This is only tangentially related but it looks like HE has surpassed Cogent on IPv4 adjacencies. That said the source probably suffers from a selection bias at the very least.
Hit reply by mistake instead of reply all.
Todd Crane
On Mar 14, 2016, at 8:40 PM, Matthew D. Hardeman <mhardeman@ipifony.com> wrote:
It looks like Google is experimenting with a change in course on this issue.
Here’s a look at the IPv6 routing table tonight on my router bordering Cogent.
*>i 2607:f8b0:4013::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 150 0 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i *>i 2607:f8b0:4014::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 110 0 6939 6461 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i *>i 2607:f8b0:4016::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 150 0 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i
This is only 3 IPv6 prefixes (out of 47 prefixes seen in my IPv6 routing table). Two of these prefixes I see via direct peering with Google and, alternatively, via Cogent through Zayo transit. One of these prefixes doesn’t advertise in Google’s direct peering session (at least not in mine, but HE picks it up via Zayo and Cogent picks it up via Zayo).
All of these are /48 subnets of their greater 2620:f8b0::/32 prefix, which does show up in both their direct session and in HE via Zayo.
On Mar 13, 2016, at 9:31 AM, Dennis Burgess <dmburgess@linktechs.net> wrote:
In the end, google has made a choice. I think these kinds of choices will delay IPv6 adoption.
-----Original Message----- From: Damien Burke [mailto:damien@supremebytes.com] Sent: Friday, March 11, 2016 2:51 PM To: Mark Tinka <mark.tinka@seacom.mu>; Owen DeLong <owen@delong.com>; Dennis Burgess <dmburgess@linktechs.net> Cc: North American Network Operators' Group <nanog@nanog.org> Subject: RE: Cogent - Google - HE Fun
Just received an updated statement from cogent support:
"We appreciate your concerns. This is a known issue that originates with Google as it is up to their discretion as to how they announce routes to us v4 or v6.
Once again, apologies for any inconvenience."
And:
"The SLA does not cover route transit beyond our network. We cannot route to IPs that are not announced to us by the IP owner, directly or through a network peer."
So if someone (say an eyeball network) was putting out a RFQ for a gig say of upstream cxn and wanted to spec full reachability to the full V6 net, what would the wording for that spec look like? Would that get $provider's attention? On Mar 15, 2016 12:50 AM, "Todd Crane" <todd.crane@n5tech.com> wrote:
This is only tangentially related but it looks like HE has surpassed Cogent on IPv4 adjacencies. That said the source probably suffers from a selection bias at the very least.
Hit reply by mistake instead of reply all.
Todd Crane
On Mar 14, 2016, at 8:40 PM, Matthew D. Hardeman <mhardeman@ipifony.com> wrote:
It looks like Google is experimenting with a change in course on this issue.
Here’s a look at the IPv6 routing table tonight on my router bordering Cogent.
*>i 2607:f8b0:4013::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 150 0 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i *>i 2607:f8b0:4014::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 110 0 6939 6461 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i *>i 2607:f8b0:4016::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 150 0 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i
This is only 3 IPv6 prefixes (out of 47 prefixes seen in my IPv6 routing table). Two of these prefixes I see via direct peering with Google and, alternatively, via Cogent through Zayo transit. One of these prefixes doesn’t advertise in Google’s direct peering session (at least not in mine, but HE picks it up via Zayo and Cogent picks it up via Zayo).
All of these are /48 subnets of their greater 2620:f8b0::/32 prefix, which does show up in both their direct session and in HE via Zayo.
On Mar 13, 2016, at 9:31 AM, Dennis Burgess <dmburgess@linktechs.net> wrote:
In the end, google has made a choice. I think these kinds of choices will delay IPv6 adoption.
-----Original Message----- From: Damien Burke [mailto:damien@supremebytes.com] Sent: Friday, March 11, 2016 2:51 PM To: Mark Tinka <mark.tinka@seacom.mu>; Owen DeLong <owen@delong.com>; Dennis Burgess <dmburgess@linktechs.net> Cc: North American Network Operators' Group <nanog@nanog.org> Subject: RE: Cogent - Google - HE Fun
Just received an updated statement from cogent support:
"We appreciate your concerns. This is a known issue that originates with Google as it is up to their discretion as to how they announce routes to us v4 or v6.
Once again, apologies for any inconvenience."
And:
"The SLA does not cover route transit beyond our network. We cannot route to IPs that are not announced to us by the IP owner, directly or through a network peer."
On Wed, Mar 16, 2016 at 9:56 AM, Dennis Bohn <bohn@adelphi.edu> wrote:
So if someone (say an eyeball network) was putting out a RFQ for a gig say of upstream cxn and wanted to spec full reachability to the full V6 net, what would the wording for that spec look like? Would that get $provider's attention?
"We would like transit services to the full ipv4 and ipv6 addressable space, we would like our prefixes to be advertised to the whole of the above space as well." then you'd by one (some) connection(s) from 'best option #1' and one(some) connection(s) to 'next best option'. I'm not sure 'rfq' is required here is it? you just call the caida top-10/15 and roll based on cost/performance. There are notable exceptions to network performance (routing performance?) but really they are all the same now, yes? perhaps you would be more concerned not with 'ipv6/v4 reachability' than with how what your customers access (may access in the future) is reachable from the providers in question? and potentially what knobs the providers expose to you for bgp TE functionality?
On Mar 15, 2016 12:50 AM, "Todd Crane" <todd.crane@n5tech.com> wrote:
This is only tangentially related but it looks like HE has surpassed Cogent on IPv4 adjacencies. That said the source probably suffers from a selection bias at the very least.
Hit reply by mistake instead of reply all.
Todd Crane
On Mar 14, 2016, at 8:40 PM, Matthew D. Hardeman <mhardeman@ipifony.com> wrote:
It looks like Google is experimenting with a change in course on this issue.
Here’s a look at the IPv6 routing table tonight on my router bordering Cogent.
*>i 2607:f8b0:4013::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 150 0 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i *>i 2607:f8b0:4014::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 110 0 6939 6461 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i *>i 2607:f8b0:4016::/48 2620:121:a000:f0::2(fe80::618:d6ff:fef1:c540) 0 150 0 15169 i * 2001:550:2:22::1d:1(fe80::12f3:11ff:fe29:2c24) 0 90 0 174 6461 15169 i
This is only 3 IPv6 prefixes (out of 47 prefixes seen in my IPv6 routing table). Two of these prefixes I see via direct peering with Google and, alternatively, via Cogent through Zayo transit. One of these prefixes doesn’t advertise in Google’s direct peering session (at least not in mine, but HE picks it up via Zayo and Cogent picks it up via Zayo).
All of these are /48 subnets of their greater 2620:f8b0::/32 prefix, which does show up in both their direct session and in HE via Zayo.
On Mar 13, 2016, at 9:31 AM, Dennis Burgess <dmburgess@linktechs.net> wrote:
In the end, google has made a choice. I think these kinds of choices will delay IPv6 adoption.
-----Original Message----- From: Damien Burke [mailto:damien@supremebytes.com] Sent: Friday, March 11, 2016 2:51 PM To: Mark Tinka <mark.tinka@seacom.mu>; Owen DeLong <owen@delong.com>; Dennis Burgess <dmburgess@linktechs.net> Cc: North American Network Operators' Group <nanog@nanog.org> Subject: RE: Cogent - Google - HE Fun
Just received an updated statement from cogent support:
"We appreciate your concerns. This is a known issue that originates with Google as it is up to their discretion as to how they announce routes to us v4 or v6.
Once again, apologies for any inconvenience."
And:
"The SLA does not cover route transit beyond our network. We cannot route to IPs that are not announced to us by the IP owner, directly or through a network peer."
On Mar 16, 2016 10:06 AM, "Christopher Morrow" <morrowc.lists@gmail.com> wrote:
On Wed, Mar 16, 2016 at 9:56 AM, Dennis Bohn <bohn@adelphi.edu> wrote:
So if someone (say an eyeball network) was putting out a RFQ for a gig
say
of upstream cxn and wanted to spec full reachability to the full V6 net, what would the wording for that spec look like? Would that get $provider's attention?
"We would like transit services to the full ipv4 and ipv6 addressable space, we would like our prefixes to be advertised to the whole of the above space as well."
then you'd by one (some) connection(s) from 'best option #1' and one(some) connection(s) to 'next best option'.
I'm not sure 'rfq' is required here is it? ....
I was thinking RFQ with specific requirements might get cogent attention more than a call. Sure they wouldn't change policy for me, but if they were unable to meet quote requirements repeatedly it might have some effect... or am I dreaming? and potentially what knobs
the providers expose to you for bgp TE functionality?
Good thought to include that. Tnx. D.
I think the RFQ idea isn’t a bad one, but I doubt it will have any effect. Cogent already knows that they have customers leaving because of their peering wars. They don’t seem to care. However, if it’s going to be effective, I think the RFQ has to be achievable by most other networks. I propose: ———— Provider must demonstrate a peering policy conducive to maintaining reachability to all publicly advertised space on the internet. Provider must show that they have reachability to all autonomous systems visible from route-views or other publicly accessible looking glass system(s). Provider must demonstrate these capabilities for both IPv4 and IPv6. ———— In this way, you’ve got a succinct, easily achievable criteria that roughly approximates full routes and a relatively clear message that restrictive “pay us or forget about reaching our subscribers” peering policy isn’t going to get them selected as your provider. Owen
On Wed, Mar 16, 2016 at 11:22 AM, Dennis Bohn <bohn@adelphi.edu> wrote:
On Mar 16, 2016 10:06 AM, "Christopher Morrow" <morrowc.lists@gmail.com> wrote:
On Wed, Mar 16, 2016 at 9:56 AM, Dennis Bohn <bohn@adelphi.edu> wrote:
So if someone (say an eyeball network) was putting out a RFQ for a gig say of upstream cxn and wanted to spec full reachability to the full V6 net, what would the wording for that spec look like? Would that get $provider's attention?
"We would like transit services to the full ipv4 and ipv6 addressable space, we would like our prefixes to be advertised to the whole of the above space as well."
then you'd by one (some) connection(s) from 'best option #1' and one(some) connection(s) to 'next best option'.
I'm not sure 'rfq' is required here is it? ....
I was thinking RFQ with specific requirements might get cogent attention more than a call. Sure they wouldn't change policy for me, but if they were unable to meet quote requirements repeatedly it might have some effect... or am I dreaming?
my guess is the same as Owen's ... 'your rfq don't mean squat'. honestly it's not like people don't ask their cogent sales folk for this sort of thing, it's just not cogent's (clearly, given how long the HE/Cogent thing along has persisted) way of doing things. Sometimes your belief system just isn't theirs.
and potentially what knobs
the providers expose to you for bgp TE functionality?
Good thought to include that. Tnx. D.
On 16/Mar/16 17:41, Christopher Morrow wrote:
my guess is the same as Owen's ... 'your rfq don't mean squat'. honestly it's not like people don't ask their cogent sales folk for this sort of thing, it's just not cogent's (clearly, given how long the HE/Cogent thing along has persisted) way of doing things.
Sometimes your belief system just isn't theirs.
The first time I considered buying from Cogent was out of One Wilshire, back in 2010. I did a 1-month PoC with them. The global IPv6 BGP table was around 2,500 routes then. Cogent had only 100 or so, IIRC. I told them I would not sign with them due to this. Fast-forward to 2012, nothing much had changed when they tried to get me to buy from them again (out of London, this time), and I told them why. Then in 2014, they tracked me down again and confirmed they then had a full IPv6 BGP table. So I added them to my network (out of Amsterdam). Cogent form part of the 7x upstreams I have (not to mention all the peering we do). So if they de-peer some network, we aren't stuck. I have them on my network because they add value to some paths on the Internet, and not because of price. I'm one guy, so I can't say whether my actions over the years prompted a warm body within the Cogent machine to rethink their IPv6 strategy. But I think if you refuse to buy from them on principles that matter to you, and you tell them why, it could help. If it doesn't, move on - it won't be your loss. I would, though, say that the amount of support this list is giving Cogent to keep their heart beating is astounding. If there was ever a time for them to listen to their environment, this would be it. But alas... Mark.
On Mar 16, 2016, at 11:43 , Mark Tinka <mark.tinka@seacom.mu> wrote:
On 16/Mar/16 17:41, Christopher Morrow wrote:
my guess is the same as Owen's ... 'your rfq don't mean squat'. honestly it's not like people don't ask their cogent sales folk for this sort of thing, it's just not cogent's (clearly, given how long the HE/Cogent thing along has persisted) way of doing things.
Sometimes your belief system just isn't theirs.
The first time I considered buying from Cogent was out of One Wilshire, back in 2010. I did a 1-month PoC with them.
The global IPv6 BGP table was around 2,500 routes then. Cogent had only 100 or so, IIRC. I told them I would not sign with them due to this.
Fast-forward to 2012, nothing much had changed when they tried to get me to buy from them again (out of London, this time), and I told them why. Then in 2014, they tracked me down again and confirmed they then had a full IPv6 BGP table. So I added them to my network (out of Amsterdam).
Please confirm that you in fact are receiving 174 * 6939 IPv6 paths from them? Seems unlikely to me. Owen
On 16/Mar/16 21:23, Owen DeLong wrote:
Please confirm that you in fact are receiving 174 * 6939 IPv6 paths from them?
Seems unlikely to me.
Nope (neither IPv4 nor IPv6) - they are about 1,500 IPv6 routes short from what we see from the others. You're welcome to poke if you want to test my perspective: http://as37100.net/ They've obviously regressed a little bit, although it appears they never did have any engagement with HE in particular, for either IP protocol. In fairness, we knew getting into bed with Cogent would bring Daily Joy, which is why we considered them last of all the major networks to on-board. But as I said before, we have sufficient transit and peering that Cogent's insufficiencies do not impact us. For now, what they have on their network offers us some value (and they aren't necessarily any cheaper than any of our other transit providers). If that value should drop below a level where having them on the network is neither here nor there, they'll get the boot. Mark.
On Mar 16, 2016, at 12:42 , Mark Tinka <mark.tinka@seacom.mu> wrote:
On 16/Mar/16 21:23, Owen DeLong wrote:
Please confirm that you in fact are receiving 174 * 6939 IPv6 paths from them?
Seems unlikely to me.
Nope (neither IPv4 nor IPv6) - they are about 1,500 IPv6 routes short from what we see from the others.
Which means that they didn’t meet your requirements, but you bought from them anyway. Even in 2014, they still don’t have a full IPv6 table, despite their claim to the contrary.
You're welcome to poke if you want to test my perspective:
I believe you.
They've obviously regressed a little bit, although it appears they never did have any engagement with HE in particular, for either IP protocol. In fairness, we knew getting into bed with Cogent would bring Daily Joy, which is why we considered them last of all the major networks to on-board.
But as I said before, we have sufficient transit and peering that Cogent's insufficiencies do not impact us. For now, what they have on their network offers us some value (and they aren't necessarily any cheaper than any of our other transit providers). If that value should drop below a level where having them on the network is neither here nor there, they'll get the boot.
Sure, that’s valid and I’m not criticizing your decision. Just saying that according to you, Cogent outright lied to you in 2014 and you let them get away with it. Owen
On 16/Mar/16 22:17, Owen DeLong wrote:
Sure, that’s valid and I’m not criticizing your decision. Just saying that according to you, Cogent outright lied to you in 2014 and you let them get away with it.
I probably should have been clearer in stating that between 2010 and 2014, Cogent's IPv6 coverage improved significantly. Although we knew it was not the complete view, it was close and had no material impact on our IPv6 capabilities re: our customers either way, as a function of the value their network offered us overall for the amount of money we pay to them. In 2010 and 2012, Cogent would have been in a position to be the sole or one of two upstreams for the networks I represented. In 2014, they are one of 7x upstreams + tons of peering. So we were more relaxed. Mark.
On 16 March 2016 at 14:56, Dennis Bohn <bohn@adelphi.edu> wrote:
So if someone (say an eyeball network) was putting out a RFQ for a gig say of upstream cxn and wanted to spec full reachability to the full V6 net, what would the wording for that spec look like? Would that get $provider's attention?
But is that even possible to deliver? I might have some address space that I only advertised with no export to a single peer - does that count? If some third party decides to stop advertising a prefix to $provider are they then in breach of contract with no way to resolve it? If so, I want to sign up and then I will pull some insignificant prefix, just so I can demand $5 million USD in ransom money. Google decided they have some prefixes they don't want to advertise to Cogent. They did offer a reasonable way for Cogent to resolve that issue, but what if Google werent reasonable? Do you still demand that Cogent cave in to anything? I see no easy way here other than let the market decide. If Cogent sucks they will get less traffic and less customers. Or maybe someone finds them useful at the pricepoint they offer. Regards, Baldur
On Wed, Mar 16, 2016 at 9:56 AM, Dennis Bohn <bohn@adelphi.edu> wrote:
So if someone (say an eyeball network) was putting out a RFQ for a gig say of upstream cxn and wanted to spec full reachability to the full V6 net, what would the wording for that spec look like?
Maybe require something roughly like this in the SLA: "Customer may notify Provider upon discovery of a network Partition. A Partition exists when correct BGP routes available via at least 90% of comparable Internet service providers are absent from Provider's BGP feed or do not otherwise function. Where such Partition persists for at least 6 hours from notification, Provider shall make a 100% service credit starting from notification. Where such Partition persists for at least 24 hours, Customer may terminate this contract without penalty until 30 days following the Partition's end."
Would that get $provider's attention?
No. They'll either agree blindly or consider you a hard case. Either way it won't change their actual behavior. Regards, Bill Herrin -- William Herrin ................ herrin@dirtside.com bill@herrin.us Owner, Dirtside Systems ......... Web: <http://www.dirtside.com/>
participants (7)
-
Baldur Norddahl -
Christopher Morrow -
Dennis Bohn -
Mark Tinka -
Owen DeLong -
Todd Crane -
William Herrin