Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
On Sat, 2009-08-01 at 16:12 +, Debian Bug Tracking System wrote: if you have to disable ipv6 to make the internet connection fast, the setup at your provider is broken - Debian comes with working out of the box support for ipv6, if you need to disable it to make the network work for you, there is something wrong on the network. While I agree in general, it is still a PITA that Debian fails miserably in these cases. It does cause a bad impression. I keep having problems with misbehaving ipv6 myself, and it seems unlikely that stuff will just start supporting ipv6 correctly in the near future. While at debconf I was debugging why epiphany-webkit (libsoup, really) was failing to get to Rhonda's blog, and it seems like DNS was resolving the name to the ipv6 address as well as the ipv4. It seems like firefox handles this by falling back, but soup doesn't. Can we do better? See you, -- Gustavo Noronha Silva k...@debian.org Debian -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Re: Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
This one time, at band camp, Gustavo Noronha Silva said: While at debconf I was debugging why epiphany-webkit (libsoup, really) was failing to get to Rhonda's blog, and it seems like DNS was resolving the name to the ipv6 address as well as the ipv4. It seems like firefox handles this by falling back, but soup doesn't. Debconf had fully working ipv6. I do wish you could have reported problems with it at the time so we could have looked at it. Cheers, -- - | ,''`.Stephen Gran | | : :' :sg...@debian.org | | `. `'Debian user, admin, and developer | |`- http://www.debian.org | - signature.asc Description: Digital signature
Re: Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
Tollef Fog Heen tfh...@err.no writes: ]] Roger Leigh | Having working local networking is important. We wouldn't consider | broken IPv4 loopback acceptable, and broken IPv6 loopback is just as | bad. Sure, having it working is important. Is it more important than keeping those (often new) users for whom Debian appears useless because of its perceived poor network performance? I don't think measuring the importance of these bugs against each other is going to be productive. Workarounds for bugs in other systems are fine, but not adding bugs in Debian should be an absolute requirement for such workarounds. Bjørn -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
]] Roger Leigh | Having working local networking is important. We wouldn't consider | broken IPv4 loopback acceptable, and broken IPv6 loopback is just as | bad. Sure, having it working is important. Is it more important than keeping those (often new) users for whom Debian appears useless because of its perceived poor network performance? Anyway, IIRC this is now solved in glibc by sending out the queries in parallel and returning the first answer you get. | The idea behind the patch isn't bad, but the implementation proposed | here is too naïve. The assumption that you only want working IPv6 | name resolution when you have a globally-scoped IPv6 address is too | simplistic. FWIW, it roughly matches what Mac OS X and Windows do. | Not only do you have the local loopback, you also have link-local | addresses which you can legitimately use. Does zeroconf support | these? Fundamentally breaking IPv6 for these use cases to work around | broken routing hardware is IMO a step too far. Does anybody use IPv6-only link-local? -- Tollef Fog Heen UNIX is user friendly, it's just picky about who its friends are -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Re: Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
On 2009-08-02, Tollef Fog Heen tfh...@err.no wrote: | Not only do you have the local loopback, you also have link-local | addresses which you can legitimately use. Does zeroconf support | these? Fundamentally breaking IPv6 for these use cases to work around | broken routing hardware is IMO a step too far. Does anybody use IPv6-only link-local? I occassionally did and I think Gobby even supports that now in conjunction with Zeroconf. Kind regards, Philipp Kern -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
On Sun, Aug 02, 2009 at 08:39:46AM +0200, Tollef Fog Heen wrote: ]] Roger Leigh | Having working local networking is important. We wouldn't consider | broken IPv4 loopback acceptable, and broken IPv6 loopback is just as | bad. Sure, having it working is important. Is it more important than keeping those (often new) users for whom Debian appears useless because of its perceived poor network performance? It would be nice if both worked. But, in the absence of good heuristics for detecting broken networking, this is probably something that should not be done automatically. Anyway, IIRC this is now solved in glibc by sending out the queries in parallel and returning the first answer you get. OK. What happens to the other answer(s) though? Does this mean that getaddrinfo(3) skips slow replies completely? This is surely equally broken if so? For anyone with broken networking, the test program here: http://bugs.debian.org/cgi-bin/bugreport.cgi?msg=5;filename=addrtest.c;att=1;bug=441857 will test local name resolving in glibc; just replace ip6-localhost with a hostname to check. | The idea behind the patch isn't bad, but the implementation proposed | here is too naïve. The assumption that you only want working IPv6 | name resolution when you have a globally-scoped IPv6 address is too | simplistic. FWIW, it roughly matches what Mac OS X and Windows do. And it's not considered broken on these platforms too? I guess the proportion of IPv6 users is vastly lower on these platforms due to the general networking clue level of the users, and that Windows has had fully functional IPv6 networking for a much shorter time than GNU/Linux. | Not only do you have the local loopback, you also have link-local | addresses which you can legitimately use. Does zeroconf support | these? Fundamentally breaking IPv6 for these use cases to work around | broken routing hardware is IMO a step too far. Does anybody use IPv6-only link-local? Probably not conciously; it's intended to be used automatically. A quick google shows Avahi does use these (with some caveats). IPv6 has the concept of Scope, and these scopes include Node/Host, Link, Site, Organisation, and Global. To assume that IPv6 networking is only active when you have one or more Global addresses is an incorrect assumption. With the previous glibc patch, my networking broke as the Global link went up and down. Regards, Roger -- .''`. Roger Leigh : :' : Debian GNU/Linux http://people.debian.org/~rleigh/ `. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/ `-GPG Public Key: 0x25BFB848 Please GPG sign your mail. signature.asc Description: Digital signature
Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
]] Roger Leigh | On Sun, Aug 02, 2009 at 08:39:46AM +0200, Tollef Fog Heen wrote: | ]] Roger Leigh | | | Having working local networking is important. We wouldn't consider | | broken IPv4 loopback acceptable, and broken IPv6 loopback is just as | | bad. | | Sure, having it working is important. Is it more important than keeping | those (often new) users for whom Debian appears useless because of its | perceived poor network performance? | | It would be nice if both worked. But, in the absence of good heuristics | for detecting broken networking, this is probably something that should | not be done automatically. Both would be nice, of course. Sometimes you have to choose between having the cake and eating it, though. | Anyway, IIRC this is now solved in glibc by sending out the queries in | parallel and returning the first answer you get. | | OK. What happens to the other answer(s) though? Does this mean that | getaddrinfo(3) skips slow replies completely? This is surely equally | broken if so? Unsure. The man page for getaddrinfo doesn't seem to imply that you get all answers if you specify AF_UNSPEC, it specifies «any». Your test program does get both the A and the record, so my recollection of this might be incorrect. [...] | | The idea behind the patch isn't bad, but the implementation proposed | | here is too naïve. The assumption that you only want working IPv6 | | name resolution when you have a globally-scoped IPv6 address is too | | simplistic. | | FWIW, it roughly matches what Mac OS X and Windows do. | | And it's not considered broken on these platforms too? You're talking about 0.07% of the users for Windows Vista (and 0.03% for XP), Mac OS X is at a whopping 2.4%. I doubt many users would see if their IPv6 was broken since there are few/no IPv6 services out there. (From http://www.ripe.net/ripe/meetings/ripe-57/presentations/Colitti-Global_IPv6_statistics_-_Measuring_the_current_state_of_IPv6_for_ordinary_users_.7gzD.pdf ) | Probably not conciously; it's intended to be used automatically. A | quick google shows Avahi does use these (with some caveats). | | IPv6 has the concept of Scope, and these scopes include Node/Host, | Link, Site, Organisation, and Global. To assume that IPv6 networking | is only active when you have one or more Global addresses is an | incorrect assumption. With the previous glibc patch, my networking | broke as the Global link went up and down. It did actually put the threshold at site, not global, since people could reasonably put site-only names into DNS, but link-local addresses shouldn't be in DNS. -- Tollef Fog Heen UNIX is user friendly, it's just picky about who its friends are -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
On Sun, Aug 02, 2009 at 03:53:48PM +0200, Tollef Fog Heen wrote: ]] Roger Leigh | On Sun, Aug 02, 2009 at 08:39:46AM +0200, Tollef Fog Heen wrote: | Anyway, IIRC this is now solved in glibc by sending out the queries in | parallel and returning the first answer you get. | | OK. What happens to the other answer(s) though? Does this mean that | getaddrinfo(3) skips slow replies completely? This is surely equally | broken if so? Unsure. The man page for getaddrinfo doesn't seem to imply that you get all answers if you specify AF_UNSPEC, it specifies «any». Your test program does get both the A and the record, so my recollection of this might be incorrect. Just tested on current sid for localhost where this is listed in /etc/hosts as both 127.0.0.1 and ::1: % addrtest localhost 3 Addrinfo for 0x21f8f90 Flags: 32 Family: 10 Socket Type:1 Protocol: 6 (tcp) Canonical name: (null) Socket Address (len=28): Port: 3 IPv6 Address: ::1 Addrinfo for 0x21f8ff0 Flags: 32 Family: 10 Socket Type:2 Protocol: 17 (udp) Canonical name: (null) Socket Address (len=28): Port: 3 IPv6 Address: ::1 Addrinfo for 0x21f9050 Flags: 32 Family: 10 Socket Type:3 Protocol: 0 (ip) Canonical name: (null) Socket Address (len=28): Port: 3 IPv6 Address: ::1 Addrinfo for 0x21f81a0 Flags: 32 Family: 2 Socket Type:1 Protocol: 6 (tcp) Canonical name: (null) Socket Address (len=16): Port: 3 IPv4 Address: 127.0.0.1 Addrinfo for 0x21f81f0 Flags: 32 Family: 2 Socket Type:2 Protocol: 17 (udp) Canonical name: (null) Socket Address (len=16): Port: 3 IPv4 Address: 127.0.0.1 ravenclaw% ./at localhost 3 5 Usage: ./at hostname port ravenclaw% ./at localhost Usage: ./at hostname port ravenclaw% ./at localhost 44 Addrinfo for 0x8bdf90 Flags: 32 Family: 10 Socket Type:1 Protocol: 6 (tcp) Canonical name: (null) Socket Address (len=28): Port: 44 IPv6 Address: ::1 Addrinfo for 0x8bdff0 Flags: 32 Family: 10 Socket Type:2 Protocol: 17 (udp) Canonical name: (null) Socket Address (len=28): Port: 44 IPv6 Address: ::1 Addrinfo for 0x8be050 Flags: 32 Family: 10 Socket Type:3 Protocol: 0 (ip) Canonical name: (null) Socket Address (len=28): Port: 44 IPv6 Address: ::1 Addrinfo for 0x8bd1a0 Flags: 32 Family: 2 Socket Type:1 Protocol: 6 (tcp) Canonical name: (null) Socket Address (len=16): Port: 44 IPv4 Address: 127.0.0.1 Addrinfo for 0x8bd1f0 Flags: 32 Family: 2 Socket Type:2 Protocol: 17 (udp) Canonical name: (null) Socket Address (len=16): Port: 44 IPv4 Address: 127.0.0.1 Testing again for via proper DNS for noc.sixxs.net: % addrtest noc.sixxs.net 44 Addrinfo for 0x142f190 Flags: 32 Family: 2 Socket Type:1 Protocol: 6 (tcp) Canonical name: (null) Socket Address (len=16): Port: 44 IPv4 Address: 213.197.29.32 Addrinfo for 0x142f1e0 Flags: 32 Family: 2 Socket Type:2 Protocol: 17 (udp) Canonical name: (null) Socket Address (len=16): Port: 44 IPv4 Address: 213.197.29.32 Addrinfo for 0x142f230 Flags: 32 Family: 2 Socket Type:3 Protocol: 0 (ip) Canonical name: (null) Socket Address (len=16): Port: 44 IPv4 Address: 213.197.29.32 Addrinfo for 0x142f280 Flags: 32 Family: 10 Socket Type:1 Protocol: 6 (tcp) Canonical name: (null) Socket Address (len=28): Port: 44 IPv6 Address: 2001:838:1:1:210:dcff:fe20:7c7c Addrinfo for 0x142f2e0 Flags: 32 Family: 10 Socket Type:2 Protocol: 17 (udp) Canonical name: (null) Socket Address (len=28): Port: 44 IPv6 Address: 2001:838:1:1:210:dcff:fe20:7c7c I've attached a slightly updated version of the test code. I'd be interested on the results for people with broken nameservers. The above tests are on a system without a Scope:Global IPv6 address but with working DNS lookups. Regards, Roger -- .''`. Roger Leigh : :' : Debian GNU/Linux http://people.debian.org/~rleigh/ `. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/ `-GPG Public Key: 0x25BFB848 Please GPG sign your mail. /* Copyright © 2007 Roger Leigh rle...@debian.org * * addrtest is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2
Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
Your message dated Sat, 1 Aug 2009 18:09:54 +0200 with message-id 200908011809.54427.hol...@layer-acht.org and subject line your provider provides buggy internet has caused the Debian Bug report #535833, regarding general: Slow internet on iceweasel, epiphany and so on... to be marked as done. This means that you claim that the problem has been dealt with. If this is not the case it is now your responsibility to reopen the Bug report if necessary, and/or fix the problem forthwith. (NB: If you are a system administrator and have no idea what this message is talking about, this may indicate a serious mail system misconfiguration somewhere. Please contact ow...@bugs.debian.org immediately.) -- 535833: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=535833 Debian Bug Tracking System Contact ow...@bugs.debian.org with problems ---BeginMessage--- Package: general Severity: important The internet connection is very slow when using the internet navigator on Debian 5.0. This problem doesn't appear on another computer of my network under ubuntu whereas the DNS used is the same for all computers. I fixed this problem: after typing 'about:config' in the adress bar of iceweasel, I modified the key 'network.dns.disableIPv6' to 'true'. This operation works for iceweasel and for epiphany too. I don't know if other applications are affected by this problem (synaptic may be affected). -- System Information: Debian Release: 5.0.2 APT prefers stable APT policy: (500, 'stable') Architecture: i386 (i686) Kernel: Linux 2.6.26-2-686 (SMP w/1 CPU core) Locale: LANG=fr_FR.UTF-8, LC_CTYPE=fr_FR.UTF-8 (charmap=UTF-8) Shell: /bin/sh linked to /bin/bash ---End Message--- ---BeginMessage--- Hi, if you have to disable ipv6 to make the internet connection fast, the setup at your provider is broken - Debian comes with working out of the box support for ipv6, if you need to disable it to make the network work for you, there is something wrong on the network. regards, Holger signature.asc Description: This is a digitally signed message part. ---End Message---
Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
]] (Debian Bug Tracking System) | if you have to disable ipv6 to make the internet connection fast, | the setup at your provider is broken - Debian comes with working out | of the box support for ipv6, if you need to disable it to make the | network work for you, there is something wrong on the network. This is probably the same bug as http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=435646, which while a bug in the DSL router of users is not possible for most of them to work around, while fairly trivial for us to work around in libc. (Heck, I believe newer libcs already have the fix in.) (I think we should fix it, but I'm not going to fight that battle again, since apparently having loopback ipv6 when no other IPv6 address is configured working is more important than making Debian usable for certain people without having to disable ipv6. See 441857 for the other bug in the story.) -- Tollef Fog Heen UNIX is user friendly, it's just picky about who its friends are -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#535833: marked as done (general: Slow internet on iceweasel, epiphany and so on...)
On Sat, Aug 01, 2009 at 09:51:26PM +0200, Tollef Fog Heen wrote: ]] (Debian Bug Tracking System) | if you have to disable ipv6 to make the internet connection fast, | the setup at your provider is broken - Debian comes with working out | of the box support for ipv6, if you need to disable it to make the | network work for you, there is something wrong on the network. This is probably the same bug as http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=435646, which while a bug in the DSL router of users is not possible for most of them to work around, while fairly trivial for us to work around in libc. (Heck, I believe newer libcs already have the fix in.) (I think we should fix it, but I'm not going to fight that battle again, since apparently having loopback ipv6 when no other IPv6 address is configured working is more important than making Debian usable for certain people without having to disable ipv6. See 441857 for the other bug in the story.) Having working local networking is important. We wouldn't consider broken IPv4 loopback acceptable, and broken IPv6 loopback is just as bad. The idea behind the patch isn't bad, but the implementation proposed here is too naïve. The assumption that you only want working IPv6 name resolution when you have a globally-scoped IPv6 address is too simplistic. Not only do you have the local loopback, you also have link-local addresses which you can legitimately use. Does zeroconf support these? Fundamentally breaking IPv6 for these use cases to work around broken routing hardware is IMO a step too far. If there's a better metric for detecting that IPv6 name resolution is broken, and then disabling it, I wouldn't be opposed to it as long as it doesn't break things which are currently working. Regards, Roger -- .''`. Roger Leigh : :' : Debian GNU/Linux http://people.debian.org/~rleigh/ `. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/ `-GPG Public Key: 0x25BFB848 Please GPG sign your mail. signature.asc Description: Digital signature