[05:02:41] lawrencestewart joins the room [05:50:04] Warren Harrop joins the room [05:53:36] test [06:02:44] dtbphayes joins the room [06:24:07] lawrencestewart leaves the room [06:32:46] Lawrence Stewart joins the room [07:01:22] Andrew McGregor joins the room [07:01:48] JLCjohn joins the room [07:02:27] Matt.Mathis joins the room [07:02:51] ꈲ joins the room [07:03:03] I am jaber scribe [07:04:15] slide: meetings [07:05:01] new slides: MulTFRC [07:05:51] S; what is mulTCP [07:06:57] Is there anybody on jabber not in the room? [07:07:12] yes [07:07:19] tnx [07:08:10] I'm remote as well [07:08:14] s: research background [07:09:33] s: draft history [07:10:48] Matt Zekauskas joins the room [07:11:38] s:overview of changes [07:12:23] s:s: section 2.1 [07:12:31] s: section 2.2 [07:13:43] s: appendis: X_Bps [07:14:10] s:next steps [07:15:53] New slides: CAIA [07:16:12] News from CAIA's NewTCP Project [07:16:32] Lars joins the room [07:16:39] David Hayes is here to [07:17:12] We know where you live Michael [07:17:13] S: Free BSD As A Research Platform [07:17:23] but he is not watching [07:19:22] lawrence: please tell me you are leaving the default in BSD at NewReno :-) [07:19:37] Lars: Absolutely [07:20:04] s: 2nd same title [07:20:51] s: delay-based CC [07:21:28] curious... will there be an audio recording of the RGs like the WGs? I may want to go listen if so... [07:21:36] yes [07:21:47] gorryf joins the room [07:21:48] Matt Zekauskas in nmrg for now [07:23:07] s: DBCC continued [07:24:35] nordmark joins the room [07:24:36] jpc@jabber joins the room [07:24:42] s:graph [07:25:22] s:DBCC continued [07:25:33] s: DBCC continued [07:27:00] s: further information [07:27:09] s: ack's [07:27:20] s: references [07:28:26] Fred Baker speaking [07:30:48] Tell him he did great [07:31:35] Next: Andreas Petlund [07:31:41] Thanks for scribing Matt [07:31:52] np [07:32:12] anywhere he wants [07:32:53] can you hear the guy remotely? [07:32:55] S; TCP modifications (title) [07:33:11] s: Anarchy server side trace trace ana [07:33:33] yup but I think we're on a 10-20s lag [07:33:37] ok [07:34:36] s: interactive thin streams over TCP [07:34:58] s: + yellow box [07:35:21] s: +yellow bandwidth [07:35:50] thomas.dreibholz joins the room [07:35:53] s: analysis of TCP for thin streams [07:36:29] s: thin stream need help [07:36:50] (Bob Briscoe) [07:38:18] s: timeouts [07:38:45] +graph Linear timeout only for thin streams [07:38:58] s: fast retransmissions [07:39:11] + art [07:39:57] s: redundant data bundeling [07:40:12] +art [07:41:03] s: thin stream detection [07:43:17] s:test results [07:44:20] mscharf joins the room [07:44:44] multiple slides w/ common titles [07:45:36] s: fairness [07:46:28] s: box: unmodified TCP [07:46:34] Warren Harrop leaves the room [07:46:59] box: unmodified TCP [07:47:22] s: fairness [07:47:27] gorryf leaves the room [07:47:29] s: questions [07:50:44] gorryf joins the room [07:50:51] Nest: Yuchung Cheng [07:50:53] Lawrence Stewart afk [07:51:27] Seeding TCP RTO with SYN RTT [07:51:39] s: TCP RTO seeding [07:52:42] s: latency problems of RTO seeding [07:53:41] s: motivation [07:54:51] s: measurement and experiments [07:55:27] s:Meas study (graph) [07:58:02] s:meas exp [07:59:02] s: wevserver exp [08:00:18] s: conclusion [08:02:55] MM: are you up next? [08:03:28] Maybe [08:04:19] can somebody else please scribe during my talk? [08:08:59] Fred Baker at Mic: How do they know there is a remote bottleneck? [08:11:06] Tim Shepherd [08:11:10] Carsten Borman [08:12:04] Aaron Falk [08:14:36] Aaron Falk [08:15:41] Bob Briscoe [08:17:24] nordmark leaves the room [08:18:23] Aaron [08:20:07] AAron [08:21:56] Briscoe [08:22:56] Tim Shepherd [08:24:07] nordmark joins the room [08:24:42] wonder at what point we should simply be doing mallman's jump start [08:26:24] also, with regards to matt's last slide: in soho networks, the host can't make predictions based on it's 1st hop link - it's the 2nd hop past the home router that is the issue (access network) [08:27:07] - well the thing about jump-start and other such schemes (which I really like when we look at them) is that they ALSO jump back when they get it wrong, and this influences subsequent cwnd choice [08:27:11] <ꈲ> Audio OK? [08:27:28] yep [08:29:48] apetlund joins the room [08:34:11] apetlund leaves the room: offline [08:34:18] Andreas Petlund joins the room [08:34:32] Has Ilpo explained what stack is used in the experiment? [08:34:32] Alexander Zimmermann joins the room [08:34:54] i'd be shocked if ilpo used anything other than the latest linux [08:35:07] but, ask him [08:35:17] Yes, but he has "simulation" in the title... [08:35:21] ah [08:35:30] nscraddle? [08:39:23] Matt.Mathis leaves the room [08:39:44] thomas.dreibholz leaves the room [08:41:37] Matt.Mathis joins the room [08:47:54] next: nandita [08:48:28] Increasing TCP initial window [08:51:23] s: new contributions and the questions addressed [08:51:39] Matt Zekauskas leaves the room [08:51:39] Matt Zekauskas joins the room [08:52:52] s:improved methodology [08:53:31] if we do any of this IW increase, we need a REALLY good argument i the document for why more is not better. i can see implementors go "well, if 16 is better an 10, and 10 is better than 3, then surely must be better than 16." [08:55:25] you can tell whether folks are form the EU or US by whether they say "i-vee" 10 or "i-dub" 10... [08:56:08] s: africa traffic [08:56:36] s: impact IW10 on Affrica [08:57:08] Yes, why not more than 16 [08:59:12] s: why imtovement in Africa [09:00:09] s; with plots [09:01:27] dtbphayes leaves the room [09:04:05] nordmark leaves the room [09:04:09] nordmark joins the room [09:04:46] For anyone that's interested, we just pushed all the code updates that were mentioned in the CAIA slides to http://caia.swin.edu.au/urp/newtcp/tools.html [09:05:11] Running a bit behind schedule but oh well [09:05:24] I am about to drop off due to power problems [09:06:27] s:IW south america [09:06:36] Matt.Mathis leaves the room [09:06:56] s:latency probs in south america [09:07:19] s: latency under packet loss [09:08:51] Lars leaves the room [09:08:51] Lars joins the room [09:09:09] s:iw3 vs iw10 [09:10:00] s:experiments with higher IWs [09:10:26] Lawrence Stewart leaves the room [09:10:49] s: comparison iw=3,10,16 [09:11:10] For more information on our thin-stream work, the relevant publications are available here: http://simula.no/research/nd/publications/Griwodz.2006.1 , http://heim.ifi.uio.no/~paalh/publications/files/lcn06.pdf and http://simula.no/people/apetlund/bibliography [09:12:25] s:comparison iw=3,10,16 in data center 2 [09:13:48] Matt Zekauskas leaves the room [09:13:49] Matt Zekauskas joins the room [09:14:13] s: tesbed topology [09:17:09] ah netperf [09:17:21] use flowgrind :-) [09:18:54] s: fairness between iw10 and iw3 [09:20:43] s: next slide, same title [09:21:23] :s conclusion [09:21:59] s: jump to "more results from testbed" [09:22:49] A difference between the presented fairness results and my difference is as follows: I didn't use fixed flow lengths, but a realisting flow size distribution with many short flows, too. For flow lengths <15K, IW=10 is not unfair at all... [09:33:04] The red line is M. Allman's Jump-Start proposal, which tends to be worse than IW=10 [09:33:38] Even though it uses rate pacing... But I couldn't study all possible combinations. [09:34:32] traces: see "TCP evaluation suite" [09:36:26] to Micahel: What was the IW with jump-start? [09:36:44] Andrew McGregor leaves the room [09:36:50] Andreas Petlund leaves the room: offline [09:37:59] In Jump-Start, the IW depends on the amount of data queued in the stack. In the original paper, it can basically very large. In my experiments, I limited the amount of data sent in the first RTT to 64K. [09:38:40] I also set the limit to 15K in some experiments, to compare IW=10 with a scheme that uses rate pacing. There is not much difference then. [09:39:00] But, note: All this is simulation only. [09:40:30] Alexander Zimmermann leaves the room [09:41:51] Ok - just working this through ... and it's not an easy comparison. [09:42:15] Gorry: If you have any questions, send them by e-mail. I have to get offline now. [09:42:56] to MS: Understood, and we'll be in touch, Gorry ! [09:43:09] mscharf leaves the room [09:45:57] jpc@jabber leaves the room [09:51:39] nordmark leaves the room [09:56:12] Matt Zekauskas leaves the room [09:59:40] ꈲ leaves the room [10:10:30] gorryf leaves the room [10:13:57] Lars leaves the room [10:31:11] ꈲ joins the room [10:37:43] ꈲ leaves the room [10:39:34] ꈲ joins the room [10:41:36] Andrew McGregor joins the room [10:43:21] Andrew McGregor leaves the room [10:44:44] ꈲ leaves the room [10:47:28] Andrew McGregor joins the room [10:48:02] Andrew McGregor leaves the room [10:52:38] ꈲ joins the room [10:55:40] JLCjohn leaves the room [10:58:45] ꈲ leaves the room [11:00:52] nordmark joins the room [11:03:08] ꈲ joins the room [11:03:52] nordmark leaves the room [11:19:17] ꈲ leaves the room [11:21:33] ꈲ joins the room [13:53:12] ꈲ leaves the room