [10:18] Tao Takashi: how is it done now?
• [10:18] Zha Ewry: Like every eyeblink
• [10:18] Zha Ewry: new UUID
• [10:18] Tao Takashi: ok, but how does the client know about the new UUID?
• [10:18] Zha Ewry: or, I think in some cases
• [10:18] Zha Ewry: new temp reference
• [10:18] Kopilo Hallard: *nods*
• [10:19] Zha Ewry: it gets pushed down as aon object update
• [10:19] Zha Ewry: *as an
• [10:19] Saijanai Kuhn: ick
• [10:19] Tao Takashi: ok, so you can do the same with URLs
• [10:19] Elric Ember: wow
• [10:19] Tao Takashi: "reload xyz, please"
• [10:19] Saijanai Kuhn: object update is ugly tihing to use for constant changes
• [10:19] Zha Ewry: How else would you do it?
• [10:19] Zha Ewry: it may be just a new texture
• [10:19] Zha Ewry: or a totally new set of prim parms
• [10:19] Zha Ewry: The sim rarely knows, it just sends "new object"
• [10:19] Tao Takashi: yes, but isn't the important part mainly to inform the client?
• [10:20] Latha Serevi: "update xyz's parameters to these, please" ? Hmm.
• [10:20] Saijanai Kuhn just saying. Perhaps partial updates packet could be defined
• [10:20] Zha Ewry: Again, its is fine grain vs coars grained
• [10:20] Zha Ewry: In thew web world
• [10:20] Zha Ewry: we'd just say "oh, they'll get the latest"
• [10:20] Zha Ewry: in a consnesus reality, you want people to see as many (ideally all) of the intermediate states
• [10:20] Tao Takashi: I don't think it's that important if the UUID has changed or the object behind an existing UUID as long as you inform the client either by telling "fetch xyz again" or directly pushing it to the client
• [10:20] Teravus Ousley: ahh, the soft stuff or the hard stuff....
• [10:20] Saijanai Kuhn: BIG packet: http://wiki.secondlife.com/wiki/ObjectUpdate
• [10:21] Enus Linden is Offline
• [10:21] Tao Takashi: you maybe even just have to send diffs
• [10:21] Zha Ewry: well, the problem is fech XYZ in web terms has very specific meaning
• [10:21] Latha Serevi: BTW, do we have a clear line between the constant-part of my flexy hair that lives on an asset server, and the variable part (position/orientation/color?/parameters) that are managed/streamed by a region?
• [10:21] Zha Ewry: http get on a url
• [10:21] Tao Takashi: sure
• [10:21] Zha Ewry: only the client
• [10:21] Saijanai Kuhn: Latha see that URL
• [10:21] Tao Takashi: I guess that would be slow
• [10:21] Zha Ewry: Fleixies aren't done on the server
• [10:22] Tao Takashi: well, the parameters are sent from the server
• [10:22] Saijanai Kuhn: definition is psased in though
• [10:22] Kopilo Hallard: flexies, omegatargets are client based
• [10:22] Teravus Ousley: additionally there's a distinction between inventory and objects 'in world'
• [10:22] Kopilo Hallard: same with particles
• [10:22] Tao Takashi: but with flexies I guess not everybody sees exactly the same at some point in time
• [10:22] Zha Ewry: At all, Tao
• [10:22] Zha Ewry: not flexies, nor anims, nor gestures
• [10:22] Zha Ewry: all are handled asynchrounsoly, on the client
• [10:22] Latha Serevi: Sorry, please replace flexy by linked-object-positions or something.
• [10:22] Saijanai Kuhn: though, you can attempt to synch animations for line dancing...
• [10:22] Zha Ewry: (thus the need for the sync messages on dance balls)
• [10:22] Kopilo Hallard: *giggles*
• [10:22] Tao Takashi: but that's another topic anyway.. so I think the URL of an object (which includes maybe the UUID) can stay the same as long as the client is informed regularly how it looks now
• [10:23] Zha Ewry: Not sure, Tao, you have to make sure HTTP get and caching and such doesn't subvert your intent
• [10:23] Tao Takashi: this can be done by pushing a complete object description to it, just a diff or an URL which it has to GET
• [10:23] Tao Takashi: sure but the server has control over the caching headers
• [10:23] Kopilo Hallard: *thinks of the network traffic and runs off screaming*
• [10:23] Tao Takashi: it's the same network traffic if not GET is used
• [10:23] Tao Takashi: might be less if diffs are used
• [10:23] Zha Ewry: Its bad, aleady koplo
• [10:24] Kopilo Hallard: exactly
• [10:24] Kopilo Hallard: lol
• [10:24] Zha Ewry: I mean this doesn't make it worse
• [10:24] Zha Ewry: and... diffs would help a fair bit
• [10:24] Teravus Ousley: though 90% of it is images going down the pipe
• [10:24] Zha Ewry: Depends on the scenegraph
• [10:24] Saijanai Kuhn: taravus, except for the ObjectUpdates, which is everything BUT images
• [10:24] Tao Takashi: but the URL can still be the same and the next client which loads it initially can still use GET and get the uptodate version at this point in time
• [10:25] Zha Ewry: for example, if its a prim which cycles three textures
• [10:25] Tao Takashi: if just a texture changes maybe just push the texture attribute to the client
• [10:25] Zha Ewry: you may get UUIDs contiunaully
• [10:25] Kopilo Hallard: just checking the url points to a grid server?
• [10:25] Zha Ewry: but they will all reference the same tecture
• [10:25] Zha Ewry: which the client *should* keep in its cache
• [10:25] Teravus Ousley: well, average usage.. means the clients get pushed many many images.
• [10:25] Tao Takashi: well, the texture is just like a hyperlink
• [10:25] Saijanai Kuhn: goes back to sculpty animation issue.
• [10:25] Tao Takashi: the texture itself can be cached separately with normal HTTP Cache headers
• [10:25] Kopilo Hallard: sorry inventory server
• [10:25] Zha Ewry nods at Teravus
• [10:25] Zha Ewry: and, textures should cash very well
• [10:25] Tao Takashi: yep
• [10:25] Zha Ewry: *cache, even
• [10:25] Zha Ewry: sheesh
• [10:26] Zha Ewry: They don't. btw, which bother me a lot
• [10:26] Teravus Ousley: http://teravus.wmcv.com/googletester/IdealistViewer5.jpg
• [10:26] Kopilo Hallard: just thinking if there would be bottleneck issues
• [10:27] Zha Ewry nods
• [10:27] Tao Takashi: btw, you can use Windows Live ID now as an Open ID, just as some unrelated news :)
• [10:27] Zha Ewry: So..right now, the major flow is that the textures come down, in the long queue
• [10:27] Zha Ewry: or on UDP
• [10:27] Zha Ewry: and, in theory they cac he
• [10:27] Teravus Ousley: UDP
• [10:27] Zha Ewry: but if you look at the flow
• [10:27] Zha Ewry: you'll see
• [10:28] Zha Ewry: that at times
• [10:28] Teravus Ousley: in addition, they're prirotized by the client
• [10:28] Tao Takashi: I am still waiting for some textures here ;-)
• [10:28] Zha Ewry: 90% of the UDP traffic is re-sennds of incomplete textures
• [10:28] Tara5 Oh is Offline
• [10:28] Teravus Ousley: .. and the client continuously re-prioritizes the images based on what's in your view at the moment
• [10:28] Tao Takashi: so did LL test how it would work if the client GETs them via HTTP?
• [10:28] Zha Ewry: They have not
• [10:28] Zha Ewry: which is kind of
• [10:28] Zha Ewry sighs
• [10:28] Teravus Ousley: .. additionally, the client requests specific 'levels' of graphivs
• [10:28] Zha Ewry nods at Teravus
• [10:29] Zha Ewry: right, based on among other things the jpg2k progressive stuff
• [10:29] Bjorlyn Loon is Offline
• [10:29] Tao Takashi: so Zha, you have influence at LL, tell them to test the HTTP stuff and to come back next week with some numbers ;-)
• [10:29] Zha Ewry: On the whole, a wicked buttload of complexity for something which often seems to fail to do its job at all well
• [10:29] Zha Ewry sighs
• [10:29] Bartholomew Kleiber: gotta run unfortunately - laters.
• [10:30] Tao Takashi: cya Barth!
• [10:30] Zha Ewry: Tao, I've gently prodded several engineers on this topic for weeks
• [10:30] Zha Ewry: cya Barth
• [10:30] Zha Ewry: "Do we know how well it will work"
• [10:30] Teravus Ousley: I dunno. I think textures are probably the prime target for http optimization.
• [10:30] Bartholomew Kleiber is Offline
• [10:30] Zha Ewry: "Can we measur eit"
• [10:30] Tao Takashi: but actually this also could be done without LL I guess, by patching the viewer and opensim
• [10:30] Zha Ewry nods
• [10:30] Misti Merryman is Online
• [10:30] Zha Ewry: We can so some of it
• [10:30] Zha Ewry: But..
• [10:30] Zha Ewry: Interestingly enough
• [10:30] Zha Ewry: when I measure things?
• [10:31] Zha Ewry: Opensim does a very much bettter job at it
• [10:31] Zha Ewry: AFK a minute, phone
• [10:31] Tao Takashi: yeah, I guessed that maybe those results might not apply to LL's grid
• [10:31] Tao Takashi: so they maybe should consider using opensim
• [10:31] Tao Takashi: ;-)
• [10:32] Tao Takashi: actually they should maybe consider to write a region from scratch ;-)
• [10:33] Kopilo Hallard: *symphony of criclets*
• [10:33] Kopilo Hallard: crickets*
• [10:33] Misti Merryman is Offline
• [10:33] Sidewinder Linden is Online
• [10:34] Tao Takashi: using opensim wouldn't be LL's thing though, they are not about reusing existing things ;-)
• [10:34] Simon Sugita is Offline
• [10:34] Kopilo Hallard: lol
• [10:34] Kopilo Hallard: idk
• [10:35] Tao Takashi: they are the mother of reinvention :-)
• [10:35] Kopilo Hallard: have you read some of the latest blog entries...
• [10:35] Tao Takashi: well, actually in this case opensim is a reinvention (but for obvious reasons)
• [10:35] Zha Ewry chuckles
• [10:35] Zha Ewry: its a mix
• [10:35] Teravus Ousley: heh, in all reality.. I'm not sure opensim had the resource to test it on a 65,000 concurrent user scale :D
• [10:35] Tess Linden is Online
• [10:35] Tao Takashi: only one way to find out! :)
• [10:35] Teravus Ousley: It would be a fun April fools joke though.
• [10:35] Simon Sugita is Online
• [10:35] Saijanai Kuhn: this talk about testing new protocols comes back to pyogp issue/argument that Tao and I have. Implementing existing or alternative group IM within pyogp would be trivial, BUT there's no defined way to send data back to the AD via caps, and no defined data discovery method for optional stuff
• [10:36] Tao Takashi: you think LL had just before the hype? ;-)
• [10:36] Aric Linden is Offline
• [10:36] Saijanai Kuhn: Taravus, sure you could. 65K users easy to fake using pyogp
• [10:36] Tao Takashi: sure, only one laptop is needed ;-)
• [10:36] Zha Ewry: Back
• [10:36] Saijanai Kuhn: in theory. Minor details to be worked out... ;-)
• [10:37] Teravus Ousley: well, except you'd also need to request all of the textures in a 256m radius for each of the clients..
• [10:37] Zha Ewry: Theory is better than practice in theory. Practice is better than theory in Praciice
• [10:37] Tao Takashi gets his Kant out
• [10:37] Aimee Trescothick: great quote
• [10:38] Jonit Ivory is Online
• [10:38] Aimee Trescothick: :)
• [10:38] Wrapp Seiling: eh?
• [10:38] Tao Takashi: now that we become philosophical :)
• [10:38] Zha Ewry chuckles a little
• [10:38] Zha Ewry: Anyway..
• [10:38] Teravus Ousley: It's a lot of stuff.. I tried to do it in a viewer.. for a sim with ~5000 prim and I got to about 4000 before it ran out of memory : http://teravus.wmcv.com/googletester/IdealistViewer7.jpg some of you might recognize the simulator.
• [10:38] Zha Ewry: I would love to actually see some coherent testing of the changes
• [10:39] Jonit Ivory is Offline
• [10:39] Zha Ewry: wright you are?
• [10:39] Saijanai Kuhn: So, its a genuine issue for us right now: there's no outgoing pie defined for UDP packets via CAP, and there's no discovery mechanims defined so Tao or I coulod add our own group IM server to an AD and let the client know baout it
• [10:39] Sidewinder Linden is Offline
• [10:39] Teravus Ousley: wright indeed
• [10:39] Tao Takashi: I defined one in OGP alternative ;-)
• [10:39] Zha Ewry chuckles at Tao
• [10:39] Tao Takashi: at least the start of one
• [10:39] Sidewinder Linden is Online
• [10:39] Latha Serevi: We're probably gonna be winding down about now, but Zha, I look forward to the next installment of the assets/permissions/inventory discussion.
• [10:39] Zha Ewry: I'd like to see if we can avoid forking the spec, but I do take the point
• [10:40] Zha Ewry: I do too
• [10:40] Zha Ewry: I think that this was enough to send me back to think some more
• [10:40] Tao Takashi: it's not really forking, more an experiment if things can be done with more "web" in it
• [10:40] Zha Ewry: and I expect all of you as well
• [10:40] Zha Ewry: Which, is often the goal
• [10:40] Zha Ewry: to get us all thinking and get a good synthesis
• [10:40] Tao Takashi: and in order to experiment it I need to write something down
• [10:40] Zha Ewry: Yes, I agree Tao, and I think we need more of that
• [10:41] Tao Takashi: and should I have time I might implement some IM integration with some mockup server/client
• [10:41] Zha Ewry: Far too much of this is ending up top down from Linden
• [10:41] Tao Takashi: and besides this, it can be reused for data portability
• [10:41] Tao Takashi: just keep the region stuff out
• [10:41] Saijanai Kuhn: all we need is service discovery mechanism and an outgoing pipe definition, I think
• [10:41] Tao Takashi: because the web will not adopt ADs
• [10:42] Tao Takashi: I also signed up for a myspace dev acount and a yahoo one to test their APIs
• [10:42] Tao Takashi: and see how they use OAuth, OpenSocial etc.
• [10:42] Latha Serevi: Tao, remind me what's the URL to your spec-writing stuff? Anybody else have URLs to mention? Thx.
• [10:42] Tao Takashi: and in November there is an identitycamp in Mountain VIew, LL should go there and meet some people :)
• [10:43] Tao Takashi: http://opengridprotocol.info/alternative
• [10:43] Tao Takashi: but it hasn't changed yet
• [10:43] Saijanai Kuhn just bangs on teh keyboard randomly 'til the monkeys say something intelligible
• [10:44] Zha Ewry: OK, and I think we'll call this a wrap.. and thanks for all the really good dsicussion today
• [10:44] Zha Ewry: Saij? You going to post a transcript?
• [10:44] Saijanai Kuhn: sure
• [10:44] Zha Ewry: Super