IETF
tcpm@jabber.ietf.org
Monday, November 10, 2014< ^ >
jishac has set the subject to: TCP Maintenance and Minor Extensions WG
Room Configuration
Room Occupants

GMT+0
[18:32:01] yoshifumi.nishida joins the room
[18:41:10] jlcJohn joins the room
[18:50:19] Michael Scharf joins the room
[18:55:46] PasiS joins the room
[18:56:15] John Leslie joins the room
[18:57:06] David Borman joins the room
[18:57:24] Michael Scharf leaves the room
[18:57:42] Michael Scharf joins the room
[18:57:49] Lennart Schulte joins the room
[18:58:37] David Borman leaves the room
[18:59:28] David Borman joins the room
[18:59:53] Martin Stiemerling joins the room
[18:59:53] Michael Scharf_ joins the room
[19:00:12] Michael Scharf_ leaves the room
[19:01:01] Alexander Zimmermann joins the room
[19:02:13] rscheff joins the room
[19:03:26] Lars joins the room
[19:07:56] Michael Tuexen joins the room
[19:10:33] Brian Trammell joins the room
[19:12:20] Brian Trammell leaves the room
[19:12:20] Brian Trammell joins the room
[19:12:37] gorryf@jabber.org joins the room
[19:13:08] bht joins the room
[19:17:45] Michael Tuexen leaves the room
[19:19:31] Michael Tuexen joins the room
[19:20:00] <PasiS> Now on TCP options space discussion, slide 3
[19:21:12] <PasiS> slide 4
[19:22:17] <PasiS> slide 5
[19:23:59] <PasiS> slide 6
[19:26:29] <PasiS> slide 7
[19:27:06] <PasiS> slide 8
[19:28:17] <PasiS> slide 9
[19:29:54] <PasiS> slide 11
[19:30:02] sowmini varadhan joins the room
[19:30:40] sowmini varadhan leaves the room
[19:30:45] <PasiS> slide 12
[19:33:20] Lennart Schulte leaves the room
[19:33:28] <PasiS> slide 13
[19:34:21] sowmini varadhan joins the room
[19:34:23] Lennart Schulte joins the room
[19:35:46] <Martin Stiemerling> Now TCP Extended Data Offset Option preso
[19:35:59] touch joins the room
[19:36:00] <Martin Stiemerling> slide 2
[19:36:06] touch leaves the room
[19:36:13] <Martin Stiemerling> slide 3
[19:36:35] <Martin Stiemerling> slide 4
[19:37:49] <Martin Stiemerling> slide 5
[19:39:09] <Martin Stiemerling> slide 6
[19:40:09] <Martin Stiemerling> slide 7
[19:42:51] <Martin Stiemerling> slide 2 of the inner space draft
[19:42:53] Lennart Schulte leaves the room
[19:43:21] Lennart Schulte joins the room
[19:43:51] <PasiS> slide 2
[19:44:12] <rscheff> anyone looking at the video stream? is bob visible?
[19:44:19] <bht> no
[19:44:29] <bht> we cannot see bob unless bob has been replaced with a chair
[19:44:41] <rscheff> how about now?
[19:44:45] <bht> (in that case, the chair does an excellent impression of bob)
[19:44:50] <bht> yep yep
[19:45:07] <bht> oh right you guys are in hawaii
[19:45:51] <PasiS> slide 3
[19:46:04] Mirja Kuehlewind joins the room
[19:46:42] sowmini varadhan leaves the room
[19:47:48] Daniele Iamartino joins the room
[19:48:15] Moe Cheaib joins the room
[19:50:08] <PasiS> slide 4
[19:50:48] <PasiS> slide 5
[19:51:15] Daniele Iamartino leaves the room
[19:52:25] <PasiS> slide 6
[19:52:32] Moe Cheaib leaves the room
[19:53:03] Moe Cheaib joins the room
[19:53:09] <PasiS> slide 7
[19:58:38] Moe Cheaib leaves the room
[19:59:43] <David Borman> mic: the dual syn also breaks simultaneous open
[20:00:14] <PasiS> will channel
[20:00:39] <David Borman> because there are two connections with 2 source ports.
[20:01:26] <PasiS> right
[20:01:43] <PasiS> somebody made that point in mic
[20:01:49] <PasiS> slide 10
[20:01:51] kivinen joins the room
[20:02:49] <Brian Trammell> ow ow ow
[20:03:56] <PasiS> slide 11
[20:04:24] bht leaves the room
[20:05:10] Gorry Fairhurst joins the room
[20:05:34] Gorry Fairhurst leaves the room
[20:06:21] Gorry  Fairhurst joins the room
[20:08:16] <David Borman> mic: TCP options that need to be processed by the kernel, e.g. SACK, can't be in the application
[20:08:37] <David Borman> it's the kernel that is usually interested in what is in the options
[20:10:53] sowmini varadhan joins the room
[20:11:54] <Brian Trammell> mic: very difficult question to answer without way more data
[20:12:41] <Brian Trammell> mic: dropping non-standard segments, splitting and coalescing are specifically not very well addressed in the literature
[20:12:57] <Brian Trammell> mic: basically what lars said
[20:13:23] <Brian Trammell> mic: and "try it" means tens of thousands to millions of paths
[20:13:36] <Brian Trammell> mic: and not just on planetlab
[20:14:15] Meetecho joins the room
[20:14:31] sowmini varadhan leaves the room
[20:14:41] <David Borman> mic: We need to be careful that we don't let the middle boxes of today dictate TCP of the future, because whatever we do, future middle boxes will be able to mess with anything we do.
[20:14:43] <Brian Trammell> mic: jana +1: there should be a (relatively standard) section in these documents stating their middlebox model
[20:15:16] <PasiS> after brandon
[20:16:11] sowmini varadhan joins the room
[20:16:24] <Brian Trammell> (preferably with a description of a suggested path transparency test: if you can run this raw packet bashing code on a path, this extension will work)
[20:18:38] <kivinen> David: For some options yes, but for example for the tcpinc use cases, there "tcp options" are not really needed by the kernel. I.e. if you can make wrapper for the socket calls (like socksify program on unix systems), that would take normal unmodified application hijack socket calls, do tcpinc processing in the application and generate the magic and framing in the application and send it out from old legacy kernel. In the other end it could be upgraded kernel doing the reverse processing, or it could be exactly same thing there too, i.e. wrapper for application stripping the magic, undoing the framing, processing the tcpinc "tcp options", decrypt and validate the tcp payload, and pass it to application.
[20:18:39] <Brian Trammell> there's another recent one in SIGCOMM
[20:18:58] gkietf joins the room
[20:19:20] sowmini varadhan leaves the room
[20:19:20] <Brian Trammell> (Craven et al, on HICCUPS, a TCP that measures path impairment as a side effect of its operation)
[20:19:44] <Michael Scharf> @kivinen: And this is why tcpinc is mostly a layer on top of TCP, unless header fields are protected
[20:19:59] Natasha Rooney joins the room
[20:20:56] <kivinen> michael: yes, and we will be discussing that more on Friday in tcpinc.
[20:22:25] <Michael Scharf> mic: I originally proposed a payload variant of MPTCP. It is somehow similar to the inner encoding.
[20:22:55] Lennart Schulte leaves the room
[20:24:23] Lennart Schulte joins the room
[20:28:49] <David Borman> mic: there is no way to include additional data in the SYN packet that a legacy server will properly ignore
[20:29:14] <David Borman> I mean, additional option space
[20:35:24] <John Leslie> +1 to Lars
[20:35:57] <Brian Trammell> mic: experimental implementations, instrumented to produce data on observed path characteristics, would be a way to do this.
[20:36:27] <Brian Trammell> (obvs with a switch to turn it off when you care about latency :) )
[20:37:12] <Brian Trammell> mic: if we have lots of data, we need all that data in one place in the same format.
[20:37:13] <David Borman> mic: It seems to me that the only way to have 100% middlebox compatibility is to do nothing, but that isn't reasonable.   Likewise, something that every middle box breaks is also unreasonable.  So how far do we bend to deal with middle boxes?
[20:38:54] gkietf leaves the room: Disconnected: Broken pipe
[20:39:00] <David Borman> Also, compatibility with middle boxes is about transition until the middleboxes catch up.
[20:39:12] <Brian Trammell> mic: +1 jana, simple solution is ignore middleboxes. the problem is necessarily complex.
[20:41:26] Yanir Weiss joins the room
[20:41:51] <Michael Scharf> For middleboxes, see e.g. RFC 3135. I am not sure if this document was ever read by middlebox vendors.
[20:43:53] <PasiS> sorry, fighting with A/V issues in room takes our attention too much
[20:44:15] <PasiS> wasn't able to follow jabber for a moment
[20:45:19] <Brian Trammell> np
[20:45:31] <Brian Trammell> i have a sense this discussion will continue on the list :)
[20:46:04] Brian Trammell leaves the room
[20:49:01] Lennart Schulte leaves the room
[20:49:14] Martin Stiemerling leaves the room
[20:49:36] Lennart Schulte joins the room
[20:49:47] Yanir Weiss leaves the room
[20:53:14] Martin Stiemerling joins the room
[20:55:28] Meetecho leaves the room
[20:57:20] Lennart Schulte leaves the room
[20:57:46] Meetecho joins the room
[20:58:05] Lennart Schulte joins the room
[21:00:57] Lars leaves the room: Disconnected: session closed
[21:01:19] <David Borman> Use two TCP options, like the original Echo and Echo Reply options.  Then it is unambiguous
[21:01:24] <David Borman> (for the mic)
[21:01:26] <PasiS> things seems to be back to normal, now echo cookie, slide 4
[21:01:47] <PasiS> david: ack
[21:02:26] Lars joins the room
[21:02:31] Lennart Schulte leaves the room
[21:02:46] Mirja Kuehlewind leaves the room
[21:03:07] Mirja Kuehlewind joins the room
[21:03:19] Lennart Schulte joins the room
[21:03:42] <rscheff> video working again? bob visible?
[21:04:14] <Michael Tuexen> I can see bob, can't see the slides...
[21:04:28] <rscheff> slide 5
[21:04:31] <PasiS> there are lots of problems with video overall at the moment
[21:04:54] <rscheff> slide 6
[21:07:31] Michael Tuexen leaves the room
[21:08:19] Michael Tuexen joins the room
[21:08:55] <PasiS> slide 3, RTO-VM
[21:09:54] shigeya joins the room
[21:10:11] <PasiS> slide 4
[21:10:57] shigeya leaves the room
[21:11:00] Martin Stiemerling leaves the room
[21:12:39] shigeya joins the room
[21:13:30] <Michael Scharf> Actually, these charts have a bias: The actual time until the timeout is triggered is one RTT larger for bulk data. (See rto-restart.)
[21:13:47] shigeya leaves the room
[21:14:45] Martin Stiemerling joins the room
[21:18:16] Lars leaves the room: Disconnected: connection closed
[21:19:37] Lennart Schulte leaves the room
[21:20:24] Lennart Schulte joins the room
[21:30:43] Natasha Rooney leaves the room
[21:36:27] Meetecho leaves the room
[21:37:31] Martin Stiemerling leaves the room
[21:38:21] Lennart Schulte leaves the room
[21:38:25] Natasha Rooney joins the room
[21:38:26] kivinen leaves the room
[21:39:21] gorryf@jabber.org leaves the room
[21:39:49] Michael Scharf leaves the room
[21:40:01] Michael Tuexen leaves the room
[21:40:06] Mirja Kuehlewind leaves the room
[21:40:08] Gorry  Fairhurst leaves the room
[21:40:19] David Borman leaves the room
[21:40:19] John Leslie leaves the room
[21:42:09] rscheff leaves the room
[21:44:29] Lars joins the room
[21:48:11] PasiS leaves the room
[21:48:28] PasiS joins the room
[21:49:55] Alexander Zimmermann leaves the room
[21:51:29] PasiS leaves the room
[22:00:29] Lars leaves the room: Disconnected: connection closed
[22:12:29] yoshifumi.nishida leaves the room
[22:34:15] Martin Stiemerling joins the room
[22:36:26] gkietf joins the room
[22:43:23] Natasha Rooney leaves the room
[22:48:19] Martin Stiemerling leaves the room
[22:53:40] rscheff joins the room
[22:56:49] Alexander Zimmermann joins the room
[22:57:07] Alexander Zimmermann leaves the room
[22:57:18] jlcJohn leaves the room
[22:57:32] gkietf leaves the room: Disconnected: closed
[22:58:36] rscheff leaves the room
[22:59:43] Lars joins the room
[22:59:47] Lars leaves the room
[23:00:07] Martin Stiemerling joins the room
[23:04:18] gkietf joins the room
[23:14:16] Martin Stiemerling leaves the room
[23:28:54] Alexander Zimmermann joins the room
[23:29:10] Alexander Zimmermann leaves the room
[23:55:29] gkietf leaves the room
[23:55:35] gkietf joins the room
[23:56:09] Natasha Rooney joins the room
Powered by ejabberd Powered by Erlang Valid XHTML 1.0 Transitional Valid CSS!