AW Groupies/Chat Logs/AWGroupies-2009-06-16

From Second Life Wiki
< AW Groupies
Revision as of 14:46, 16 June 2009 by Saijanai Kuhn (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
  • [2009/06/16 9:41] Morgaine Dinova: Hi Aimee :-)
  • [2009/06/16 9:41] Zha Ewry: I noticed, a while back, that most live music venues don't have them. A few do, and the tiny community loves that. So.. I figured. I can be smart too ;-)
  • [2009/06/16 9:41] Mojito Sorbet: It sure shows up in 1.23.4 (which shipped waay before it was ready)
  • [2009/06/16 9:41] Aimee Trescothick: hey :)
  • [2009/06/16 9:41] Zha Ewry: So....
  • [2009/06/16 9:41] Twisted Laws: i'm running 1.23.4
  • [2009/06/16 9:41] Zha Ewry: This looks quorate
  • [2009/06/16 9:41] Morgaine Dinova: lol
  • [2009/06/16 9:42] Twisted Laws: can't ar anyone tho on nvidea card
  • [2009/06/16 9:42] Rex Cronon: hello everyboy
  • [2009/06/16 9:42] Morgaine Dinova: Hi Rex
  • [2009/06/16 9:42] xstorm Radek: test
  • [2009/06/16 9:42] Kwame Oh: Hi all , just here to pretend I know what you are talking about lol
  • [2009/06/16 9:42] Zha Ewry: So.. this is an attempt that I've been slowly bashing around with Infinity to define a proper set of mechansims to address client side caps
  • [2009/06/16 9:42] xstorm Radek: hi Rex
  • [2009/06/16 9:42] Rex Cronon: hi morgaine, xstrom
  • [2009/06/16 9:42] Morgaine Dinova: Kwame: join the club :P
  • [2009/06/16 9:43] Aimee Trescothick: ssssh, don't tell anyone Kwame, we're all pretending
  • [2009/06/16 9:43] Kwame Oh: Doh
  • [2009/06/16 9:43] Zha Ewry: So... as we looekd at OGP at MMOX this spring
  • [2009/06/16 9:43] Rex Cronon: have a fine time kwame:)
  • [2009/06/16 9:43] Zha Ewry: the "bi-directional" pipe story was kind of fictional
  • [2009/06/16 9:43] Zha Ewry: As in we've said for a while: Oh, there should be am odel of one
  • [2009/06/16 9:43] xstorm Radek: i still see group chat doing the timeout on finding the DNS at times
  • [2009/06/16 9:43] Zha Ewry: and we should think we've go one, but.. we haven't really done anything with it
  • [2009/06/16 9:44] Zha Ewry: So... as I was busy coding in event queues last month, I got semi serious about it
  • [2009/06/16 9:44] xstorm Radek: is that like a packet problem ?
  • [2009/06/16 9:44] Zha Ewry: so.. lets start with the architectrual problem for a moment...
  • [2009/06/16 9:45] Zha Ewry: We've sort of said, in AWG, and OGP that he pattern we're shooting for is RESTish
  • [2009/06/16 9:45] Zha Ewry: with URI addressed web resources against which we apply HTTP operations
  • [2009/06/16 9:45] Zha Ewry: and.. the specs to date, do a decent job of that on the servcie side
  • [2009/06/16 9:45] Zha Ewry: Caps, and the like, are pretty good mechanisms to do REST with security to "web" hostes resources
  • [2009/06/16 9:46] Zha Ewry: *hosted
  • [2009/06/16 9:46] Morgaine Dinova: Nice to see that REST is very much the preferred model in Opensim too, at least in discussion and intent.
  • [2009/06/16 9:46] Zha Ewry: Very much so, Morgaine, and that;'s important, too
  • [2009/06/16 9:46] Zha Ewry: Now.. on the client side, we've honestly said

[2009/06/16 9:46] Pixel Gausman perks

  • [2009/06/16 9:46] Zha Ewry: "Throw UDP or LLSD over event queues at the client, and hope it sticks"
  • [2009/06/16 9:47] Zha Ewry: Whichi s to say, we haven't really got much formalism for REST on the client at all
  • [2009/06/16 9:47] Morgaine Dinova: Hi Pixel ;-)
  • [2009/06/16 9:47] Zha Ewry: What, no snowglobe ave?
  • [2009/06/16 9:47] Mojito Sorbet: Could not the same addressing system work in reverse?
  • [2009/06/16 9:47] Zha Ewry: well, not precisely
  • [2009/06/16 9:47] Zha Ewry: and. thus the slabs of prim behind you
  • [2009/06/16 9:47] Zha Ewry: and the discussino
  • [2009/06/16 9:48] Zha Ewry: IF
  • [2009/06/16 9:48] Zha Ewry: we could host http and https
  • [2009/06/16 9:48] Zha Ewry: endpoints on the client
  • [2009/06/16 9:48] Mojito Sorbet: Setting aside how the pipes get the bits from one end to another, why not?
  • [2009/06/16 9:48] Zha Ewry: we could just say "client.fqdn.host.org://client/caps/f123-dce4/cap
  • [2009/06/16 9:48] Zha Ewry: err
  • [2009/06/16 9:48] Zha Ewry: http:// that
  • [2009/06/16 9:48] Zha Ewry: But
  • [2009/06/16 9:49] Zha Ewry: http won't get through 90% of the corporate, home, or machien firewalls
  • [2009/06/16 9:49] Mojito Sorbet: Sure it will, if the TCP　pipe stays open
  • [2009/06/16 9:49] Zha Ewry: not hosted on the client
  • [2009/06/16 9:49] Mojito Sorbet: Client oipens the pie from its end.
  • [2009/06/16 9:49] Mojito Sorbet: pipe
  • [2009/06/16 9:49] Zha Ewry: That gets you r-http
  • [2009/06/16 9:49] Zha Ewry: IF you can get the proxy firewalls to do upgrade
  • [2009/06/16 9:49] Pixel Gausman: Mojito: thus the event queue, right?
  • [2009/06/16 9:49] Zha Ewry: And.. it isn't generally addressable
  • [2009/06/16 9:50] Mojito Sorbet: The sserver knows where the client is.
  • [2009/06/16 9:50] Dzonatas Sol: open pie... i think you just figured out why they call it blueberry
  • [2009/06/16 9:50] Mojito Sorbet: So in your exmaple "client.fqdn.host.org://client/caps/f123-dce4/cap"
  • [2009/06/16 9:50] Mojito Sorbet: The part that says "client" is really just a dummy
  • [2009/06/16 9:50] Mojito Sorbet: "/client/"
  • [2009/06/16 9:51] Lily Nozaki: Hello!
  • [2009/06/16 9:51] Mojito Sorbet: IOt is not like the server needs to do a name lookup on every exchange like a web browser would
  • [2009/06/16 9:51] Rex Cronon: hi
  • [2009/06/16 9:51] Zha Ewry: So..
  • [2009/06/16 9:51] Zha Ewry: The problem is that you can't listen to http on client side in any reliable way
  • [2009/06/16 9:51] Zha Ewry: nor.. can you open an http connection to the client in any sane and useful way.
  • [2009/06/16 9:51] Mojito Sorbet: Not general purpose HTTP, no
  • [2009/06/16 9:52] Mojito Sorbet: But there aere two parts to setting up an HTTP exchange
  • [2009/06/16 9:52] Zha Ewry: nor, in 2009, and probably 2010, reverse-http
  • [2009/06/16 9:52] Mojito Sorbet: Connecting to the server is a separate steo from running GETs and POSTs over it
  • [2009/06/16 9:52] Zha Ewry: So. thus, the eventq
  • [2009/06/16 9:52] Zha Ewry: Which
  • [2009/06/16 9:52] Morgaine Dinova: Let's assume we have the reverse transport into the client, be it over RHTTP or WebSockets or the old COMET long polls. It's more interesting to talk about where we go with REST in that direction.
  • [2009/06/16 9:52] Zha Ewry: We all agree is painful and hacy
  • [2009/06/16 9:52] Zha Ewry: right
  • [2009/06/16 9:53] Zha Ewry: But its what actually is deployable.
  • [2009/06/16 9:53] Zha Ewry: So.. the problem is two fold
  • [2009/06/16 9:53] Zha Ewry: or 3 or 4
  • [2009/06/16 9:53] Zha Ewry: first)
  • [2009/06/16 9:53] Pixel Gausman: Zha: with the IETF pitch fits over the event queue? or is this a problem other spaces have as well?
  • [2009/06/16 9:53] Zha Ewry: model it properly, as a two way pipe
  • [2009/06/16 9:53] Pixel Gausman: will the IETF, even
  • [2009/06/16 9:53] Zha Ewry: the http-bis session at ITEF-74 was very open to fixing this

[2009/06/16 9:54] Pixel Gausman googles http-bis

  • [2009/06/16 9:54] Zha Ewry: Now.. will the IANA request for eventq and rhttp as URI schemes be as open? Interesting question
  • [2009/06/16 9:54] Pixel Gausman: do we really need a URI scheme?
  • [2009/06/16 9:54] Zha Ewry: But.. at the discussions on r-http, we had 4 or 5 (depending how you count) the groups needing it
  • [2009/06/16 9:55] Zha Ewry: We need something for the servcies to manage
  • [2009/06/16 9:55] Morgaine Dinova: Pixel: well it's been an active topic ever since COMET came out. They're going the W3C route with HTML5 instead of IETF, but it seems they've finally reached their goal. I've not tried WebSockets yet though.
  • [2009/06/16 9:55] Zha Ewry: And. websockets is only about 1/3 of the space
  • [2009/06/16 9:55] Pixel Gausman: i just poke on anything that is going to cause additional work in the viewer.

[2009/06/16 9:55] Zha Ewry chuckles

[2009/06/16 9:56] Zha Ewry sighs and tries to recall the notes from 2 months back.

  • [2009/06/16 9:56] Zha Ewry: Its kind of painfully short on the http semantics
  • [2009/06/16 9:56] xstorm Radek: looking at it i say yes its a needed step
  • [2009/06/16 9:56] Pixel Gausman: ...and i wonder if there are preliminary steps we can chew off without marching down the full URI thing
  • [2009/06/16 9:56] Morgaine Dinova: Zha: ignore the question, distracts from this REST topic :P

[2009/06/16 9:56] Saijanai Kuhn blushes and makes note to ask for transcripts of last month's meetings

  • [2009/06/16 9:56] Zha Ewry: I think the URI is actually very lightweight
  • [2009/06/16 9:56] Zha Ewry: and..
  • [2009/06/16 9:57] Zha Ewry: The painpoint for the client, is mapping "resource" paths to handlers
  • [2009/06/16 9:57] Morgaine Dinova: Hi Sai :-)
  • [2009/06/16 9:57] Zha Ewry: which.. we're going to have in one form or another, anyway

[2009/06/16 9:57] Pixel Gausman pushes her snowglobe across the table to Zha

[2009/06/16 9:57] Zha Ewry looks for a table

  • [2009/06/16 9:57] Zha Ewry: So..
  • [2009/06/16 9:57] Zha Ewry: Lets walk through this a bit at a time

[2009/06/16 9:58] Aimee Trescothick hold on tight in case someone shakes their snowglobe

  • [2009/06/16 9:58] xstorm Radek: only problem i see is if some one is on a 486 pc with dos still on it ;-)
  • [2009/06/16 9:58] Zha Ewry: I've got a "web side" service which wants to route messages down to the client, to be handled
  • [2009/06/16 9:58] Zha Ewry: The example I'm using is "friendship offers" becuase they are pluasible
  • [2009/06/16 9:59] Zha Ewry: (and assume none of the details are cannonical, just illustrative)
  • [2009/06/16 9:59] Zha Ewry: What we want to do, is give the web side servcie a way to address the client, down tot he REST resource, so it can POST LLSD encoded messages on it.
  • [2009/06/16 10:00] Zha Ewry: That's the fundamental goal.
  • [2009/06/16 10:00] Zha Ewry: And.. we want to permit the deployers of web side services
  • [2009/06/16 10:00] Zha Ewry: to have a *lot* of flexability in how they deploy these services int he computational fabric
  • [2009/06/16 10:00] Zha Ewry: we don't want to dictate tot hem how they build the server farms
  • [2009/06/16 10:00] Morgaine Dinova: Zha: I'm not countering that, but at some time I'd like to hear the reason why that relationship has to be symmetrical, other than just apparent elegance :-)
  • [2009/06/16 10:00] Mojito Sorbet: I　have a suggestion taken from the world of VOIP
  • [2009/06/16 10:01] Kwame Oh: will beat a hasty retreat at this stage but hopes he is welcome when topic is less strenuous, like life the universe and everything, less digital :-)
  • [2009/06/16 10:01] Rex Cronon: tc
  • [2009/06/16 10:01] Pixel Gausman: Margaine: +1
  • [2009/06/16 10:01] Pixel Gausman: Morgaine, even
  • [2009/06/16 10:01] Morgaine Dinova: Pixel: I actually have reasons for wanting the symmetry, but wondered what Zha's were
  • [2009/06/16 10:01] Zha Ewry: Its a good questoin, and there are actually some pretty compeling reasons, roughly, because the problems are not asymetric
  • [2009/06/16 10:02] Zha Ewry: The only reason they even appear assymetric, is becase we're stuck with clients behind firewalls
  • [2009/06/16 10:02] Mojito Sorbet: So,l this is after the client has logged into begin with, some service somewhere wants to contact that client?

[2009/06/16 10:03] Zha Ewry nods "rouhgly"

  • [2009/06/16 10:03] Mojito Sorbet: ok
  • [2009/06/16 10:03] Mojito Sorbet: How about this
  • [2009/06/16 10:03] Zha Ewry: Most things get provoked client side, but the servcies may not be deployed so that the client has a clue about them
  • [2009/06/16 10:03] Mojito Sorbet: As part of loggin in, the client has connected to something, lets call it the Agent Domain Server
  • [2009/06/16 10:03] Lily Nozaki: You want to let unsolicited external services make direct http connections with a client of their choice?..
  • [2009/06/16 10:03] Zha Ewry: God no. Lily
  • [2009/06/16 10:03] Morgaine Dinova: (My main reason is that, post-interop, there is nothing special about the remote server. Some services could easily be running on the local client, or on the local LAN with the client as a proxy. So the remote server has exactly the same relationship to us as us to it.)
  • [2009/06/16 10:03] Lily Nozaki: I just jumped in, I'm missing details, sorry. :3
  • [2009/06/16 10:04] Mojito Sorbet: This friendship service has some way of locating the Agent Server for the account in question
  • [2009/06/16 10:04] Zha Ewry: EKR would be foaming at the mouth 30 seconds after I said that
  • [2009/06/16 10:04] Mojito Sorbet: So it asks the Agent Domani Server to tell the client (over rHTTP or some such)
  • [2009/06/16 10:04] Lily Nozaki: Me too. :p
  • [2009/06/16 10:04] Mojito Sorbet: to "Please contact service XYZ at URL
  • [2009/06/16 10:04] Zha Ewry: well
  • [2009/06/16 10:04] Mojito Sorbet: So the client always initates the cconnection
  • [2009/06/16 10:04] Morgaine Dinova: That's why assymetry is actually bad for interop.
  • [2009/06/16 10:04] xstorm Radek: ? some way ? i can see a new AI scrit in the works now ;-) he he he
  • [2009/06/16 10:04] Mojito Sorbet: But, bueiong rHTTP, the service at the other end can use it
  • [2009/06/16 10:04] Zha Ewry: With some careful caveats, that's pretty much what a eventq: URI says
  • [2009/06/16 10:05] Zha Ewry: ansd the pattern would work identicaly with rhttp:// URIs
  • [2009/06/16 10:05] Zha Ewry: and.. if done properly, the code is identical with the excpeion of the endpoints between the client and the hostin service
  • [2009/06/16 10:05] Mojito Sorbet: It is just like how a SL rparcel media URL　makes the viewer contact an outside streaming service.
  • [2009/06/16 10:05] Zha Ewry: BUT..
  • [2009/06/16 10:05] Zha Ewry: several issues show up along the way
  • [2009/06/16 10:06] Zha Ewry: So..
  • [2009/06/16 10:06] Pixel Gausman: Morgaine: what's an example of a service that is something that should be available on the client, but is instead on the local lan?
  • [2009/06/16 10:06] Mojito Sorbet: voice?
  • [2009/06/16 10:06] Zha Ewry: I did this for event queues first, because, they actually work, while r-http is eaten alive by most proxies.
  • [2009/06/16 10:07] Zha Ewry: The principal is essentially identical, tho
  • [2009/06/16 10:07] Zha Ewry: So.. peek a the gray slide in the middle of the prim wall
  • [2009/06/16 10:08] Mojito Sorbet: The one on top?
  • [2009/06/16 10:08] Zha Ewry: where you have the blue handlers
  • [2009/06/16 10:08] Zha Ewry: yes
  • [2009/06/16 10:08] Zha Ewry: You've got events queues taking in requests, which look, just like the LLSD map to the right of it
  • [2009/06/16 10:08] xstorm Radek: if your inback of a fire wall some things may not work like voice for one will there not be setting for that or better auto setting may be ?
  • [2009/06/16 10:08] Morgaine Dinova: Pixel: the first immediate local service is asset and inventory. The remote server is not a special case at all post-interop. Asset services could you at S3 or Yahoo or your local ISP or on your client machine or on your LAN.
  • [2009/06/16 10:09] Morgaine Dinova: s/could you/could run/
  • [2009/06/16 10:09] Zha Ewry: So, the slide "POST method in EQ"
  • [2009/06/16 10:10] Zha Ewry: Shows a map of pretyty normal POST to a resource, mapped out as an array elelement in a set of events flowing down the eventq
  • [2009/06/16 10:10] Mojito Sorbet: So everything between the outer <map> tags is one Q entry?
  • [2009/06/16 10:10] Pixel Gausman: Morgaine: so u want a remote service to be querying/changing assets and inventory stared locally by going thru the viewer?
  • [2009/06/16 10:10] Zha Ewry: The "resource" is exactly what a cap looks like sim side, and part of the eventq:// URI
  • [2009/06/16 10:11] Zha Ewry: Yes, Mojito
  • [2009/06/16 10:11] Zha Ewry: the outer map, is in the array, the inner map, is the arbitrary LLSD payload we're posting
  • [2009/06/16 10:11] Zha Ewry: (for added fun, it could be a string, and we woudl post that too, or any LLSD encodable content)
  • [2009/06/16 10:12] Zha Ewry: you're in a LLSD MAP, so its <key<payload</key> {any valid LLSD payload}
  • [2009/06/16 10:12] Zha Ewry: Here its a totally too weak for real use friendship request
  • [2009/06/16 10:13] Zha Ewry: (There should at least be an openID, or similar connection for the ownership of "Hippo The Avatarian")
  • [2009/06/16 10:13] Zha Ewry: anyway
  • [2009/06/16 10:13] Pixel Gausman: Zha: where does the viewer post it's reply to the friendship request?
  • [2009/06/16 10:13] Morgaine Dinova: Pixel: each "world" chooses its own configuration of sets of services, and it varies on a case by case basis. If you use an S3-hosted asset service, you would expect the world you're visiting to obtain those assets directly from S3, since that's most scalable. If you have an asset service running locally, or behind a firewall without ingress connectivity, then you'll want your client connection to be proxying the accesses to the asset service.
  • [2009/06/16 10:13] Mojito Sorbet: Is not the UserUUID in this case a sort of handle to that?
  • [2009/06/16 10:14] Zha Ewry: You'd like both the UUID, and the owner of the UUID/Name mapping, in a broader metevaerse I xpect. I merely mention that it's not a totaly realistic reequest, is all
  • [2009/06/16 10:14] Mojito Sorbet: Pixel, is the resource tag an indicator of how to reply?
  • [2009/06/16 10:15] Zha Ewry: In fact, you have two choices, pixel
  • [2009/06/16 10:15] Zha Ewry: The reply needs to go back up the eventq, since its the only element which can fully route back the response
  • [2009/06/16 10:16] Zha Ewry: The sequence number is there for that reason
  • [2009/06/16 10:16] Mojito Sorbet: So the server side of the eventq channel has to know where to redistribute any replies coming back form the client
  • [2009/06/16 10:16] Zha Ewry: Not any, Mojito
  • [2009/06/16 10:16] Zha Ewry: Only ones it delievred
  • [2009/06/16 10:16] Mojito Sorbet: yes
  • [2009/06/16 10:16] Zha Ewry: whihc is imporant
  • [2009/06/16 10:16] Mojito Sorbet: That is what I meant
  • [2009/06/16 10:16] Zha Ewry: k
  • [2009/06/16 10:16] Pixel Gausman: Zha: and then the AD event queue handler would forward it on to the service cloud?
  • [2009/06/16 10:16] Mojito Sorbet: An error if it receives a reply to a request it did not handle in the firts place
  • [2009/06/16 10:16] Zha Ewry: Yes, Pixel, that;'s the expected behaviro
  • [2009/06/16 10:17] Pixel Gausman: Zha: u said two options....
  • [2009/06/16 10:17] Zha Ewry: so,t he caller gets back a 404, 501, or a 200, or whateve
  • [2009/06/16 10:17] Zha Ewry: right
  • [2009/06/16 10:17] Zha Ewry: The other option, is to construct the "call back" for the handler, and let it post it directly. That is not pretty
  • [2009/06/16 10:17] Pixel Gausman: gee, the wooly caterpillar is fussy today. :)
  • [2009/06/16 10:17] Zha Ewry: Insanely helpful
  • [2009/06/16 10:17] Zha Ewry: That goes right into the next draft
  • [2009/06/16 10:18] Mojito Sorbet: So, restrict replies to going back the way the request came in
  • [2009/06/16 10:18] Zha Ewry: Now
  • [2009/06/16 10:18] Zha Ewry: Some other bits of fun
  • [2009/06/16 10:18] Zha Ewry: Since, the eventq is likely shared
  • [2009/06/16 10:18] Mojito Sorbet: But such a reequest could, if necessary, have a side effect of opening some entirely new channel elsewhere.
  • [2009/06/16 10:18] Pixel Gausman: Zha: what's unpretty abt posting to the callback?
  • [2009/06/16 10:18] Zha Ewry: Breaking the rule of 2, mostly
  • [2009/06/16 10:18] Zha Ewry: Also forcing some ugly things into the re-try loop
  • [2009/06/16 10:19] Zha Ewry: (the rule of 2 http pipes between a client/server pair)
  • [2009/06/16 10:19] Pixel Gausman: rule of two: "only two people agree with Zha at a time"?
  • [2009/06/16 10:19] Pixel Gausman: Oo
  • [2009/06/16 10:19] Pixel Gausman: :)
  • [2009/06/16 10:19] Mojito Sorbet: The AD handler needs to see the reply coming back, so it knows when it can remove that id number form its list.

[2009/06/16 10:19] Zha Ewry rolls her eyes, smokes them and inserts a fresh set from inventory

  • [2009/06/16 10:19] Morgaine Dinova: Pixel: continuing the answer, you have an example of it in SL right now. The regions proxy our access to the SL asset service, we don't access it directly. But when LL starts running asset services on S3 or wherever, then we'll be accessing them directly instead of through the sim. Same applies locally. That's why the situation is very symmetrical.
  • [2009/06/16 10:19] Zha Ewry: so.. a couple of other bits of fun:
  • [2009/06/16 10:20] Zha Ewry: the cap, on the client, just like in the sims, is unguessable in the large
  • [2009/06/16 10:20] Zha Ewry: noteice the resouce string has a lovely UUID in themiddle
  • [2009/06/16 10:20] Zha Ewry: not:
  • [2009/06/16 10:20] Zha Ewry: ogpx.friendship.listner
  • [2009/06/16 10:20] Mojito Sorbet: But the Request Type in this case is a partial way of identifying which Handler needs to see it
  • [2009/06/16 10:20] Zha Ewry: but 128 bits of ungeussable path
  • [2009/06/16 10:21] Zha Ewry: Ah.
  • [2009/06/16 10:21] Twisted Laws: I like the idea of anything wanting to talk with the client on its own, it sends a connection request to the agent domain and it passes that to client on existing circuit.. client then connects to the request with basically an http listener. Same if something wants to talk to my asset server then it could send the request to the agent domain and it would send the message to my asset server to establish the connection. i like security aspect of this as the agent domain could refuse to send some requests and the client would have told agent domain at login what caps it could support.
  • [2009/06/16 10:21] Mojito Sorbet: Presumably "friendship request" is a subset of the things that UUID can handle
  • [2009/06/16 10:21] Zha Ewry: CAPS style, it's telling the handler which message it is, not enough to tell you the handler
  • [2009/06/16 10:21] Zha Ewry: and.. the UUID, is a one shot, per sesssion
  • [2009/06/16 10:22] Mojito Sorbet: The client needs a way to vector the request. It has to know all possible request types in advance
  • [2009/06/16 10:22] Zha Ewry: so that other services sharing your eventq can't stuff forged packets onto your handlers
  • [2009/06/16 10:22] Infinity Linden: you're talking about a separate "friendship service" than the agent domain?
  • [2009/06/16 10:22] Zha Ewry: Just a service
  • [2009/06/16 10:22] Zha Ewry: Maybe part of ther AD, maybe not
  • [2009/06/16 10:22] Mojito Sorbet: Its an example in this case
  • [2009/06/16 10:22] Infinity Linden: right
  • [2009/06/16 10:22] Infinity Linden: oaky
  • [2009/06/16 10:22] Zha Ewry: Doesn't mattter
  • [2009/06/16 10:22] Infinity Linden: okay
  • [2009/06/16 10:22] Mojito Sorbet: Might be coming from faceBook say
  • [2009/06/16 10:22] Zha Ewry: So, the CAP
  • [2009/06/16 10:22] Zha Ewry: routes it to the resource
  • [2009/06/16 10:22] Zha Ewry: which means, we need to havbe passed the CAP up
  • [2009/06/16 10:23] Infinity Linden: cause it's important to remember... the agent domain MAY use non OGP protocol to talk to other services
  • [2009/06/16 10:23] Zha Ewry: (As a eventq or rhttp: URI
  • [2009/06/16 10:23] Zha Ewry: )
  • [2009/06/16 10:23] Zha Ewry: (or others, if we define them)
  • [2009/06/16 10:23] Infinity Linden: but yeah. it's good to get these examples down pat
  • [2009/06/16 10:23] Zha Ewry: So..
  • [2009/06/16 10:23] Mojito Sorbet: What happens behind the AD is not the client's worry
  • [2009/06/16 10:23] Zha Ewry: Correct, mojito
  • [2009/06/16 10:23] Zha Ewry: and.. its part of why the eventq: scheme is messy
  • [2009/06/16 10:23] Zha Ewry: Becuase we don't just include the path
  • [2009/06/16 10:23] Zha Ewry: we also identify the client
  • [2009/06/16 10:23] Zha Ewry: So..
  • [2009/06/16 10:24] Mojito Sorbet: It is like a cellphone
  • [2009/06/16 10:24] Zha Ewry: if the service is actually handed off to another computational resource in the cloud
  • [2009/06/16 10:24] Mojito Sorbet: My AD is like the tower my phone i stalking to
  • [2009/06/16 10:24] Zha Ewry: it has a cap, which it can use to route it back
  • [2009/06/16 10:24] Infinity Linden: thought it MAY be the client's worry if a service is deployed such that the client needs to speak OGP to it
  • [2009/06/16 10:24] Morgaine Dinova: Car analogy please. ;-)
  • [2009/06/16 10:24] Mojito Sorbet: The network knows my phone is reachable thru that tower
  • [2009/06/16 10:24] Zha Ewry: Or possibly several others
  • [2009/06/16 10:24] Zha Ewry: And that's insanely sueful
  • [2009/06/16 10:24] Zha Ewry: *useful
  • [2009/06/16 10:24] Infinity Linden: mm.. actually through the MSC, not the BSC, but i get your point
  • [2009/06/16 10:25] Zha Ewry: So..
  • [2009/06/16 10:25] Zha Ewry: The missing slide in all this is the one to the far right
  • [2009/06/16 10:25] Zha Ewry: see the cap/URI pair?
  • [2009/06/16 10:25] Zha Ewry: THAT is used to tie the "services" side to the client cap
  • [2009/06/16 10:26] Zha Ewry: We post that onto the frienship service saying
  • [2009/06/16 10:26] Zha Ewry: "Here is my CLIENT side listener cap"
  • [2009/06/16 10:26] Zha Ewry: its a URI
  • [2009/06/16 10:26] Zha Ewry: Woot, we can replace eventq with
  • [2009/06/16 10:26] Zha Ewry: rhttp, ptth, xmpp
  • [2009/06/16 10:26] Zha Ewry: down the road
  • [2009/06/16 10:26] Zha Ewry: The resource path stays the same
  • [2009/06/16 10:26] Pixel Gausman: so the viewer posts that to the friendship service?
  • [2009/06/16 10:26] Zha Ewry: Yep
  • [2009/06/16 10:26] Zha Ewry: It creates a cap, with a UUID
  • [2009/06/16 10:27] Zha Ewry: and posts it to the service saying 'Here I am"
  • [2009/06/16 10:27] Zha Ewry: sets up its routing internally
  • [2009/06/16 10:27] Mojito Sorbet: So ther friendship servbice is telling the client how to contact it firectly?
  • [2009/06/16 10:27] Zha Ewry: (and can hash the cap as need be)
  • [2009/06/16 10:27] Infinity Linden: wait... the client is doing a service establishment dance with the foreign service and it's using it's AD's event queue?
  • [2009/06/16 10:27] Zha Ewry: not directly, but via a URI
  • [2009/06/16 10:27] Mojito Sorbet: The AD event queue mediates the connection, but does not participate in the connection once established.
  • [2009/06/16 10:27] Zha Ewry: that, infinity, depends on the caps that were granted
  • [2009/06/16 10:28] Zha Ewry: The cap grant which led to this servcie
  • [2009/06/16 10:28] Zha Ewry: may have included an eventqu on it
  • [2009/06/16 10:28] Pixel Gausman: Mojito: i'd rather avoid having a 1-1 correspondence between viewer event queue and service that needs to make resource requests of it
  • [2009/06/16 10:28] Infinity Linden: if the client is going to talk to a foreign service, why not just do regular service establishment?
  • [2009/06/16 10:28] Zha Ewry: or. maybe it's local to the AD in the deployment in question
  • [2009/06/16 10:28] Zha Ewry: We're way into the flow, Infinity
  • [2009/06/16 10:28] Mojito Sorbet: In VOIP, the SIP server finds the other party you are calling, and passes the port numbers back and forth. But the RTP geos caround the SIP server.
  • [2009/06/16 10:28] Zha Ewry: The cap, and the eventq
  • [2009/06/16 10:28] Zha Ewry: may be on a AD
  • [2009/06/16 10:28] Zha Ewry: on a region
  • [2009/06/16 10:29] Zha Ewry: on a third party servcie
  • [2009/06/16 10:29] Infinity Linden: well.. OGP is not SIP
  • [2009/06/16 10:29] Zha Ewry: we may have eventqs local to them
  • [2009/06/16 10:29] Zha Ewry: if..
  • [2009/06/16 10:29] Mojito Sorbet: It was ananlogy
  • [2009/06/16 10:29] Zha Ewry: we're on a third party cap
  • [2009/06/16 10:29] Lily Nozaki: Being ported.. have fun. :3
  • [2009/06/16 10:29] Infinity Linden: but are you thinking abotu a trapezoidal architecture?
  • [2009/06/16 10:29] Zha Ewry: we're not going to be using the AD's event qy
  • [2009/06/16 10:29] Pixel Gausman: how does the viewer discover ( and sort) the need for different event queues?
  • [2009/06/16 10:29] Zha Ewry: Because, when it asks for a cap set
  • [2009/06/16 10:29] Zha Ewry: in the replay
  • [2009/06/16 10:29] Zha Ewry: is an eventq cap
  • [2009/06/16 10:29] Zha Ewry: which it can open
  • [2009/06/16 10:29] Zha Ewry: So..
  • [2009/06/16 10:30] Zha Ewry: if I ask for "org.zha.super.special.cap"
  • [2009/06/16 10:30] Zha Ewry: I may get back
  • [2009/06/16 10:30] Mojito Sorbet: Loggin in, the client creates the EQ to the AD. All others get created on demand after that.
  • [2009/06/16 10:30] Zha Ewry: both the URI of the special cap
  • [2009/06/16 10:30] Zha Ewry: and a bog standard event q cap
  • [2009/06/16 10:30] Zha Ewry: just like
  • [2009/06/16 10:30] Zha Ewry: when you ask the AD for its cap set
  • [2009/06/16 10:30] Zha Ewry: one of them is an eventq cap
  • [2009/06/16 10:31] Zha Ewry: and..
  • [2009/06/16 10:31] Zha Ewry: because I have the clientID in the eventq URIs
  • [2009/06/16 10:31] Pixel Gausman: i request the EventQueueGet cap
  • [2009/06/16 10:31] Zha Ewry: I can figure out if I have a path locally, or need to use one that;s hosted off to the end of the world
  • [2009/06/16 10:31] Zha Ewry: Shshss.
  • [2009/06/16 10:31] Zha Ewry: When you ask for a cap, you may get more than you expect ;-)
  • [2009/06/16 10:31] Pixel Gausman: how do i know who i need to request it from? are there other services that i need to also request it from?
  • [2009/06/16 10:32] Pixel Gausman: or just region and AD?
  • [2009/06/16 10:32] Zha Ewry: The way I'm leaning
  • [2009/06/16 10:32] Zha Ewry: is that, when you do a cap request, you may get back a bunch of caps, and eventqs, are ones you have to use
  • [2009/06/16 10:32] Zha Ewry: But.. Thats' why this is very much a 0.1 draft
  • [2009/06/16 10:33] Infinity Linden: @pixel. i think the idea is the viewer is "just supposed to know" where to do service establishment
  • [2009/06/16 10:33] Zha Ewry: What it starts to do, is fully seperate out resource routing from adressing, from the two way pipe
  • [2009/06/16 10:33] Morgaine Dinova: You know, in many ways, OGP has nothing to do with VWs --- it's a distributed services protocol and architecture. Only the sims deal with the VW concept directly. The rest can be really generic.
  • [2009/06/16 10:33] Infinity Linden: and in the future if peeps wanna use XRDS or whatever.. .they're more than welcome too
  • [2009/06/16 10:33] Pixel Gausman: Infinity: that's what frightens me. :)
  • [2009/06/16 10:33] Mojito Sorbet: If a viewer is able to support a service at all (say playing music) it has to know how to go about sertting it up
  • [2009/06/16 10:33] Zha Ewry: so.. we have some future proofing as rhttp etc happens
  • [2009/06/16 10:33] Infinity Linden: right now we have the loginuri and helperuri that can be specified from the command line
  • [2009/06/16 10:34] Infinity Linden: OUR client will probably always use the AD for all it's caps
  • [2009/06/16 10:34] Zha Ewry: My feeling is that we're going to use the seedcap/cap pattern
  • [2009/06/16 10:34] Infinity Linden: though something like the Hippo Viewer's grid picker would be cool
  • [2009/06/16 10:34] Zha Ewry: and some caps may yield new seedcaps, and its easily recurisve
  • [2009/06/16 10:34] Morgaine Dinova: The commandline links are just an early implementation. It'll be dynamic in time.
  • [2009/06/16 10:34] Pixel Gausman: Infinity: so someone has to fork your viewer (again) to support something different?
  • [2009/06/16 10:34] Zha Ewry: Note that this also
  • [2009/06/16 10:34] Zha Ewry: (I want this in snowglobe, neatly)
  • [2009/06/16 10:35] Pixel Gausman: GridInfo in snowglobe would be coolness
  • [2009/06/16 10:35] Zha Ewry: allows us to do things like merge the eventqs of sims onto a single handler for a busy region and offload the traffic entirely from the region
  • [2009/06/16 10:35] Techwolf Lupindo: Is there a Hippo Viewer?
  • [2009/06/16 10:35] Infinity Linden: hmm.. i should publish some of teh stuff JH and i have been gabbing about RE: OAuth and service establishment. it might make things easier
  • [2009/06/16 10:35] Pixel Gausman: tech: yeah, it's geared towards OpenSIm
  • [2009/06/16 10:35] Techwolf Lupindo: Ok.
  • [2009/06/16 10:35] Infinity Linden: @pixel.. i'm actually a big fan of it
  • [2009/06/16 10:36] Zha Ewry: What I want to see
  • [2009/06/16 10:36] Zha Ewry: is caps, client side resourcepatterns, and such
  • [2009/06/16 10:36] Infinity Linden: i have an OpenSim instance running in my house, starting the client from the command line all the time is annoying
  • [2009/06/16 10:36] Pixel Gausman: Infinity: me too. the command line is getting a lil crowded
  • [2009/06/16 10:36] Infinity Linden: oh
  • [2009/06/16 10:36] Infinity Linden: and
  • [2009/06/16 10:36] Zha Ewry: become an underpinnign for OpenSim, and Second Life, and anyone who wants to use this style of plumbing
  • [2009/06/16 10:36] Zha Ewry: and get the basic routing out in nice lbraries
  • [2009/06/16 10:36] Zha Ewry: and in Snowglowb
  • [2009/06/16 10:36] Infinity Linden: i've been working no a MIME-type and related definition intended to "launch" the viewer
  • [2009/06/16 10:36] Techwolf Lupindo: I usally start from the console to get console output and --coreoncrash
  • [2009/06/16 10:37] Zha Ewry: Provide the core handlers for it on OpenSim in C#, in Python, Java, and so on
  • [2009/06/16 10:37] Infinity Linden: @techwolf. yeah. sadly. me too

[2009/06/16 10:37] Pixel Gausman wonders if Janus would consider porting GridInfo to SnowGlobe.... and if the patch would be accepted

  • [2009/06/16 10:37] Zha Ewry: So.. we can just share it all widely
  • [2009/06/16 10:37] Infinity Linden: that's philip and robla's call about accepting it
  • [2009/06/16 10:37] Infinity Linden: but, dang it would be nice
  • [2009/06/16 10:37] Rex Cronon: zha what do u mean by this "Provide the core handlers for it on OpenSim in C#, in Python, Java, and so on"?
  • [2009/06/16 10:37] Morgaine Dinova: Well Aimee's here, and she's on the Snowglobe committers list :P
  • [2009/06/16 10:38] Zha Ewry: I'd think that the basic listener for this pattern
  • [2009/06/16 10:38] Techwolf Lupindo: The grid list given on starup could be reduced a lot. currently, I think there only two usfull entires on there.
  • [2009/06/16 10:38] Zha Ewry: really, really, really, ought to b abstracted out to
  • [2009/06/16 10:38] Zha Ewry: so you have easy to use code in a number of environments
  • [2009/06/16 10:38] Pixel Gausman: Morgaine: but she shouldn't start slapping in wild changes to snowglobe. :) it's a community process.
  • [2009/06/16 10:38] Zha Ewry: I'm clearly going to end up doing one for C# for OpenSim, if we go this route
  • [2009/06/16 10:39] Morgaine Dinova: Pixel:have you ever seen Aimee make a "wild" change? ;-)
  • [2009/06/16 10:39] Zha Ewry: No, but it's a community, with a BDFL

[2009/06/16 10:39] Aimee Trescothick hides under the bench

  • [2009/06/16 10:39] Zha Ewry: Granted the BDFL has a very NEWBish avatar, but still
  • [2009/06/16 10:40] Zha Ewry: Seriously, one of the reasons to do OGPX
  • [2009/06/16 10:40] Morgaine Dinova: Since Philip started off the interop process, I don't see him blocking it now. The 'B' stands for Benevolent .... fingers crossed :P
  • [2009/06/16 10:40] Zha Ewry: and be at the IETF is to get common building blocks for this stuff
  • [2009/06/16 10:41] Zha Ewry: /The nice thing, is this woudl work for anything from

[2009/06/16 10:41] Pixel Gausman decides to fess up

  • [2009/06/16 10:41] Zha Ewry: the current AD/Region split in the OGP spec, to
  • [2009/06/16 10:41] Pixel Gausman: i've been working on porting the existing OGP code to snowglobe
  • [2009/06/16 10:42] Morgaine Dinova: Wooho!!!!!
  • [2009/06/16 10:42] Morgaine Dinova: Pixel++
  • [2009/06/16 10:42] Zha Ewry: pushign a random service onto a cloud host, and having it work
  • [2009/06/16 10:42] Ahuva Heliosense: honesty - the best policy. :)
  • [2009/06/16 10:42] Zha Ewry: (and transparently, I might add)
  • [2009/06/16 10:42] Zha Ewry: Of course, we get to blend in the X.509, and policy stuff when handing off a cap
  • [2009/06/16 10:42] Zha Ewry: but.. you can, if we do this right
  • [2009/06/16 10:42] Zha Ewry: imaine going to the AD
  • [2009/06/16 10:43] Zha Ewry: getting a cap grant, which includes a cloud eventq
  • [2009/06/16 10:43] Zha Ewry: and a bunch fo cloud bnased services sharing that eventq
  • [2009/06/16 10:43] Zha Ewry: and the client never knowing
  • [2009/06/16 10:43] Zha Ewry: (which ought to be compelling to all the players here)
  • [2009/06/16 10:43] Pixel Gausman: the client is happiest that way
  • [2009/06/16 10:43] Zha Ewry: Well, the client might be able to guess that all the
  • [2009/06/16 10:44] Zha Ewry: https://amazon-e3.coulserver67.amazong.com caps might mean somethign weas going on
  • [2009/06/16 10:44] Zha Ewry: But.. it wouldn't care
  • [2009/06/16 10:44] Zha Ewry: (And hopefully, amazon, would spell its URLs correclyt)
  • [2009/06/16 10:45] Zha Ewry: I'm plannong on discussing this broadly, and posing the draft 0.1 of it to the OGXP mailing list
  • [2009/06/16 10:45] Zha Ewry: So..
  • [2009/06/16 10:45] Zha Ewry: feedback is devoutly desired
  • [2009/06/16 10:46] Zha Ewry: Oh, and for fun:
  • [2009/06/16 10:46] Zha Ewry: 1. Abstract away a specific single bidirectional pipe from the OGPX specs

2. Provide a REST/Caps style model for describing resources on the client 3. Permit service side deployers to pass around references to these caps within the service cloud safely and transparently a. Prevent services from spoofing messages to resources others than those they have been granted access to b. Provide a endpoint independent path to the client (via any eventq available, or any URI representing the client resource) c. Preventing guessing of endpoint resource addresses 4. Permit deployers to make deployment time choices about where resources are serviced, and how events are routed within their own services 5. Closely model the current event queue implementations

  • [2009/06/16 10:46] Zha Ewry: Those are the requirements I took on in trying to define this
  • [2009/06/16 10:47] Rex Cronon: zha. your lifespan might not allow u to see all of those become real:)
  • [2009/06/16 10:47] Zha Ewry: Once again I've stunned the crowd into silence
  • [2009/06/16 10:47] Zha Ewry: @Rex, I'm an optimist
  • [2009/06/16 10:48] Rex Cronon: i think u made everbody rofl:)
  • [2009/06/16 10:48] Pixel Gausman: Zha: u r always the optimist. that's why i double every sizing you give me. :)
  • [2009/06/16 10:48] Zha Ewry: heh. Being within 50% is pretty good for a developer/architect
  • [2009/06/16 10:49] Zha Ewry: so.. one concern Pixel expressed, which is very rational
  • [2009/06/16 10:49] Zha Ewry: is "how much cost is there in doing this, client side"
  • [2009/06/16 10:49] Mojito Sorbet: 3a, public key technology on the caps?
  • [2009/06/16 10:50] Zha Ewry: Well, the CAPs style handle, should prevent even needing that
  • [2009/06/16 10:50] Pixel Gausman: my concern is that if we want to move things off of UDP, whatever me move it to has to be quick
  • [2009/06/16 10:50] Infinity Linden: @mojito.. what do you mean? the use of HTTPS, client certs or S/MIME
  • [2009/06/16 10:50] Infinity Linden: ?
  • [2009/06/16 10:50] Pixel Gausman: otherwise nothing will move off UDP
  • [2009/06/16 10:50] Zha Ewry: The caps are one shots, per client sessions only passed on to people you "trust" but. its agood discussion point
  • [2009/06/16 10:50] Mojito Sorbet: Given q choice hbetween quick and reliable...
  • [2009/06/16 10:51] Mojito Sorbet: @infinitty - about services being able to forge cap identifiers.
  • [2009/06/16 10:51] Zha Ewry: So, an array of 20 LLSD messages in a post, vs 100 UDP posts
  • [2009/06/16 10:51] Zha Ewry: They can't Mojito
  • [2009/06/16 10:51] Zha Ewry: becuase the client creatse the UUID per session
  • [2009/06/16 10:51] Zha Ewry: and passes it up to the servcie susing it
  • [2009/06/16 10:51] Zha Ewry: So. the cap, just like the service side ones
  • [2009/06/16 10:51] Zha Ewry: are (pretty much) unguessable
  • [2009/06/16 10:51] Zha Ewry: acutally, even more so, given 1-2 day client sessions
  • [2009/06/16 10:52] Zha Ewry: You can always set the expire time below your best guess at the NSA's current cracking computers
  • [2009/06/16 10:52] Mojito Sorbet: ok,. I forgot the UUIDs were not long-term handles.
  • [2009/06/16 10:52] Pixel Gausman: just sayin, the UDP path is tried and tested. i dont recall anyone doing performance analysis of moving some of the time critical messages to HTTP
  • [2009/06/16 10:52] Zha Ewry: SIP peeps have done well, actually doing that
  • [2009/06/16 10:53] Zha Ewry: But, yes, its a very valid concern
  • [2009/06/16 10:53] Infinity Linden: we've done some of that, but i don't know if we published the results
  • [2009/06/16 10:53] Infinity Linden: i'll see if we can
  • [2009/06/16 10:53] Mojito Sorbet: Some things need to be fast, like me hitting arrows to move.
  • [2009/06/16 10:53] Pixel Gausman: so if SIP design matches ours, we're good. not sure it does
  • [2009/06/16 10:53] Mojito Sorbet: Other things do nopt need to be so fast, but DO need to be reliable.
  • [2009/06/16 10:53] Morgaine Dinova: Pixel UDP+circuits is dead slow compared to TCP which is an optimized kernel stack. Don't compared TCP to raw UDP, that's not the situation we have here.
  • [2009/06/16 10:53] Infinity Linden: some aspects of SIP are noteworthy
  • [2009/06/16 10:53] Infinity Linden: others are not
  • [2009/06/16 10:53] Zha Ewry: So..right
  • [2009/06/16 10:53] Zha Ewry: its 1 http get
  • [2009/06/16 10:54] Zha Ewry: which gets you maybe 50-100 high level messages
  • [2009/06/16 10:54] Zha Ewry: vs 1,000 UDP packets and assembly ans circuit management and retry code
  • [2009/06/16 10:54] Mojito Sorbet: and doing your own acks

[2009/06/16 10:54] Morgaine Dinova nods

  • [2009/06/16 10:54] Infinity Linden: but right. HTTP(S) to establish a connection with a simulator, then using RTP to carry object updates
  • [2009/06/16 10:54] Mojito Sorbet: yuk
  • [2009/06/16 10:54] Mojito Sorbet: Yes.
  • [2009/06/16 10:54] Infinity Linden: RTP _can_ go over either UDP or TCP
  • [2009/06/16 10:54] Mojito Sorbet: Use RTP for the stuff that has to be fast.
  • [2009/06/16 10:55] Infinity Linden: interesting.. HTTP _can_ be fast
  • [2009/06/16 10:55] Infinity Linden: in some cases, faster than RTP
  • [2009/06/16 10:55] Zha Ewry: So...
  • [2009/06/16 10:55] Mojito Sorbet: The limitation is not in the protocol, but in the code on etiher end
  • [2009/06/16 10:55] Zha Ewry: Yes, we need to be sane
  • [2009/06/16 10:55] Pixel Gausman: oh dear. RTP acronym mismatch. i doubt we're talking Research Triangle Park in NC
  • [2009/06/16 10:55] Zha Ewry: but.. we also
  • [2009/06/16 10:55] Mojito Sorbet: We are not trying to serve up web pages here
  • [2009/06/16 10:55] Infinity Linden: but the semantics of RTP are what interest me... "here's some data that's got to get there before XXX time"
  • [2009/06/16 10:55] Zha Ewry: want to be careful about footsteps
  • [2009/06/16 10:55] Zha Ewry: note:
  • [2009/06/16 10:55] Rex Cronon: the question is. what cant match the speed of UDP:)
  • [2009/06/16 10:55] Infinity Linden: RTP == Real Time Protocol
  • [2009/06/16 10:55] Zha Ewry: if we get down to only discardable packets in the UDP stack, we're much happier
  • [2009/06/16 10:56] Rex Cronon: what can*
  • [2009/06/16 10:56] Infinity Linden: if you look at the way that a lot of ISPs tune their networks, you'll find that there's a point where UDP gets dropped in preference to TCP
  • [2009/06/16 10:56] Mojito Sorbet: I like that approach
  • [2009/06/16 10:56] Zha Ewry: so,packets which are "agent stepped forward from x,y,z to x1,y1,z1, at time x"
  • [2009/06/16 10:56] Zha Ewry: are really nice
  • [2009/06/16 10:56] Zha Ewry: Because if you lose one
  • [2009/06/16 10:56] Zha Ewry: You can just sync up without a resend
  • [2009/06/16 10:57] Zha Ewry: "Agent took 1 step forward, packet #1978"
  • [2009/06/16 10:57] Morgaine Dinova: Since almost nothing in SL is discardable, UDP was a bad choice from the start. It's still usable for intersticial movement packets (in between the mandatory ones), but little else.
  • [2009/06/16 10:57] Mojito Sorbet: But it would be nice is we could elimiate messages saying "user is holding down the arrow key"
  • [2009/06/16 10:57] Zha Ewry: you're not so happy
  • [2009/06/16 10:57] Infinity Linden: right. data that doesn't have to be confirmed. that's a good candidate for RTPification
  • [2009/06/16 10:57] Infinity Linden: also
  • [2009/06/16 10:57] Infinity Linden: remember. RTP can go over UDP _or_ TCP
  • [2009/06/16 10:57] Zha Ewry: and where you dno't care about the stack chosing what ets discarded if you go UDP
  • [2009/06/16 10:58] Infinity Linden: though it's usually sent over UDP
  • [2009/06/16 10:58] Zha Ewry: One of the killers of UDP is the total lack of control over what gets discarded
  • [2009/06/16 10:58] Mojito Sorbet: If you put Too much in UDP, would you have to worry about routers timing out? Maybe a keepalive message..
  • [2009/06/16 10:58] Mojito Sorbet: Sorry
  • [2009/06/16 10:58] Mojito Sorbet: if you put too much in TCP, you cant let the UDP channel go dry
  • [2009/06/16 10:58] Zha Ewry: (Murphy assuring you'll often discard the stuff you need, and not the stuff you could ignroe if it got lost)
  • [2009/06/16 10:58] Mojito Sorbet: or else the router in uyour house drops the port association

[2009/06/16 10:58] Zha Ewry sighs

  • [2009/06/16 10:59] Zha Ewry: Yeah, as long as you're doing UGP
  • [2009/06/16 10:59] Zha Ewry: *UDP
  • [2009/06/16 10:59] Zha Ewry: you need, one packet every, 2 seconds?
  • [2009/06/16 10:59] Zha Ewry: Its low.
  • [2009/06/16 10:59] Mojito Sorbet: I do not know what that timeout is
  • [2009/06/16 10:59] Techwolf Lupindo: And some ISP are intentelly messing up UDP to kill voip so they can promote there own voip.
  • [2009/06/16 10:59] Mojito Sorbet: yeah
  • [2009/06/16 10:59] Zha Ewry: So.. feedback
  • [2009/06/16 10:59] Infinity Linden: yeah. one of the many reasons AT&T doesn't route packets for me
  • [2009/06/16 10:59] Zha Ewry: IMs and notecards
  • [2009/06/16 10:59] Zha Ewry: here,
  • [2009/06/16 10:59] Zha Ewry: I'll happily chat with people
  • [2009/06/16 10:59] Zha Ewry: or:
  • [2009/06/16 11:00] Zha Ewry: zha.ewry@gmail.com or dwl@us.ibm.com
  • [2009/06/16 11:00] Zha Ewry: (The gmail is on my profile tab too)
  • [2009/06/16 11:00] Infinity Linden: ack. it's 11:00 already

[2009/06/16 11:00] Pixel Gausman realizes she skipped lunch [2009/06/16 11:01] Zha Ewry gets lunch *before* this starts for a reason ;-)

  • [2009/06/16 11:01] Zha Ewry: Hopefully, this was helpful..
  • [2009/06/16 11:01] Infinity Linden: i have a busy day. nice chatting with everyone. zha... these diagrams are up on a wikipage somewhere, right?
  • [2009/06/16 11:01] Zha Ewry: Expect more discussions, more huge prims with sequence diagrams
  • [2009/06/16 11:01] Rex Cronon: tc
  • [2009/06/16 11:01] Infinity Linden: or even an old webpage
  • [2009/06/16 11:01] Zha Ewry: (You have them all in your e-mail, Inf)
  • [2009/06/16 11:02] Morgaine Dinova: Thanks Zha
  • [2009/06/16 11:02] Pixel Gausman: laters, kiddos
  • [2009/06/16 11:02] Infinity Linden: i was just thinking about community comment
  • [2009/06/16 11:02] Zha Ewry: and I'll post some to the wiki soon
  • [2009/06/16 11:02] Morgaine Dinova: Cya Pixel
  • [2009/06/16 11:02] Zha Ewry: OGXP mailing list and wiki
  • [2009/06/16 11:02] Rex Cronon: i am leaving too
  • [2009/06/16 11:02] Rex Cronon: bye everybody
  • [2009/06/16 11:02] Zha Ewry: but. the groupies get first crack
  • [2009/06/16 11:02] Infinity Linden: cheers, folkies!
  • [2009/06/16 11:02] Morgaine Dinova: Cya Rex
  • [2009/06/16 11:02] Morgaine Dinova: Cyu Infinity
  • [2009/06/16 11:02] Zha Ewry: Hopefully useful Morgaine?
  • [2009/06/16 11:03] Morgaine Dinova: Yeah, this was GREAT stuff!
  • [2009/06/16 11:03] Zha Ewry: it gets out to OGPX next
  • [2009/06/16 11:03] Zha Ewry: but.. I'mt rying to always preview stuff here first
  • [2009/06/16 11:03] Morgaine Dinova: I expect to be thinking about it all week, there's lots of stuff down various alleys
  • [2009/06/16 11:04] Morgaine Dinova: Very cool
  • [2009/06/16 11:04] Zha Ewry: Good
  • [2009/06/16 11:04] Zha Ewry: WB Nyx.
  • [2009/06/16 11:04] Morgaine Dinova: Hi Nyx, bye Nyx :P
  • [2009/06/16 11:05] Mojito Sorbet: Oh, did you find the URL to the most recent OGP proposals?
  • [2009/06/16 11:06] Zha Ewry: oh, one second
  • [2009/06/16 11:08] Mojito Sorbet: I see teleport draft 5, from October
  • [2009/06/16 11:09] Zha Ewry: * OGP : Intro and Requirements (

http://tools.ietf.org/html/draft-hamrick-ogp-intro-00 )

[2009/06/16 11:12] Zha Ewry tries to recall the last Paris standards meeting she was at

  • [2009/06/16 11:13] Mojito Sorbet: Dinner right next to Notre Dame....

[2009/06/16 11:13] Zha Ewry preferes rue moufetard, or some very good stuff near the eifel tower ;)

Retrieved from "https://wiki.secondlife.com/w/index.php?title=AW_Groupies/Chat_Logs/AWGroupies-2009-06-16&oldid=395193"