User:Which Linden/Office Hours/2009 Apr 16

[11:04] Which Linden: Good morning, Techwolf and Sai
[11:04] Morgaine Dinova: 'Morning Which, Sai, Tech :-)
[11:04] Morgaine Dinova: Haha
[11:04] Which Linden: hi morgaine
[11:04] Techwolf Lupindo: Not many here today
[11:05] Techwolf Lupindo: lol
[11:05] Morgaine Dinova: Hey Which, read these two lines in the AW Groupies chat log frm overnight:
[11:05] Which Linden: :-)
[11:05] Morgaine Dinova: [2009/04/16 9:10
[2009/04/16 9:10] Linden: yes: we are working on that.
[11:05] Techwolf Lupindo: Love the avatar Which.
[11:05] Which Linden: Thanks.
[11:06] Which Linden: Morgaine: hah well philip would know more than I
[11:06] Techwolf Lupindo: I had to use my werewolf tracking skills to find you. :-)
[11:06] Saijanai Kuhn: hey all
[11:06] Morgaine Dinova: Hiya Sai!
[11:06] Morgaine Dinova: Sai, did you read Philip's comment above?
[11:06] Which Linden: should get a seat that's hidden in the branches of maximum hideability
[11:06] Saijanai Kuhn: yeah, and I noticed he shut up as soon as you jumped on it
[11:07] Morgaine Dinova: I was asleep, but he didn't answer those who asked more.
[11:07] Which Linden: It's amazing to me that you log the chat while you sleep, bTW
[11:08] Morgaine Dinova: My machines are always on. That's what broadband is for :-)
[11:08] Yann Dufaux: hello :)
[11:08] Morgaine Dinova: Hi Yann
[11:09] Which Linden: ha ha ha
[11:09] Which Linden: hi yann
[11:10] Yann Dufaux: bambooWhich :)
[11:10] Yann Dufaux: same of beta grid:)
[11:10] Saijanai Kuhn: the nicest part of Philip speaking up is that he DID speak up in Groupies
[11:10] Which Linden: It's great to see him around more
[11:10] Morgaine Dinova: Yeah, that's cool
[11:11] Which Linden: I, by the way, am extremely happy with the way the http-texture viewer development is going ... that's how we should have done it from the start
[11:11] Saijanai Kuhn: very kool
[11:11] Which Linden: But I wasn't a viewer dev at the time so I didn't push it that way
[11:11] Saijanai Kuhn: makes things much more flexible in the long run, too, I suspect
[11:11] Techwolf Lupindo: Its nice to see someone high up paying attenction to there company core product.
[11:11] Saijanai Kuhn: you can create more than one channel to prioritize textures by size or urgency for example
[11:12] Which Linden: hm yes
[11:13] Saijanai Kuhn: while that massive 4mb texture is downloading and being parsed, you can still receive a dozen lsmaller textures and process them
[11:13] Which Linden: can't we do that with the current udp system now as well?
[11:13] Saijanai Kuhn: hmmmm
[11:14] Saijanai Kuhn: Not sure exactly. There's only one socket per sim and there's no priority as far as I know
[11:15] Which Linden: well but it can deliver packets out of order
[11:15] Techwolf Lupindo: UDP is good for state data, like in a FPS when the data is outdated in seconds and therefore drop data is of no consern. TCP is good for moving data intact. Whoever desided to use UDP for data transfer was out of there mind in my opion.
[11:15] Which Linden: Ha ha well it probably wasn't a conscious decision, more of a "hey we have this transport mechanism, let's use it"
[11:16] Which Linden: notes for the nth time that this explains many design decisions at many companies
[11:16] Morgaine Dinova: Techwolf: UDP is not good for state data. UDP is not good for any data that must be transferred reliably.
[11:17] Saijanai Kuhn: avatar movement keys FTW
[11:17] Morgaine Dinova: Only for intersticial avatar movement, not for mandatory movement.
[11:17] Morgaine Dinova: Ie. fine-grain movement in between the mandatory position updates.
[11:18] Techwolf Lupindo: I may have my terms messed up. I was refering to data that expires quickly.
[11:18] Techwolf Lupindo: I"me not perfet. :-)
[11:19] Which Linden: :-)
[11:19] Morgaine Dinova: Nope, it's not good for data that expires quickly even. If data *has to arrive reliably* but expires quickly, then UDP is not good enough.
[11:19] Saijanai Kuhn: I'm not a parfait either
[11:19] Which Linden: so, I have a bit to continue discussing w.r.t last week's discussion on group chat architecture
[11:19] Morgaine Dinova: Super!
[11:19] Which Linden: as you may recall, I laid out 4 top-level architectures
[11:19] Saijanai Kuhn: sounds good
[11:20] Which Linden: we've begun looking at architecture 2 again
[11:20] Which Linden: 2 is the one where each client listens to a particular server, and every message sent to the system gets sent to all servers
[11:20] Which Linden: the servers act sort of as filters so the clients only get the messages they desire
[11:21] Which Linden: we noticed that our base message rate is kinda low, low enough that we might not have to worry about scalability along that axis for a while
[11:22] Which Linden: so....we're investigating whether that would actually work for us
[11:22] Which Linden: basically the base setup would be: each user gets a queue on a designated server
[11:22] Which Linden: there's about N servers, each of which is not part of a cluster
[11:23] Which Linden: each server subscribes to a "broadcast" server
[11:23] Which Linden: all incoming messages go into the broadcast server and get replicated to all user severs
[11:23] Which Linden: and...that's it
[11:24] Techwolf Lupindo: digests it all...
[11:24] Which Linden: since our base message rate is around 10 msg/sec, and these servers can frequently handle 10,000 msgs/sec, we might have enough orders of magnitude that we don't have to worry about overwhelming the broadcast server for a while
[11:24] Which Linden: testing that hypothesis is the next order of business
[11:25] Techwolf Lupindo: I wasn't here last time. What messages are you refering too? IM, group chat, etc?
[11:26] Which Linden: oh..sorry, group chat, techwolf
[11:26] Which Linden: here's the transcript: https://wiki.secondlife.com/wiki/User:Which_Linden/Office_Hours/2009_Apr_9
[11:26] Which Linden: this is a conversation that's gone on for a while
[11:26] Techwolf Lupindo: Ok. Figure that. I need crib notes. lol]
[11:27] Which Linden: yeah.....basically I keep hoping for a solution that's truly scaleable but keep discovering that there isn't anything already built
[11:28] Saijanai Kuhn: Which, Morgaine and I were arguing about next steps for interop, and group IM and trans-Group IM were a strong topic
[11:28] Which Linden: trans-group IM?
[11:28] Saijanai Kuhn: trans-Group IM = text + media
[11:29] Saijanai Kuhn: whiteboards amongst participating group members, video conferencing, etc
[11:29] Saijanai Kuhn: Morgaine didn't see it as VW interop per se, but I think the problem spaces overlap quite a bit
[11:30] Which Linden: oh so it's more like collaborative editing
[11:30] Which Linden: hm yeah that's an interesting point
[11:30] Saijanai Kuhn: or collaborative ivewing or whatever
[11:30] Morgaine Dinova: Well it's not a VW mashup. That's why I give it a specific name: MEDIA INTEROP.
[11:30] Which Linden: one oculd argue that Second Life is geared towards "spatial" collaboration rather than group-based collaboration
[11:30] Morgaine Dinova: I agree with Which
[11:30] Saijanai Kuhn: you could have a white board with passive and active subscribers, for example
[11:31] Saijanai Kuhn: which, but gorup IM IS group collaboration
[11:31] Saijanai Kuhn: everyone shares a text-only whiteboard
[11:31] Morgaine Dinova: It's spatial mashup that is at the heart of VW interop. I agree that media interop is HIGHLY useful, but it's not the end goal, just a midway one.
[11:32] Which Linden: Sai: yeah, we have groups, but as a feature every aspect of them seems tacked-on and broken
[11:32] Saijanai Kuhn: MOrgaine, sure, but useful to the extreme (when it works ) and very extensible as a metaphor
[11:33] Saijanai Kuhn: I mean, metanomics mashup between sims is a combination of video feed, group IM and consolidated local chat
[11:33] Morgaine Dinova: Very useful, but focussing on it defocusses from the "real" work.
[11:33] Which Linden: At some point when we consider media collaboration we get into discussing 3d desktops
[11:33] Saijanai Kuhn: the fact that we STILL don't have a metanomics connectionb etween SL and OpenSim is quite sad
[11:33] Eddy Stryker: whatever happened with LL's experiments with AQMP?
[11:34] Which Linden: Eddy: was just discussing that at the start
[11:34] Morgaine Dinova: Eddy!!!!!
[11:34] Morgaine Dinova: About time you showed up :P
[11:34] Eddy Stryker: ah, i'm too late
[11:34] Eddy Stryker: morgaine: i've been working :p
[11:34] Wyn Galbraith: more like timely
[11:34] Morgaine Dinova: Eddy: weak excuse :P
[11:35] Which Linden: Wyn!
[11:35] Wyn Galbraith: WHICH!! :D
[11:35] Which Linden: I didn't see you arrive. :-)
[11:35] Wyn Galbraith: I sneeked in all green like bamboo
[11:35] Which Linden: :-)
[11:36] Morgaine Dinova: Eddy: I really needed you here at Zero's last office hours. I was trying to get Cable Beach discussed, but Zero refused even to touch on the topic of asset distribution. Really need you around more.
[11:36] Wyn Galbraith: Nice to see you. Truly
[11:36] Which Linden: Eddy, you may wish to check out the transcript from last week https://wiki.secondlife.com/wiki/User:Which_Linden/Office_Hours/2009_Apr_9
[11:37] Which Linden: Wyn: well, glad you could make it, glad to see you again
[11:37] Morgaine Dinova: While you're at it, check out Zero's last transcript.
[11:37] Which Linden: It seems to me that our best two group chat architectures are 2 and 3, and we need to find out which one we want to chooes
[11:38] Saijanai Kuhn: which I finally got up :-/
[11:38] Wyn Galbraith: smiles and listens to catch up with what the heck is goin' on now days.
[11:38] Morgaine Dinova: Sai++
[11:38] Which Linden: Sai: finally got up what?
[11:39] Saijanai Kuhn: which active desktops seems to be a rather distant goal. Before that you could have croquet-like data sharing though group IM
[11:39] Saijanai Kuhn: ah, I was behind on getting the transcripts posted
[11:39] Which Linden: wow zero's office hours are so different from mine
[11:39] Which Linden: dude's got an agenda!
[11:39] Saijanai Kuhn: don't let it go to your head, er, stalk
[11:40] Saijanai Kuhn: Zero cut his from an hour a week to two hours once per month
[11:41] Morgaine Dinova: Which: we don't normally have prepared agendas at Zero's, rarely bother. But I made official ones to try to get MMOX matters discussed, because Zero is so AWOL from MMOX. Not altogether successful.
[11:41] Morgaine Dinova: No, it was two hours a week. So he cut it to 1/8th
[11:41] Eddy Stryker: saijanai: was the last office hour the one where zero wouldn't let the LLSD criticism be discussed? do you have a link to the transcript?
[11:42] Morgaine Dinova: And expects us to discuss things in depth. Impossible in the time. That's why Zero's OH are now just PR. No time for anything else.
[11:42] Saijanai Kuhn: ah you're right, he went form 2 meetings a week to one per month
[11:42] Saijanai Kuhn: eddy, just secd
[11:43] Which Linden: he seems to discuss things in more depth than we do here
[11:43] Which Linden: Honestly OH are pretty difficult to get anything done in
[11:43] Which Linden: Just the nature of the medium
[11:43] Saijanai Kuhn: I try to keep a pointer to al of them here, Eddy: [1]
[11:43] Lillie Yifu: It's the nture of office hours
[11:43] Morgaine Dinova: Eddy, no, that was some time back. This is last one, with no success discussing asset distribution: https://wiki.secondlife.com/wiki/User:Zero_Linden/Office_Hours/2009_apr_07
[11:43] Lillie Yifu: I did mine for the week earlier today
[11:44] Lillie Yifu: and nothing got doen there either
[11:44] Lillie Yifu: here leet me make it official
[11:44] Lillie Yifu: raises hand
[11:44] Which Linden: They're good for philosophizing though
[11:44] Lillie Yifu: is this going to be on the final?
[11:44] Lillie Yifu: raises hand again
[11:45] Lillie Yifu: can I get an extension onthe pper?
[11:45] Lillie Yifu: raises hand again
[11:45] Lillie Yifu: I still don't undertand this fotnoting thing
[11:45] Lillie Yifu: there that's office hours irl.
[11:45] Which Linden: ha ha true
[11:45] Morgaine Dinova: Which: yeah, not surprised about OH difficulties. But Mark needs to be active inMMOX. Not engaging to analyse his own proposals in open IETF discussion is, well, <no comment>.
[11:46] Techwolf Lupindo: tries to keep up with the other meeting his alt is at at the same time.
[11:46] Lillie Yifu: so maybe we need to schedule a meta OH where people can complain about things not getting done in OH, so that we can get things done inOHs?
[11:47] Morgaine Dinova: Lillie++
[11:47] Which Linden: The thing about Zero is that his contributions are usually so well-thought-out and topical that it seems as though nothing can be done without him
[11:48] Lillie Yifu: takes out note card
[11:48] Lillie Yifu: takes out push pin
[11:48] Lillie Yifu: writes "avoid single point of failure" on it
[11:48] Lillie Yifu: sticks notecard in bamboo stalk.
[11:48] Which Linden: One time the guy helped me debug something in 2 hours that had eluded me for 2 weeks
[11:48] Morgaine Dinova: Sure, Zero's great in the forward direction, but doesn't take kindly to feedback in the reverse direction. And that's what the IETF is about.
[11:49] Which Linden: Yet -- the point being that he is not the end matter, just like Philip is not
[11:49] Yohan Pintens: lol why did you wait 2 weeks to get help?
[11:49] Which Linden: I've always found him to be receptive to criticism, and I don't really see any evidence that he hasn't been
[11:49] Which Linden: Yohan: hah well I was trying to hack it myself that whole time
[11:49] Which Linden: thought I could do it
[11:50] Which Linden: probably could have, given enough time
[11:50] Morgaine Dinova: It's nice to solve problems oneself.
[11:50] Eddy Stryker: what are we talking about here, MMOX?
[11:50] Saijanai Kuhn: once invented teh barebones basics of TIFF. Then found a TIFF manual and facepalmed for a while
[11:50] Morgaine Dinova: Eddy: nah, only getting you back to speed.
[11:50] Eddy Stryker: up to speed on what?
[11:50] Morgaine Dinova: Here we were talking about group IM
[11:51] Morgaine Dinova: Up to speed on transcripts
[11:51] Lillie Yifu: kk and Which was saying that groups seem broken in every way
[11:51] Eddy Stryker: i'm glancing over some chat logs of morgaine pointing out the obvious and zero and zha cleverly dodging questions
[11:51] Yohan Pintens: on the information superhighway, what else?
[11:51] Eddy Stryker: it looks like MMOX crap to me
[11:51] Eddy Stryker: so the current question is how to fix LL's group IM?
[11:51] Morgaine Dinova: Well tht's how I see it too, but I can't fight the fight alone.
[11:52] Which Linden: Eddy: it's larger than that, but that's the guinea pig problem
[11:52] Lillie Yifu: I think the current question is how to move Zero hout of the dependency loopithout gettign a fence post error
[11:52] Eddy Stryker: keeping at least the same feature set as the current group IM features, minimizing the cost to the company to switch to a new architecture, and provide a rollout plan?
[11:53] Morgaine Dinova: Or even do it incrementally.
[11:53] Morgaine Dinova: It doesn't have to be a Big Bang swithcover
[11:53] Which Linden: Eddy: well at the very first we'd like to pick a technology that handles it assuming that the number of groups and number of concurrent users continues to grow
[11:53] Morgaine Dinova: switchover*
[11:53] Lillie Yifu: what are the options which?
[11:54] Which Linden: Our current technology already scales when number of groups grows, it just doesn't scale when the number of members of a group grow
[11:54] Lillie Yifu: And is there some way that this could be opened for people outside of ll to work on?
[11:54] Which Linden: Lillie: funny you should ask: https://wiki.secondlife.com/wiki/Message_Queue_Evaluation_Notes
[11:54] Lillie Yifu: settles down to read
[11:54] Eddy Stryker: great page, thanks
[11:55] Morgaine Dinova: Which: any updates on that evaluation, in terms of where things might be heading internally?
[11:55] Which Linden: yeah...oh, that reminds me I said I'd put up a section on proprietary tech
[11:56] Which Linden: Morgaine: well, internally we were stuck against the bulwark of model 3, which is not very good, and then we just realized that maybe model 2 isn't so bad
[11:56] Morgaine Dinova: I suggest the models be given names, or we'll perpetually have to keep going back to the list :P
[11:56] Which Linden: the problem with architecture 3 is that it requires N tcp connections for a user in N groups
[11:56] Morgaine Dinova: Hahaha
[11:56] Which Linden: hah um, go ahead and suggest names for them
[11:57] Saijanai Kuhn: Also, Infinity (or someone) set loose an internal memo about a proposed generic gorup IM protocol in OGP that could be used to test competing protocols without stepping on anyone's toes
[11:57] Morgaine Dinova: heads back to the transcript
[11:57] Techwolf Lupindo: I have a an IBM e-server pSeries 690, 256 cpu 1024 core 4T ram. Think LL could use that somewhere in there infrastructure? LIke groups chats? :-)
[11:57] Lillie Yifu: I think saij has a great point
[11:57] Lillie Yifu: what if some kind of gateway is exposed
[11:57] Lillie Yifu: and different models can be tested
[11:58] Which Linden: Techwolf, if you were to give that to us, I'm sure we could use it
[11:58] Saijanai Kuhn: gave you Leaked memo.
[11:58] Morgaine Dinova: Sai: is that memo available?
[11:58] Saijanai Kuhn: did I miss anyone?
[11:58] Morgaine Dinova: Muchas gracia Sai
[11:59] Morgaine Dinova: gracias*
[11:59] Which Linden: We don't really do memos, this looks like a strawman proposal for discussion to me
[11:59] Techwolf Lupindo: I would prefer to sell. Got to survide in this ecomity somehow. hehe
[12:00] Saijanai Kuhn: I think that's what it was. The "leaked meemo" thing was a joke
[12:00] Techwolf Lupindo: Or lease
[12:00] Which Linden: oh hah
[12:01] Saijanai Kuhn: it was just referred to as one, and passed around as one
[12:01] Eddy Stryker: question: would it be possible for a grid operator to just plug into an existing IM infrastructure instead of running and administering their own?
[12:01] Morgaine Dinova: How official a rough proposal is that? Should I file it under Sai, or under Infinity? :-)))))
[12:01] Which Linden: Yeah, so, I kinda don't actually care about what the protocol looks like to the client :-)
[12:01] Saijanai Kuhn: anonymous but Infinity mentioned it
[12:01] Morgaine Dinova: Under Sai then
[12:02] Eddy Stryker: for example, connecting all the IM connectors to an IRC backend or an AIM backend (assuming everyone had an AIM login associated with their account)
[12:02] Which Linden: Eddy: yes, though finding an existing chat room infrastructure that has the same properties has proven difficult
[12:02] Saijanai Kuhn: *I* nver heard of it til infinity mentioned it...
[12:02] Eddy Stryker: which: for LL's use case. but if your grid only supports 100 people to begin with it's a diffferent scenario
[12:03] Saijanai Kuhn: Eddy, that's one thing this memo does: provide the endpoints for a client to establish group IM with ANY architecture, not just the existing one
[12:03] Which Linden: Eddy: sure.....um, but I can't really support the notion of intentionally disregarding scalability
[12:03] Morgaine Dinova: Well that's a worthy goal fer sure, Sai
[12:03] Eddy Stryker: which: how is supporting multiple backends disregarding scalability?
[12:04] Saijanai Kuhn: with a functional non-LL Agent DOmain you couldl start playing with alternatives
[12:04] Which Linden: Eddy: I thought that's what you mean by your "only 100 people" case
[12:04] Saijanai Kuhn: weems to me we've discussed this before
[12:04] Which Linden: I mean, yes, obviously pluggable backends are classy
[12:05] Eddy Stryker: which: i just mean that different grids will have different requirements. some might only have a requirement of 100 peak concurrency, some will have a requirement of "you *must* use our internal jabber server for communication", etc.
[12:05] Which Linden: Ok sure
[12:06] Morgaine Dinova: Note that scalability discussions for interop need to embrace connection to LESS scalable systems too. For example, you need to be able to cope with your outbound queues jamming solid, and not have your highly scaled infrastructure die as a result.
[12:06] Which Linden: I thought you meant that having a pluggable backend would mean that we'd carve up the multiverse into chat room islands rather than deal with the scalability problem
[12:07] Eddy Stryker: morgaine: i'm still catching up in the logs.. is messaging (IM) across trust domains something that is on the table?
[12:07] Which Linden: Morgaine: I wouldn't call anything scaleable that couldn't handle choke points gracefully
[12:07] Saijanai Kuhn: eddy, that memo was a first attempt at proposing the generic protocol for that
[12:07] Morgaine Dinova: Eddy: not expressed in technical terms, no. But it's the intention, I think it's fair to say.
[12:08] Eddy Stryker: which: in my opinion, the multiverse will ultimately end up as chat room islands with some being larger than others and some being more open than others. but that's a different discussion
[12:08] Which Linden: Actually to me this seems targeted at Viewer<->AD chat communication
[12:08] Saijanai Kuhn: client requests a gorup IM service from teh Agent DOmain and the Agent Domain passes something back to the client
[12:08] Saijanai Kuhn: WHich, yeah, but aside from authorization, there's no requirement that the AD has anything to do with the actual innards of the system
[12:09] Eddy Stryker: ok, different people are saying different things :-). do we want scalability in a single trust domain, or do we want to make the N*N problem of everyone in every grid talking to every other person scale?
[12:09] Which Linden: Sai: right, but, I don't see anything in there about messaging across agent domains
[12:09] Saijanai Kuhn: a bit of both, perhaps
[12:09] Saijanai Kuhn: Which that would be up to the ADs I would think
[12:10] Which Linden: Great question Eddy; I'm not the right person to answer that though
[12:10] Morgaine Dinova: Eddy: although Zha is notable by her absence, every response does offer the copout that "nil trust is a valid trust policy", so I guess there's hope. I want something more concrete though.
[12:10] Which Linden: My concern is currently for scalability in our current trust domain
[12:11] Morgaine Dinova: No concern is primarily for the future VW universe, although I sure would like it if group IM worked well here too :P
[12:11] Eddy Stryker: which: ok, that's an actionable item
[12:11] Which Linden: It's a bit of a microcosm, really
[12:13] Which Linden: I should fly.
[12:13] Which Linden: Great discussion!
[12:13] Eddy Stryker: and you've arrived at a decision on message queueing systems and are going to announce the future direction now?
[12:13] Morgaine Dinova: SL isn't *really* a walled garden, as all the expressed intentions are to have all the door, windows and gates open. It just *seems* closed because of the very slow rate of movement on opening holes.
[12:13] Which Linden: Eddy: far from it
[12:13] Eddy Stryker: oh :(
[12:13] Wyn Galbraith: be like a tree and leaf Which!
[12:13] Which Linden: sorry
[12:13] Which Linden: Well thanks all for coming, glad to see y'all again
[12:13] Morgaine Dinova: Sadly, just a design study, Eddy :-(
[12:13] Wyn Galbraith: it was great
[12:13] Morgaine Dinova: But a very good one.
[12:14] Eddy Stryker: ok, wasn't sure on how far along things were. the evaluation page on the wiki is a great start
[12:14] Morgaine Dinova: Thanks Which!
[12:14] Which Linden: Morgaine: that's a great analogy btw
[12:14] Eddy Stryker: take care
[12:14] Which Linden: Peace out!
[12:14] Which Linden: Have a great one!

User:Which Linden/Office Hours/2009 Apr 16

Navigation menu

Search