Monday, August 5, 2013< ^ >
gmaxwell has set the subject to: HTTPbis Working Group |
Room Configuration
Room Occupants

[01:43:23] lear leaves the room: Replaced by new connection
[01:43:23] lear joins the room
[06:53:56] Eliot Lear joins the room
[07:07:01] Julian joins the room
[07:07:46] cabo joins the room
[07:07:49] hiro1017 joins the room
[07:08:00] Barry Leiba joins the room
[07:08:01] martin.thomson joins the room
[07:08:33] alexey.melnikov joins the room
[07:09:00] <alexey.melnikov> I will be trying to scribe during the interim
[07:09:12] <alexey.melnikov> Mark showed a pretty picture of Berlin and Note Well
[07:09:48] <alexey.melnikov> Mark N: is asking people to talk about their implementation experience
[07:10:43] <alexey.melnikov> Not clear if we can usefully do interop testing in Hamburg
[07:10:50] <Barry Leiba> Hi, Eliot!
[07:11:30] <alexey.melnikov> Several people have QA people testing HTTP2, but they don't seem to participate in HTTPBis discussions.
[07:11:44] <alexey.melnikov> Several companies rather
[07:12:41] <alexey.melnikov> Mark is talking about Berlin IETF summary
[07:14:00] <alexey.melnikov> In the next month or so outstanding issues on HTTP/1.1 should be closed. 4 weeks IETF LC, then 4 weeks till IESG telechat after
[07:14:33] <alexey.melnikov> A new issue about use of TLS in HTTP/1.1 and HTTP/2.0
[07:15:19] <alexey.melnikov> Server have control about whether a connection is encrypted or not. Clients don't have much control. This need to change
[07:16:11] <alexey.melnikov> A joined session with Transport Area people. Alison presented HTTP/2.0 issues from Transport prospective.
[07:16:35] joins the room
[07:16:54] <alexey.melnikov> It was a lively discussion. Transport people want to help HTTPBis.
[07:16:59] willchan joins the room
[07:19:13] <alexey.melnikov> Mark is showing wiki with the list of known implementations
[07:20:08] <alexey.melnikov> Mark is asking about compression and draft quality in general
[07:21:06] <alexey.melnikov> Patrick: 2 flow controls - not clear about the difference
[07:21:19] <alexey.melnikov> Continuations are broken
[07:21:54] <alexey.melnikov> Compression is not hard to implement. Implemented in 2 days (both sides), 400 lines of C++.
[07:23:03] <alexey.melnikov> No obvious bugs in the compression. Variable length encoding was better understood after the code was written.
[07:24:51] <alexey.melnikov> ALPN versa NPN - the ordering of protocol is opposite.
[07:25:52] <alexey.melnikov> Hitting some limit which causes bugs. ALPN can just cause timeouts
[07:26:30] <alexey.melnikov> "Hello world" is running between Google, Microsoft and Mozilla implementations.
[07:27:10] <alexey.melnikov> 2 laptops sacrificed to HTTP2 development so far :-)
[07:27:26] <alexey.melnikov> Mark N: need a separate mailing list for developers?
[07:27:54] <alexey.melnikov> People not convinced. Just use existing bug tracking (e.g. in GIT)
[07:28:38] <alexey.melnikov> Patrick is generally happy
[07:30:25] <alexey.melnikov> Fred from Google: no surprises. It would have been nice to test the base spec without compression.
[07:31:27] <alexey.melnikov> Doing SPDY 2, SPDY 3 and HTTP/2.0 in the same code base is tricky
[07:35:00] <alexey.melnikov> Jeff (Twitter): a continuation frame modifies information in previous frames.
[07:36:31] <alexey.melnikov> Hassan: I haven't implemented continuations yet. Do we even need them? There is one use cases for them.
[07:37:05] <alexey.melnikov> Continuations only apply to headers. Should we just have a new frame type ("more headers") instead of the continuation bit.
[07:38:36] <alexey.melnikov> Roberto: restrict use of continuation to HEADERS
[07:40:04] <alexey.melnikov> It would be good for a new op code that omits duplicate information (e.g. priority)
[07:42:53] <martin.thomson> eliot: that was hasan who promised a write up
[07:44:24] lef joins the room
[07:47:22] <alexey.melnikov> The latest HTTP/2.0 requires libraries to know whether they are client side or server side in more places.
[07:49:21] Hervé joins the room
[07:53:07] <alexey.melnikov> State transition table ( - people found it to be useful
[07:56:18] <alexey.melnikov> Akamai: implementation in Perl from scratch. It was very easy. Compression was quite complicated (lots of small details).
[07:57:26] <alexey.melnikov> Window update: need to always send two: one per connection and one per stream
[07:58:04] <alexey.melnikov> Which seemed to be a bit verbose
[07:58:33] <alexey.melnikov> Some disagreement on whether always sending 2 window updates is needed
[08:00:45] <alexey.melnikov> Jeff: (Compression) Inserting a single big header (over 4Kb) will cause connection close (4Kb buffer for all headers). But writing a stupid implementation is too easy.
[08:02:36] <alexey.melnikov> Akamai implementation is a proxy
[08:03:13] <alexey.melnikov> Roberto: did anybody not implement flow control?
[08:04:00] <alexey.melnikov> Patrick: disabled flow control on some stream and not others. An interesting case to test.
[08:04:50] <alexey.melnikov> Markn N: who did NPN? 7 hangs
[08:04:57] <alexey.melnikov> ALPN: 5 hands
[08:05:07] <alexey.melnikov> Upgrade: 3 hands
[08:06:53] <alexey.melnikov> Akamai person is Steven
[08:07:28] <alexey.melnikov> Mark is editing implementation wiki
[08:09:04] <alexey.melnikov> Barry: is there a reference to the implementation wiki in the HTTP/2.0 draft?
[08:10:12] <Barry Leiba> "Implementation Status" is, I believe, the title for the section.  See RFC 6982
[08:14:44] <Barry Leiba> I think for this, I suggest putting in a note that −04 was the first draft version for which implementations were encouraged, and that [this URL] contains the list of implementations and details of those implementations.  That's all you need: small and simple.
[08:15:52] <alexey.melnikov> Microsoft OpenTech contributed ALPN patch to OpenSSL, but it was not integrated yet. OpenSSL people are quite conservative/slow.
[08:16:36] <Barry Leiba> BTW, response to "circular references suck": Circular normative references suck.  But the reference from the http2 draft to the implementation page is not normative.  'snot a problem.
[08:20:26] <alexey.melnikov> Roberto: different NPN/ALPN strings "I really want to use this" versa "this is for testing"
[08:23:46] martin.thomson leaves the room
[08:24:57] <alexey.melnikov> Discussing finer points about large scale testing and deploying experimental code
[08:25:17] <Eliot Lear> can't servers simply query their base at that hour to opt in?
[08:30:33] <alexey.melnikov> Hasan: a variety of different applications/uses is better than just increase in the number of test users (without introducing new uses).
[08:36:41] <alexey.melnikov> Microsoft has no "experimental" channel. Probably will delay supporting this in IE by default until the spec is more stable. It is hard to deprecate features, due to customer demand.
[08:40:31] Hervé leaves the room
[08:52:25] alexey.melnikov leaves the room
[08:55:57] <Eliot Lear> offline til 11:20
[09:09:38] Hervé joins the room
[09:18:31] alexey.melnikov joins the room
[09:20:43] <alexey.melnikov> Discussion about how to identify particular implementations. Are we adding user-agent strings where HTTP/2.0 version is used (in ALPN)?
[09:21:33] <alexey.melnikov> Roberto: Add a new GOAWAY with "don't use this string again when talking to me" or "fallback to HTTP/1.1"?
[09:22:59] <alexey.melnikov> Hasan: But GOAWAY might not fix flow control bugs. It would be nice to know client string at the beginning of a HTTP/2 session.
[09:25:20] <alexey.melnikov> Mark N.: closing the discussion about implementation experience.
[09:25:40] Julian leaves the room
[09:29:13] <alexey.melnikov> Martin: I sent some comments to ALPN authors about strings which are different from what HTTP/2.0 draft says.
[09:31:41] <alexey.melnikov> Gabriel is talking about another TLS extension for conveying settings. At the moment there is no agreement among HTTPBis participants about whether this is the way to go.
[09:33:59] <alexey.melnikov> Mike: should we ask TLS WG to reevaluate the ALPN versa NPN decision (i.e. to go back to NPN), because of new privacy awareness. I.e. NPN encrypts the choice of protocol names.
[09:35:05] <alexey.melnikov> Patrick: ALPN might be tricky to deploy (the bug he mentioned earlier), prefer NPN.
[09:40:21] <alexey.melnikov> Mark. N. is trying to shut down the discussion about NPN versa ALPN because HTTPBis is not the right forum.
[09:40:44] <alexey.melnikov> And the TLS WG mailing list is.
[09:47:41] <alexey.melnikov> Marting: the list of TLS 1.3 requirements talk about encrypted negotiation and fewer round trips. Which seems what people want.
[09:48:41] <Eliot Lear> who is that speaking?
[09:49:09] <alexey.melnikov> I forgot the name, Eliot
[09:50:57] joins the room
[09:51:11] <alexey.melnikov> Mark: "we love the TLS WG :-)"
[09:52:51] <Eliot Lear> is there an agreed time for the wg to break for lunch?
[09:53:33] <alexey.melnikov> Between noon and 1pm...
[09:54:10] molnarg joins the room
[09:55:10] <Eliot Lear> i can hear you
[09:55:18] <Eliot Lear> the problem is that adobe won't let me unmute
[09:55:21] molnarg leaves the room
[09:55:26] csirkus joins the room
[09:55:30] <Eliot Lear> yes
[09:55:52] <alexey.melnikov> We can hear you
[09:57:35] csirkus leaves the room
[09:57:36] molnarg joins the room
[09:59:27] <alexey.melnikov> Issue # 1 (Upgrade mechanism)
[10:00:08] <alexey.melnikov> Issues are in GIT
[10:01:03] <alexey.melnikov> "Transport mapping" filter
[10:01:10] <alexey.melnikov> Issue 172
[10:01:26] <alexey.melnikov> "Set a baseline for TLS implementations".
[10:01:50] <alexey.melnikov> Martin: we would like to require SNI TLS extension
[10:02:23] <molnarg> Hello everyone, can I join in on skype as listener? (I could not configure my mic properly, but probably will not have much to say anyway)
[10:02:56] <alexey.melnikov> Are specifying a profile of TLS?
[10:03:08] <alexey.melnikov> Minimal version of TLS to support?
[10:03:59] <alexey.melnikov> Mark N: "MUST send SNI" - any objections?
[10:05:31] <alexey.melnikov> molnarg: I just asked Mark
[10:05:43] martin.thomson joins the room
[10:05:55] <alexey.melnikov> molnarg: What is your country code?
[10:05:59] <molnarg> HU
[10:06:10] <molnarg> skype name: molnargabor714
[10:06:39] <alexey.melnikov> molnarg: I think you will be asked to dial a number
[10:07:00] <alexey.melnikov> No RC4
[10:07:29] <Barry Leiba> molnarg: it's not Skype; it's a telephone conference
[10:08:25] <molnarg> okay I see... then I probably will not be able to do it (home telephone, and its not a free number I guess)
[10:08:38] <molnarg> no problem
[10:09:15] <Barry Leiba> Alexey was asking for your country code to see if there's a free number in your country.
[10:09:32] <alexey.melnikov> Dial in info:
[10:10:50] <molnarg> thanks
[10:10:52] <Barry Leiba> molnarg: Assuming that when you said "HU" you meant Hungary, +36, then there is a Budapest number you can use.
[10:11:35] <molnarg> Yes, I'll try in a minute, thanks
[10:13:49] <alexey.melnikov> TLS 1.2 is required, but this might evolve over time
[10:14:10] <alexey.melnikov> Twitter: we use 1.0 and 1.1/1.2 are current disabled.
[10:20:31] <molnarg> "Incorrect conference code or called a wrong number. code: 8603022084" Is this the right code?
[10:21:47] <Barry Leiba> Hm, I seem to remember that we had this problem here as well.  Checking.
[10:22:30] Julian joins the room
[10:22:47] <Eliot Lear> i used conference code 9091928374
[10:23:54] <molnarg> trying again
[10:26:57] <molnarg> Very noisy but listening now, thanks :)
[10:27:16] <Barry Leiba> Glad you could get in
[10:28:27] <Eliot Lear> stepping away for lunch
[10:28:37] <Barry Leiba> Bon appetit.
[10:28:41] <Eliot Lear> u2
[10:30:27] alexey.melnikov leaves the room
[10:30:49] <Hervé> Lunch break up to 1:20
[10:30:56] <Barry Leiba> Gabor, you were just in time for the lunch break!
[10:31:03] <molnarg> :D yes
[10:31:19] Hervé leaves the room
[10:32:01] martin.thomson leaves the room
[10:32:07] lef leaves the room
[10:32:27] lef joins the room
[10:47:38] lef leaves the room
[11:07:27] martin.thomson joins the room
[11:22:23] Hervé joins the room
[11:22:41] lef joins the room
[11:25:29] <Julian> ping
[11:25:32] <cabo> Meeting is resuming
[11:25:35] <Barry Leiba> Any remote people here?
[11:26:23] <Julian> we're ready to get back to work :-)
[11:27:44] <molnarg> pong!
[11:27:57] <Barry Leiba> Gabor, are you getting back on the phone?
[11:28:19] <molnarg> yeah, I would try
[11:29:34] <molnarg> but sometimes it's quite hard to understand what you guys say (noisy) so probably muted, as an observer
[11:29:47] alexey.melnikov joins the room
[11:30:56] <molnarg> is the code the same this time?
[11:31:09] <alexey.melnikov> Issue # 133
[11:31:12] <Barry Leiba> Yes.
[11:31:19] <alexey.melnikov> "GOAWAY status code or similar meaning "don't use HTTP/2 when reconnecting" would be good"
[11:31:21] <Barry Leiba> If you're on and we're not, let us know.
[11:32:01] <alexey.melnikov> Patrick: is this flag persistent?
[11:32:07] <alexey.melnikov> Google people: no
[11:33:43] <alexey.melnikov> Patrick: this is a cookie that identifies a user?
[11:33:52] <alexey.melnikov> Others seem to disagree.
[11:35:02] <alexey.melnikov> Patrick: it is a bit persistent between connections. Privacy team in Mozilla will object.
[11:37:26] <alexey.melnikov> Jeff: most of my clients would not respect the feature. For clients I control, I can do this in other ways. Not convinced.
[11:38:15] <alexey.melnikov> Steven: not high enough on a priority list.
[11:39:23] <alexey.melnikov> Jeff: my expectations is that we are dealing with evil clients that try to consume resources.
[11:43:46] <alexey.melnikov> Roberto: this is useful to deal with permanent problems in clients (or clients + server combination)
[11:44:29] <alexey.melnikov> Is this for debugging purposes or long term purposes? Hasan: both
[11:45:16] <Eliot Lear> where are we?  issue#?
[11:45:33] <alexey.melnikov> 133
[11:45:40] <Eliot Lear> thanks
[11:45:49] <alexey.melnikov> "GOAWAY status code or similar meaning "don't use HTTP/2 when reconnecting" would be good"
[11:52:42] <alexey.melnikov> Mark N.: "can clients can just ignore this code?" Others: yes, it is unenforceable.
[11:55:28] <alexey.melnikov> Mike: not much objection, but don't see much point
[12:03:03] <alexey.melnikov> Steven: the number of implementation that will be intelligent about this feature is going to be small, in comparison to naive implementations who would just do user-agent blocking
[12:04:27] <alexey.melnikov> Eliot: need to document this feature better if it is to be included
[12:05:42] <alexey.melnikov> Mark. N. going around the room asking who and how is going to use this
[12:11:40] <alexey.melnikov> Close the issue. Google people can play with this feature in SPDY 4. Then this can be discussed again.
[12:16:11] <alexey.melnikov> 215
[12:26:56] <alexey.melnikov> Side discussion about mapping to TCP minions
[12:31:18] <alexey.melnikov> Hasan: compression is THE biggest performance win. Even for mobile devices
[12:34:17] <alexey.melnikov> Mark N.: reasonable happy with compression, but need to play more with it.
[12:37:09] <alexey.melnikov> Most people who implemented compression didn't implement eviction.
[12:39:16] <alexey.melnikov> People are asking for a test suite
[12:46:41] <alexey.melnikov> Action item: Herve to start test corpus. Roberto will help.
[12:52:14] <alexey.melnikov> Issue 23
[12:52:59] <alexey.melnikov> Roberto: have a different compression context for :<header-field>s and the rest? This will help proxies to inspect/extract specific header fields
[12:59:55] <molnarg> An idea: if we would require headers to appear in alphabetical order in header blocks (and nothing else would change compared to current spec), then 1) ':' headers would always come first 2) it would be possible to implement a streaming decoder 2) it is possible to implement a streaming encoder, if its input is in alphabetical order (otherwise first you would have to reorder them).
[13:03:47] <Eliot Lear> i heard that just fine
[13:04:21] g.e.montenegro joins the room
[13:07:13] <molnarg> okay, agree that hard to enforce
[13:10:14] <alexey.melnikov> Issue # 187
[13:10:22] <alexey.melnikov> "String Literal"
[13:16:10] Hervé leaves the room
[13:16:16] <Eliot Lear> back in 30
[13:33:18] Hervé joins the room
[13:40:34] <Barry Leiba> Gabor, you still hearing us?
[13:41:10] <molnarg> No, sorry, I got disconnected
[13:41:21] <molnarg> But try to come back
[13:41:40] <Barry Leiba> Discussing issue 11 in the compression spec.
[13:41:45] <Barry Leiba> Contemplating closing the issue.
[13:41:55] <Barry Leiba> (It's yours...)
[13:42:16] <molnarg> yes
[13:42:21] <molnarg> I'm here
[13:42:41] <Barry Leiba> Issue was just closed.  If you want to say something, please do.
[13:42:52] <molnarg> yeah I hear you now
[13:43:17] <molnarg> sorry, I will have to come up with concrete text
[13:43:26] <molnarg> maybe revisit then
[13:43:28] <Barry Leiba> OK.
[13:58:09] Julian leaves the room
[13:58:44] Julian joins the room
[14:01:13] Julian leaves the room
[14:01:15] Julian joins the room
[14:04:39] alexey.melnikov leaves the room
[14:06:47] alexey.melnikov joins the room
[14:07:07] Julian leaves the room
[14:07:09] Julian joins the room
[14:12:52] <molnarg> Have to go now, thanks for letting me join!
[14:14:14] molnarg leaves the room
[14:15:15] Julian leaves the room
[14:19:42] <alexey.melnikov> Talking about disabling stream control
[14:43:24] willchan leaves the room
[14:43:30] Barry Leiba leaves the room
[14:44:04] lef leaves the room
[14:44:55] Hervé leaves the room
[14:45:31] leaves the room
[14:45:40] Hervé joins the room
[14:45:42] hiro1017 leaves the room
[14:45:45] Hervé leaves the room
[14:45:58] martin.thomson leaves the room
[14:46:00] Hervé joins the room
[14:46:02] Hervé leaves the room
[14:50:08] cabo leaves the room
[15:04:39] Eliot Lear leaves the room
[15:06:32] g.e.montenegro leaves the room
[15:43:50] lef joins the room
[15:55:10] joins the room
[15:56:35] hiro1017 joins the room
[16:21:19] leaves the room
[16:23:07] g.e.montenegro joins the room
[16:23:41] g.e.montenegro leaves the room
[17:10:09] lef leaves the room
[17:10:41] hiro1017 leaves the room
[18:55:37] cabo joins the room
[18:56:42] cabo leaves the room
[19:57:00] cabo joins the room
[20:12:46] cabo leaves the room
[21:00:09] willchan joins the room
[21:04:24] cabo joins the room
[21:12:23] cabo leaves the room
[21:32:31] martin.thomson joins the room
[21:45:39] martin.thomson leaves the room
[21:54:23] willchan leaves the room
[22:04:57] cabo joins the room
[22:12:57] cabo leaves the room
[22:31:59] lef joins the room
[22:45:51] hiro1017 joins the room
[22:46:37] hiro1017 leaves the room
[23:05:28] cabo joins the room
[23:15:13] cabo leaves the room
Powered by ejabberd Powered by Erlang Valid XHTML 1.0 Transitional Valid CSS!