[06:48:22] --- Cullen Jennings has joined [06:49:19] --- Nicholas Weaver has joined [06:49:29] --- gregorylebo has joined [06:49:33] test message [06:49:36] ack [06:49:59] --- Gonzalo has joined [06:50:04] --- danny has joined [06:50:04] --- danny has left [06:50:05] --- sob has joined [06:51:05] --- danny has joined [06:55:04] --- danny has left: Replaced by new connection [06:55:04] --- danny has joined [06:56:16] --- spencerdawkins has joined [06:56:50] jon/cullen - could you post the current agenda into jabber? [06:57:12] 9:30 - Welcome/Note Well/Agenda Bash 9:45 - Service Provider Perspective (Rich Woundy/Jason Livingood) 10:45 - Application Designer Perspective (Stanislov Shalunov) 11:15 - Lightning Talks & General Discussion Robb Topolski Nick Weaver Leslie Daigle 11:45 - Lunch 12:45 - Localization (Laird Popkin,Yu-Shun Wang, and Vinay Aggrawa) 2:15 - Afternoon Break 2:30 - New Approaches to Congestion Bob Briscoe/ Marcin Matusze 3:15 - Quality of Service Mary Barnes and Henning Schulzrinne 4:00 - Conclusions & Wrap 4:30 - Meeting Ends [06:57:30] --- Lars has joined [06:57:36] --- danny has left [06:57:45] --- danny has joined [06:57:57] --- Magnus Westerlund has joined [06:58:15] BTW - all the slides being presented will be made public but they are not up on a server yet because I don't even have all of them yet [07:00:00] --- Nicholas Weaver has left [07:00:36] --- schulzrinne has joined [07:00:59] mini slot; http://www.cisco.com/warp/public/109/cable_faq_rf.html [http://www.cisco.com/warp/public/109/cable_faq_rf.html] [07:02:08] --- bauer1j has joined [07:06:02] --- Jean-Francois has joined [07:10:55] --- EKR has joined [07:11:57] --- Jean-Francois has left: Lost connection [07:12:58] QoS doesn't matter unless there's contention... ya [07:13:21] --- Nicholas Weaver has joined [07:17:11] --- Cullen Jennings has left [07:19:01] --- Cullen Jennings has joined [07:22:16] --- amalis has joined [07:24:11] --- schulzrinne has left: Replaced by new connection. [07:24:12] --- Lars has left [07:24:26] --- spencerdawkins has left [07:24:31] --- schulzrinne has joined [07:24:33] --- Lars has joined [07:25:55] --- Gonzalo has left: Replaced by new connection [07:25:55] --- gcamaril has joined [07:26:33] --- Magnus Westerlund has left [07:26:35] --- schulzrinne has left: Disconnected. [07:26:47] --- schulzrinne has joined [07:26:49] --- amalis has left [07:27:14] --- danny has left: Replaced by new connection [07:27:14] --- danny has joined [07:27:29] --- gregorylebo has left [07:28:56] --- gcamaril has left: Replaced by new connection [07:28:56] --- gcamaril has joined [07:29:18] A P2P caching infrastructure can eliminate/reduce the need for the upload stream from the DOCSIS domain I think [07:29:34] --- Nicholas Weaver has left [07:30:09] --- gregorylebo has joined [07:30:15] --- amalis has joined [07:30:23] --- danny has left: Replaced by new connection [07:30:23] --- danny has joined [07:30:24] --- danny has left [07:30:37] how big is this problem for non-cable operators/providers, in comparison? [07:31:48] i.e. do other types of providers (T1, ADSL, etc.) see the last mile congestion issue with similar heat, or is it due to the bandwidth sharing properties of cable architectures that makes it so hairy? [07:32:16] Danny made a comment at mic a minute ago - can we make sure that get's captured. We might need Danny to repost it here [07:32:55] Nick: totally agree about caching. [07:33:14] I just don't see how locality per se removes upstream [07:35:22] Locality helps the WAN infrastructure, not the local upstream congestion [07:36:16] --- danny has joined [07:37:09] well, it might help the local upstream in that if a cache is in use, then the same file that an endpoint has doesn't need to uploaded more than once, right? Otherwise, it would be uploaded many times, one for each other peer that wanted it [07:37:57] so, won't help for the first upload, but should cut down on uploads overall, right? [07:38:02] Danny's comment: If you have more localized distribution via caching, or location techniques, or whatever, then burst rates (i.e., peak <> trough ratios) may actually be larger and you may increase HFC contention (note: HFC in particular, v. DSL, etc..) [07:38:13] --- ThinkingCat has joined [07:38:13] --- ThinkingCat has left: Lost connection [07:38:47] well, you could do packet inection into their HTTP connections to put in a banner ad telling htem :) [07:39:09] to Danny: you could do some TCP throttling at the cache to mellow down the last mile congestion, no? [07:39:22] Sure, you could.. [07:39:54] well, you could set their downstream bandwidth to zero. they'd sure notice that [07:40:01] --- Leslie Daigle has joined [07:40:16] greg: but then, the cache has no idea about the HFC network contention... [07:40:32] does the client? [07:41:06] no.. the cable modem does when request physical layer timeslots [07:41:23] i guess what I was brainstorming was that the cache, being hosted by the provider who knows the last mile characteristics, will know the... [07:41:53] optimal settings for TCP throttling to optimize the last mile of that client, no? [07:42:10] no, because there are many other apps, including voip, other p2p, http, flash, etc.. all contending for resources.. [07:43:50] well yes, on this solution thread of thinking, you'd have to set the throttling based upon some optimal consumption rate, where that rate considered spare capacity for the other apps. [07:44:11] to the "how big is this problem for non-cable operators/providers, in comparison?" question -- not an answer, but an observation. The 2 major Canadian ISPs (at least one of which is DSL-based) are doing "traffic shaping" [07:44:17] --- schulzrinne has left: Disconnected. [07:44:20] --- Lars has left [07:44:22] and looking at tiered access [07:44:27] --- amalis has left [07:44:32] --- danny has left: Replaced by new connection [07:44:32] --- danny has joined [07:44:38] --- danny has left [07:44:46] --- Lars has joined [07:45:36] --- gcamaril has left: Replaced by new connection [07:45:36] --- gcamaril has joined [07:46:25] --- schulzrinne has joined [07:48:13] --- danny has joined [07:49:20] --- Leslie Daigle has left [07:49:32] --- Leslie Daigle has joined [07:49:50] The local NAT seems to kick people off the chat every few minutes... [07:50:01] --- amalis has joined [07:50:14] nice [07:50:26] Dave needs better control of his son. [07:50:33] I've not been kicked off (yet) but watch lots of off & on of others [07:50:58] I've been kicked off several times. I'm on the "a" network, if that makes any difference. [07:51:10] If I get kicked off again, I'll try switching to "g". [07:51:11] Might depend on the client (iChat, in my case) and OS [07:51:31] I'm also using iChat [07:51:50] i'm using adium & noticing same [07:53:09] --- Nicholas Weaver has joined [07:54:16] --- Nicholas Weaver has left [07:58:41] He said that the "most popular" client was on his laptop - does anyone know which client he was referring to? [07:59:10] I assume he's referring to bittorrent's client since he works for them. [08:00:51] Leslie: re the Canadian ISPs: is the Traffic shaping aimed at transit across their backbone, or at helping the last mile congestion? My question about non-cable operators was trying to get at the last mile issue that was the focus for much of their preso [08:01:18] Greg: ACK re. the distinction [08:01:23] Let me dig up info re. Canadian [08:01:49] --- weitzner has joined [08:03:26] i was tole today that the congestion point in Canada is the (large layer 2) access net - not the backbone [08:04:31] --- Nicholas Weaver has joined [08:04:41] --- weitzner has left [08:05:15] I don't have concrete data -- but this article is informative: http://www.theglobeandmail.com/servlet/story/RTGAM.20080325.wgtinternet26/BNStory/Technology [08:05:42] --- weitzner has joined [08:05:46] Well, for a lot of the users you have teh "Install dees stuupid software" [08:05:48] Essentially: Bell is claiming "p2p is clogging our network for our customers", and it is throttling for their customer ISP networks (to SOB's point) [08:06:03] 50% is not good enough, BTW. 50% means you still have as much transit as UNCACHED HTTP [08:06:30] --- Magnus Westerlund has joined [08:06:42] ping [08:06:44] because you have the 2x magnification because you have to do the uplink [08:06:46] pong [08:06:54] thx [08:07:20] (oh, I meant the "stoopid softwaer" for user notification. You can have NEW customers have a notification agent that will mostly get installed, but "rodgering" the web surfing is probably useful) [08:08:48] (the problem is also not the POPULAR torrents, but the UNPOPULAR ones. If you only have one user on a torrent in your ISP, no caching in the world removes theproblem that the user needs to upload as much as he downloads) [08:10:04] one could argue that this might cost some ISPs more.. e.g., cable operators that don't own their own backbone but use the same AS in multiple locations... [08:14:15] --- spencerdawkins has joined [08:16:55] got it.. [08:17:14] Caches for P2P are ALWAYS worse than caches for HTTP from the ISP viewpoint, however [08:17:21] so if it's not p2p and its centralized then other "peers" aren't using bw [08:17:34] didn't understand that argument [08:17:39] got it now.. [08:21:06] --- Jean-Francois has joined [08:21:16] Why not use any of the many friendlier-than-tcp already implemenetde protocols? [08:21:38] well, in this case the cache is looks to the rest of the peers like just another P2P peer [08:21:50] just that it has LOTS of storage and bandwidth [08:21:55] A cache for generic BitTorrent is asking for a lawsuit [08:22:10] And any ISP would make the cache only peer with the local nodes and leech from remote nodes [08:22:22] add something like Ono, something to optimize the cache choice to a close (closest?) peer/cache, and we make headway, no? [08:22:25] nicolas: agree, and then folks wouldn't use it.. [08:22:59] danny: Well, being the ISP, you can use DPI to detect the torrent request, AND snarf the unrequested stuff, AND play lying games between peers [08:23:12] Its just not worth the lawsuit to do [08:23:36] or, the ISPs work out the policy/legal issues and promote that their network offers faster P2P service than others, where as others are crushing P2P, and the churn turns in their favor [08:23:41] Oh, AND use DPI to get unencrypted blocks off the wire [08:23:55] gregorylebo: You want to be the test case? [08:24:20] lawyers are couple hundred K /yr, bandwidth upgrades all over the network are millions. ROI is there to solve it in the policy/legal arena... [08:24:52] so let's not drop the idea due to blanket "policy / legal" issues argument that we don't fully understand and definitely don't control [08:24:59] yes, this enables RIAA and IFPI to require ISPs to implement policy [08:25:03] also, why should Comcast/AT&T risk it? They have strong incentives to be close with the MPAA because they are also content providers [08:26:36] and therefore they will have a summitt, like this one, for the legal/policy side and find a solution. Why? Because it behooves ALL of them to get the content out there. Content distribution is a positive. P2P is merely a mechanism for rapidly doing it. Used in the right way, it is a the best thing that can happen for the content industry [08:27:07] for the _Legal content industry, yes.. [08:27:24] the problem is, they'd now have a platform to go after the illegal content [08:27:32] you can't stop the illegal content industry ever, so that's a non-goal, right? [08:28:00] there's a difference between ignoring it and enabling it.. [08:28:01] gregory: You can do a big bite out of open-world piracy however [08:28:22] http://nweaver.blogspot.com/2008/01/security-thought-at-copyright-fighting.html [08:28:36] caching arguably enables it, we've already seen the _legal content folks argue this with DPI, give them caches.. [08:28:43] I'm finding myself arguing vehemently for something I really don't have that much passion about, so I'm going to mellow out now [08:28:55] --- hawkinsw@jabber.org has joined [08:28:55] [08:29:38] he [08:29:43] you do that a lot.. [08:31:23] Why hasn't anyone done the simple BitTorrent client that just throttles down when YOU do other stuff. It would perform really well. :) [08:32:05] *agrees* [08:33:31] --- hawkinsw@jabber.org has left [08:37:31] agrees 2 [08:37:56] 3 min left [08:38:37] Some clients can be easily configured so that it only uses the BW that you want it to (upstream and downstream separately) [08:38:51] Its 3-4 PER TORRENT [08:38:59] Users will often open up multiple torrents (we've witnessed it) [08:39:09] it's google maps and itunes that nuke it.. _then p2p apps [08:39:23] Also, if it ISN'T trying to use all the bandwidth, then why are the bitorrent people complaining about full buffers? [08:39:42] on the send side [08:40:34] umm, yeah, that's sorta implicit in the process, methinks.. understand the problem first [08:42:12] 250 GB per month is a HELL of a lot of bandwidth [08:42:16] You *might* be able to find most the slides in ZIP file called P2PI-Slides.zip at http://idisk.mac.com/cullenfluffyjennings-Public?view=web [08:43:15] time check? [08:43:23] Thats 1.8 hours/day of 10 Mbps download every day [08:43:45] time check .. agenda contention.. cheap fix.. [08:44:03] HEY - TIME CHECK [08:44:08] the chair are weak [08:44:12] sheesh.. [08:44:24] could you ask him to summarize? [08:44:50] i think that's my presentation [08:44:58] (it's much shorter ;-) ) [08:45:48] Got the slides - thanks! [08:46:05] thanks - good to know it sort of worked [08:46:42] Worked fine, not just "sort of" [08:47:01] And this is on WinXP, not a Mac :-) [08:47:04] how is it perfectly topologically aware? [08:47:17] in-band cached.. [08:47:31] So, there's another reason for P2P: Liability shifting [08:47:43] ekr: yeah, like that [08:49:52] ++nick [08:51:24] A perfectly topologicallyaware P2P app: Only sends ONE copy up and down any given link [08:51:44] per AS if you have a cache [08:51:51] Per user if you don't [08:52:05] with minimum transit distance. [08:53:53] --- Lars has left [08:53:53] --- danny has left [08:53:56] --- danny has joined [08:53:57] --- schulzrinne has left: Disconnected. [08:54:01] --- EKR has left [08:54:08] --- spencerdawkins has left [08:54:12] --- danny has left [08:54:16] --- Nicholas Weaver has left [08:54:24] --- amalis has left [08:54:27] --- Cullen Jennings has left [08:54:31] --- Lars has joined [08:54:46] --- bauer1j has left: Computer went to sleep [08:54:47] --- Lars has left [08:55:33] --- gcamaril has left: Replaced by new connection [08:55:45] --- Jean-Francois has left [08:56:31] --- weitzner has left [08:58:25] --- Leslie Daigle has left [09:00:55] --- sob has left [09:05:15] --- Magnus Westerlund has left [09:36:31] --- amalis has joined [09:44:55] --- gregorylebo has left [09:45:11] --- amalis has left [09:47:05] --- amalis has joined [09:51:19] --- amalis has left [09:51:19] --- sob has joined [09:51:19] --- amalis has joined [09:51:21] --- Nicholas Weaver has joined [09:51:27] --- EKR has joined [09:51:57] --- danny has joined [09:51:59] --- schulzrinne has joined [09:52:29] --- Lars has joined [09:52:58] --- lisa has joined [09:54:15] --- gregorylebo has joined [09:54:47] --- Magnus Westerlund has joined [09:55:24] --- Lars has left [09:56:01] --- Cullen Jennings has joined [09:56:05] --- Lars has joined [09:56:30] All of the slides are now available in P2PI-Slides-All.zip file at http://idisk.mac.com/cullenfluffyjennings-Public?view=web [09:57:17] --- Cullen Jennings has left [09:58:00] --- weitzner has joined [09:58:08] --- Jean-Francois has joined [09:58:43] --- weitzner has left [09:58:44] --- Lars has left [09:58:48] --- schulzrinne has left: Disconnected. [09:58:48] --- danny has left [09:58:51] --- danny has joined [09:58:53] --- amalis has left [09:58:53] --- lisa has left: Replaced by new connection [09:58:53] --- lisa has joined [09:59:03] --- danny has left [09:59:33] --- gregorylebo has left [09:59:45] --- sob has left: Replaced by new connection [09:59:47] --- sob has joined [09:59:48] --- Lars has joined [09:59:49] --- lisa has left: Replaced by new connection [09:59:49] --- lisa has joined [10:00:00] --- amalis has joined [10:00:45] --- Lars has left [10:00:55] --- Lars has joined [10:02:36] --- EKR has left [10:02:36] --- EKR has joined [10:02:46] --- Leslie Daigle has joined [10:03:38] --- gregorylebo has joined [10:04:41] --- Magnus Westerlund has left [10:06:12] --- danny has joined [10:06:52] ping [10:07:01] pong [10:07:49] pang [10:08:01] --- schulzrinne has joined [10:08:34] . [10:09:08] --- JF has joined [10:09:55] where is the P4P-WG? [10:10:02] ie. consortium only? [10:11:00] Uhh, is there any significant non-infringing use for trackerless P2P? [10:13:02] --- schulzrinne has left: Disconnected. [10:13:02] --- lisa has left [10:13:05] --- Lars has left [10:13:10] --- lisa has joined [10:13:17] --- amalis has left [10:13:18] no. This has been another issue of simple answers to simple questions [10:13:29] --- Lars has joined [10:15:16] --- robb has joined [10:15:26] --- amalis has joined [10:16:42] --- Magnus Westerlund has joined [10:16:51] Amusingly I just got this email on a WebDAV mailing list: [10:16:53] My ISP has just announced that it is abandoning its 20-year-old web-cache, on the grounds that: - In a broadband world, end-users don't perceive much benefit - In a world of SSL, web-apps and sessions, caches don't work anyway - You can't seriously expect to cache any useful propertion of  multimedia content The impact of web-caches on global bandwidth usage is presumably marginal, nowadays; HTTP page requests must be a pretty small fraction of general bandwidth usage. [No stats to hand - ed.] -- Jack. [10:17:01] --- Jean-Francois has left: Disconnected [10:17:23] greg - i think it starts here http://www.dcia.info/activities/p4pwg/membership.html [10:17:44] There always seem to be huge differences in peoples' beliefs how well caching works even in long-existing systems. [10:18:09] --- Leslie Daigle has left [10:18:12] yep.. caching infrastructure vendors believes it works best [10:18:33] lisa: Bret Glass w ould disagree. You can ask him for his cache effectiveness, but its a big deal for him (OTOH, he bans P2P pretty heavily) [10:19:02] JF: thx [10:19:11] BitTorrent has a choke-and-unchoke and sampling technology that finds the best reciprocating peers. When they're not the closest peers, there generally is a reason (congested link, overtaxed CPU on node, bad client configuration, etc). [10:19:37] --- sob has left [10:21:31] Brett Glass does not have an interest in studying this problem, he has made up his mind and any facts he has will only support his views. [10:22:58] But for any cache, it can only be as effective as its size and users' willingness to use it. [10:27:48] --- Leslie Daigle has joined [10:29:17] If you build using the cache into the client, then the user doesn't need to think about it. [10:29:23] --- Paul has joined [10:30:12] but the isps don't own the end systems, and these are isp-centric problems [10:30:23] wtf? [10:30:38] The answer to that is *crypto* [10:30:48] ja [10:30:58] Slides are at http://www.funchords.com/p2pi/slides/ [10:32:05] --- Cullen Jennings has joined [10:32:16] Repost of agenda by request ... [10:32:17] 9:30 - Welcome/Note Well/Agenda Bash 9:45 - Service Provider Perspective (Rich Woundy/Jason Livingood) 10:45 - Application Designer Perspective (Stanislov Shalunov) 11:15 - Lightning Talks & General Discussion Robb Topolski Nick Weaver Leslie Daigle 11:45 - Lunch 12:45 - Localization (Laird Popkin,Yu-Shun Wang, and Vinay Aggrawa) 2:15 - Afternoon Break 2:30 - New Approaches to Congestion Bob Briscoe/ Marcin Matusze 3:15 - Quality of Service Mary Barnes and Henning Schulzrinne 4:00 - Conclusions & Wrap 4:30 - Meeting Ends [10:32:40] Cullen -- I got the slides up -- http://www.funchords.com/p2pi/slides/ [10:32:48] this guys has 45 min more? [10:32:56] ahhh [10:33:03] --- lisa has left [10:33:36] --- sob has joined [10:33:53] dnssec [10:34:09] bcp 38 [10:36:22] --- weitzner has joined [10:44:45] --- Cullen Jennings has left [10:47:40] --- Cullen Jennings has joined [10:48:42] might be just a wee bit of copryright IP problem with the name of this thing, eh? [10:49:10] ;-) [10:50:04] :-D possibly, but the use of the name oracle as any internet service that knows something is quite established. It is the last O in Y.A.H.O.O. [10:53:10] --- weitzner has left [10:53:15] --- Leslie Daigle has left [10:53:20] --- Leslie Daigle has joined [10:53:24] --- robb has left [10:53:37] --- Robb Topolski has joined [10:53:40] --- schulzrinne has joined [10:53:49] --- Lars has left [10:54:02] Good idea Nicholas [10:54:37] Azureus and uTorrent already can do this. It's on by default in Azureus. [10:55:04] --- Paul has left: Replaced by new connection [10:55:04] --- Paul has joined [10:57:49] so, there seem to me to be 3 questions here (for all the presentations in this session): [10:57:51] . how much of the P2P traffic reduction they would produce . whether that would solve the larger problem of dealing with . whether they would in fact be deployed/deployable [11:01:50] also: whether they solve the right problem, whether they impact task performance significantly, whether they affect user privacy, whether they affect user choice [11:03:09] --- Lars has joined [11:03:18] well, yes :-) [11:03:33] the "right problem" was hwat I was getting at in the second bullet (which appears to be a fragment) [11:03:40] --- EKR has left [11:03:41] --- hawkinsw@jabber.org has joined [11:03:50] I want to believe user privacy & choice would be part of "deployability", but admit it probably isn't. sigh [11:08:01] --- Robb Topolski has left [11:08:13] --- Robb Topolski has joined [11:10:13] --- EKR has joined [11:10:18] --- Robb Topolski has left [11:12:29] --- Robb Topolski has joined [11:13:30] -test- [11:13:42] test received [11:15:34] tracker stickiness... that's the solution.. [11:15:40] now, what was the problem again? [11:16:40] the problem is for the IETF to be seen as being involved in the real Internet world of today? [11:16:57] yes... i get that... [11:17:30] we need to add bittorrent to the successful protocols slides [11:17:38] :-) [11:19:00] --- Lars has left [11:20:09] Okay, someone tell me what is "layer 8" and "layer 9" ? [11:20:24] politics/law/ etc [11:20:43] above the application (layer 7) [11:21:46] ah, thanks! [11:21:54] http://blog.layerate.net/?p=1 [11:23:31] Thanks danny [11:23:54] --- weitzner has joined [11:24:23] yeah, i'm certain you're far more informed after reading that :-) [11:24:42] --- Robb Topolski has left [11:27:35] --- Leslie Daigle has left [11:27:35] --- schulzrinne has left: Disconnected. [11:27:40] --- Leslie Daigle has joined [11:28:00] --- weitzner has left [11:28:14] --- hawkinsw@jabber.org has left [11:29:48] HAHAHAH. "When the non paranoid go 'here's how to hack it', you have a problem" [11:30:51] The ISP could actually probably do a "forced free" localization... [11:30:58] --- Paul has left: Replaced by new connection [11:30:59] --- Paul has joined [11:31:01] Do an IDS to find tracker requests... [11:31:30] Have it go to a modified client which communicates with the local peers, does NOT claim to have any content (so cacheless), but tells the clients about the other clients [11:31:48] But that might still be suiable. :( [11:33:14] --- JF has left: Computer went to sleep [11:33:28] --- Cullen Jennings has left [11:35:18] --- Nicholas Weaver has left: Computer went to sleep [11:37:25] --- EKR has left [11:38:10] --- Magnus Westerlund has left [11:39:09] --- sob has left [11:48:46] --- Cullen Jennings has joined [11:48:55] --- EKR has joined [11:49:21] --- Nicholas Weaver has joined [11:49:52] poof [11:49:58] --- sob has joined [11:51:20] --- JF has joined [11:51:58] --- Lars has joined [11:53:03] --- don.waterloo has joined [11:53:34] --- Cullen Jennings has left [11:53:40] --- EKR has left [11:53:40] --- EKR has joined [11:53:50] --- EKR has left [11:53:51] --- EKR has joined [11:55:35] --- schulzrinne has joined [11:55:49] --- weitzner has joined [12:03:58] --- Robb Topolski has joined [12:11:16] --- Lars has left [12:11:16] --- Nicholas Weaver has left [12:11:16] --- Robb Topolski has left [12:11:22] --- schulzrinne has left: Disconnected. [12:11:27] --- schulzrinne has joined [12:11:31] --- EKR has left [12:11:36] --- weitzner has left [12:11:57] --- Leslie Daigle has left [12:12:06] --- Leslie Daigle has joined [12:12:21] --- Robb Topolski has joined [12:12:57] --- EKR has joined [12:12:58] --- Lars has joined [12:13:01] --- Paul has left: Replaced by new connection [12:13:01] --- Paul has joined [12:14:16] --- EKR has left [12:14:16] --- EKR has joined [12:22:40] --- Nicholas Weaver has joined [12:23:08] --- spencerdawkins has joined [12:26:54] That's a blank page to me Cullen [12:27:22] the chat dropped almost everyone a bit ago [12:27:43] --- JF has left: Disconnected [12:29:04] --- gregorylebo has left [12:35:08] --- gregorylebo has joined [12:36:09] --- Cullen Jennings has joined [12:37:06] --- don.waterloo has left [12:37:32] --- don.waterloo has joined [12:39:56] --- don.waterloo has left [12:42:52] --- weitzner has joined [12:46:42] --- weitzner has left [12:48:16] --- Cullen Jennings has left [12:48:52] --- Robb Topolski has left [12:49:29] chat drops, chaps restart [12:50:32] --- Cullen Jennings has joined [12:51:04] --- Robb Topolski has joined [12:54:44] --- schulzrinne has left: Disconnected. [12:54:59] --- spencerdawkins has left [12:55:00] --- Robb Topolski has left [12:55:59] --- Paul has left: Replaced by new connection [12:55:59] --- Paul has joined [12:56:27] --- weitzner has joined [12:57:25] --- gregorylebo has left [12:57:25] --- weitzner has left [12:57:40] --- amalis has left [12:58:10] ping [12:58:19] pong [12:58:20] --- Robb Topolski has joined [12:58:23] thanks! [12:58:26] --- amalis has joined [13:00:16] --- gregorylebo has joined [13:03:14] on the DSCP preso (last one) the comment was made that you simply mark the traffic statelessly at the provider edge... [13:03:21] how do you know which traffic to mark? [13:03:57] if the P2P app "hides" or tunnels in http or some other proto, or an unknown port, then you have to use DPI again to really identify it [13:04:05] I expect the marking would be done (or redone) by the 1st hop [13:04:12] router in the organization [13:04:15] Nah, you just do the comcast hack: Anything beyond a threshhold is no longer priority [13:04:17] which org? [13:04:24] the org where the host is? [13:04:42] You mark the traffic that you want to get the non-default handling. Example: VOIP would probably be marked with the Expedited Forwarding marks [13:04:45] --- Lars has left [13:04:52] the org in which the sender sits - e.g., your cable modem [13:04:58] nic: per flow? then you are into state keeping [13:05:08] or at least counter keeping, per 5 tuple [13:05:16] a: don't be afraid of state, its not THAT bad at the edge (1 Gbps is pretty tolerable) [13:05:20] b: You do it per user. [13:05:22] no - you are not keeping state - you decide to mark based on the application [13:05:25] (usel == IP) [13:05:25] agree with nicholas -- after 200 MB in a day (or pick a different limit), stop acknowledging the EF markings [13:05:27] sob: so you upgrade all the cable modems? [13:05:47] nic: I make routers. State costs a lot, vs stateless [13:05:52] at this point the end system (your PC) coud do the marking and [13:06:07] the cambel modems (I think) will ignore teh marks [13:06:08] nic: great for me, it means more revenues, but it doesn't necessarily work a huge scales [13:06:10] For the edge, 1 Gbps, a PC works pretty well [13:06:17] edge is okay, but will cost more [13:06:23] --- amalis has left [13:06:35] why would you believe the endpoint? [13:06:39] sob: yes.. it's timeslots for inserting bytes, not frames [13:06:40] last was to sob: [13:07:09] --- Robb Topolski has left [13:07:11] A simple decaying count would actually work well. Give the service levels a bucket, when [13:07:13] edge may cost more but the edge is the only place that knows that a stream is "important" or "not important" [13:07:16] nic: how many hosts can you support with 1 Gig at the edge? [13:07:18] the bucket fills, just downgrade the packets [13:07:31] gregory: Do you assume statistical multiplexing or not? [13:07:37] I don't know, really. [13:07:40] Edges are usually much more than that in aggregate, or else it's not economically feasible [13:07:41] the question of believeing the endpoint has been a discussion from day 0 in the diffserve world [13:07:53] But I look at UC Berkeley, and the entire net link is 2x 1 Gbps [13:08:02] (although its being upgraded to 10 Gbps). [13:08:08] and thats a lot of users [13:08:19] there is no easy answer for the home - somewhat easer for teh enterprise [13:08:38] nic: downgrade the user (read: source IP) as a whole, yes. Per flow intelligently, no. User gets whacked. [13:09:24] I'd ask Brett Glass how well it works to actually do per flow intelligently, as he does at Lariat to make VoIP work in his WISP [13:09:25] UC Berkeley, in all its glory, is a small part of a community compared to an average pop for a provider [13:10:04] I stand corrected then [13:10:09] sob: for the enterprise: so the router/switch/network-box needs to do the intelligent application identification [13:10:18] and it still involves state [13:10:26] if you don't believe the host, which you shouldn't [13:11:23] that is one way - Microsoft had code in XP that let the host do teh determination and then ask the local router for what DSCP to mark the packets (then the router could police the actual packets) [13:11:53] is that a usage of UPnP? [13:12:33] yes, host signaling to the network for a code is one way, but [13:12:42] can the network then trust the host? [13:13:06] I ask for a marking code for bittorrent [13:13:11] no - this was pre UPnP [13:13:11] no - this was pre UPnP [13:13:13] network says 07 [13:13:17] i use 01 [13:13:20] then what? [13:13:56] if you use 01 the router knows that you were not given 01 as the code for anything so it drops thos packets or it zaps the DSCP [13:14:05] Unless we let go the requirement that the network enforces policy, i.e. unless we decide to trust the hosts' markings, then [13:14:29] the network needs to accurately ID the traffic flow / app in order to enforce the right marking and thus enforcement [13:14:33] the Microsoft assumption was that the router was an enforcement agent [13:14:52] agreed. Router = network in my thoughts here [13:14:56] --- Paul has left [13:15:08] but look at my example of the host asking, getting 07, then using 0 [13:15:10] 01 [13:15:20] but note that Microsoft removed the code at some point [13:15:20] see the issue? [13:15:38] just because the host asked doesn't mean he is trustworthy [13:16:09] if 07 is less than best effort that would be a problem - if 07 is special (better) handling for bittorrent then using 0 would provide worse service [13:16:38] sorry, 07 is best effort (in this example) and 01 is for voip [13:16:53] --- Nicholas Weaver has left [13:16:55] host is sending file sharing [13:17:01] no - the host is not trustworthy - the key is to set things up so that the user is incented to do the "right" thing [13:17:11] ahhh [13:17:31] again this is far easier in an enterprise with good network managers than for home users [13:17:34] well, i'll think about incentives, like free chickens for x years of honest packet marktings... [13:17:49] meanwhile, I have a flight, so disconnecting, [13:17:51] I think I'd learn to not like chicken [13:17:58] LOL [13:18:03] good exercise [13:18:05] more later [13:18:10] --- gregorylebo has left [13:18:10] --- gregorylebo has joined [13:19:22] --- gregorylebo has left [13:20:02] --- schulzrinne has joined [13:23:59] --- Nicholas Weaver has joined [13:25:44] scavenger class is generally not economically viable [13:26:02] because it is too slow for "get it now", and if you want to "get it tomorrow", Netflix is CHEAP [13:26:05] --- Cullen Jennings has left [13:27:54] If I remember the I2 experience slides correctly, the delays were actually fairly comparable to BE, but self-interference was smaller for large (physics data set) transfers [13:28:52] Skype is able to bill because it bills at the gateway (skypeout/skypein) which are under the corporate control [13:29:51] --- schulzrinne has left: Disconnected. [13:31:07] but scavenger class could be very useful within a network even if the user never selects (or sees) it [13:34:48] --- EKR has left [13:42:36] --- sob has left [13:48:04] --- danny has left [13:49:20] --- Nicholas Weaver has left [13:53:30] --- Leslie Daigle has left [14:27:40] --- schulzrinne has joined [14:28:32] --- schulzrinne has left: Lost connection [16:24:48] --- Robb Topolski has joined [16:45:29] --- Robb Topolski has left