I am looking for user experiences for people who have purchased transit from cogent in the 300Mbps or up range as far as performance, stability, and any other measurable metric of quality you can come up with. We have heard a lot of negatives about them, about their pricing model, about their network, about de-peering with Level 3, etc. What we really need is actual information. Thank you for your time. -Drew
At 09:30 PM 3/7/2006, Drew Weaver wrote:
I am looking for user experiences for people who have purchased transit from cogent in the 300Mbps or up range as far as performance, stability, and any other measurable metric of quality you can come up with.
We have heard a lot of negatives about them, about their pricing model, about their network, about de-peering with Level 3, etc. What we really need is actual information.
Thank you for your time.
From a global perspective[1], the top 12 (I stopped at Cogent since you are asking about them) service providers whose customers and peering partners reach the largest number of networks are listed below. You can make some fairly interesting assumptions on your own: 1 Level 3 Communications, LLC AS 3356 2 Sprint AS 1239 3 UUNET Technologies, Inc. AS 701 4 AT&T WorldNet Services AS 7018 5 Qwest AS 209 6 NTT America, Inc. AS 2914 7 Global Crossing AS 3549 8 China Telecom AS 4134 9 TeliaNet Global Network AS 1299 10 DoD Network Information Center AS 721 11 Savvis AS 3561 12 Cogent Communications AS 174 [1] http://tinyurl.com/lepnl YMMV, -M< -- Martin Hannigan (c) 617-388-2663 Renesys Corporation (w) 617-395-8574 Member of Technical Staff Network Operations hannigan@renesys.com
On Wed, 8 Mar 2006, Martin Hannigan wrote:
I am looking for user experiences for people who have purchased transit from cogent in the 300Mbps or up range as far as performance, stability, and any other measurable metric of quality you can come up with.
We have heard a lot of negatives about them, about their pricing model, about their network, about de-peering with Level 3, etc. What we really need is actual information.
Much of the negatives is from jaded competitors who don't want to fairly compete. Other than that, the answer is 'it depends'.
At certain cities, your experience will be worse - Cogent doesn't have peers with big boys in every city they are at - so you'll have more chance of being backhauled to sfo/iad than if you bought from $bigger-carrier. With regard to depeerings: they are a fact of life on the internet - and as a service provider, you should always have multiple transits, for this and other reasons. Yes, you obviously will have more risk of being caught in a depeering fight if you are buying from $low-price-leader-du-jour, because these are the ones more likely to be depeered by $big-boys for being "too-competitive". ;) With regard to network stability: It *appears* (from number of recent fiber cuts) that Cogent doesn't have enough redundancy on intercity or metro transports - fairly recently network was cut in half for extended period of time due to two concurrent cuts. Not to say that doesn't happen to anyone else, happened to Sprint too, but, losing nyc->iad transport (and having everything go through ord) due to metro fiber cut in nyc is somewhat unexpected. With regard to peers: I can't say that cogent's peers are more congested than any other carrier's peers. With regard to price: There are others who sell at about the same price. Cogent is far better than them. :) Overall: Cogent can be a good part of a transit mix. (from Marty)
From a global perspective[1], the top 12 (I stopped at Cogent since you are asking about them) service providers whose customers and peering partners reach the largest number of networks are listed below. You can make some fairly interesting assumptions on your own: <snip>
This gotta be the most meaningless metric ever. What does "reach" mean? More ASNs seen behind given network? What does it tell, precisely? There are ASNs which have significant chunks of intarweb (say, AS1668) behind them, while AS721 is not likely to matter in a grand scheme of things, even though all .mil installations are behind it. Note that many Cogent customers, while using Cogent for outbound, prefer not to announce any routes to Cogent for political reasons (or prepend or depref their routes). So, that metric won't be exactly helpful.
On Wed, Mar 08, 2006 at 01:56:00AM -0500, alex@pilosoft.com wrote:
(from Marty)
From a global perspective[1], the top 12 (I stopped at Cogent since you are asking about them) service providers whose customers and peering partners reach the largest number of networks are listed below. You can make some fairly interesting assumptions on your own: <snip>
This gotta be the most meaningless metric ever. What does "reach" mean?
now, that's a bit harsh. 'incompletely specified in this email', perhaps, but 'meaningless'? come on. :-) the marketing description for those rankings is: "Identify service providers who are responsible for meeting Internet transit needs (directly or indirectly) of significant numbers of large customer networks within a given market, including both customer and peering relationships." in actual, technical detail, what that means is that we take global routing data (routeviews-style), determine the relationship of every edge (customer, provider, peer), and weight the 'downstream cone' (a caida term that is useful here) of each provider by the scaled prefix space and the degree to which that provider actually provided transit to that prefix space over the course of time. it's critical to integrate across time and across lots of peers in this process. otherwise, the position of your peers and leaks come to dominate the equation. this metric (one of several that we've begun calculating daily) isn't perfect, but we've found that for large SPs it matches expectations of people familiar with traffic flows.
Note that many Cogent customers, while using Cogent for outbound, prefer not to announce any routes to Cogent for political reasons (or prepend or depref their routes). So, that metric won't be exactly helpful.
this is, in fact, a useful point. detecting and compensating for asymmetric routing is difficult in a metric such as this, although it's probably not impossible. t. -- _____________________________________________________________________ todd underwood +1 603 643 9300 x101 renesys corporation chief of operations & security todd@renesys.com http://www.renesys.com/blog
On 3/8/06, Todd Underwood <todd-nanog@renesys.com> wrote:
On Wed, Mar 08, 2006 at 01:56:00AM -0500, alex@pilosoft.com wrote:
(from Marty)
From a global perspective[1], the top 12 (I stopped at Cogent since you are asking about them) service providers whose customers and peering partners reach the largest number of networks are listed below. You can make some fairly interesting assumptions on your own: <snip>
This gotta be the most meaningless metric ever. What does "reach" mean?
now, that's a bit harsh. 'incompletely specified in this email', perhaps, but 'meaningless'? come on. :-)
the marketing description for those rankings is: "Identify service providers who are responsible for meeting Internet transit needs (directly or indirectly) of significant numbers of large customer networks within a given market, including both customer and peering relationships."
in actual, technical detail, what that means is that we take global routing data (routeviews-style), determine the relationship of every edge (customer, provider, peer), and weight the 'downstream cone' (a caida term that is useful here) of each provider by the scaled prefix space and the degree to which that provider actually provided transit to that prefix space over the course of time. it's critical to integrate across time and across lots of peers in this process. otherwise, the position of your peers and leaks come to dominate the equation.
this metric (one of several that we've begun calculating daily) isn't perfect, but we've found that for large SPs it matches expectations of people familiar with traffic flows.
Note that many Cogent customers, while using Cogent for outbound, prefer not to announce any routes to Cogent for political reasons (or prepend or depref their routes). So, that metric won't be exactly helpful.
this is, in fact, a useful point. detecting and compensating for asymmetric routing is difficult in a metric such as this, although it's probably not impossible.
t.
-- _____________________________________________________________________ todd underwood +1 603 643 9300 x101 renesys corporation chief of operations & security todd@renesys.com http://www.renesys.com/blog
Drew: What i would do is construct an RFP with specific questions about capacity, availability and locations (plus other things i can't remember right now) so that I knew what i was getting into with any provider on if they were congested. We see them all the time for large deals, and i'm sure other providers do too. A decision to purchase from supplier A vs. B should not be done lightly, and without some due diligence pertaining to where YOUR traffic may be coming from or going to. Perhaps a spread sheet of some sort that allows you to consolidate all the responses. Good Luck Peter Cohen
On Mar 8, 2006, at 1:56 AM, alex@pilosoft.com wrote:
At certain cities, your experience will be worse - Cogent doesn't have peers with big boys in every city they are at - so you'll have more chance of being backhauled to sfo/iad than if you bought from $bigger- carrier.
It's not just cities, it's entire countries. Try being on a DSL line in France and getting to a Cogent web server in France.
With regard to depeerings: they are a fact of life on the internet - and as a service provider, you should always have multiple transits, for this and other reasons. Yes, you obviously will have more risk of being caught in a depeering fight if you are buying from $low-price-leader-du-jour, because these are the ones more likely to be depeered by $big-boys for being "too-competitive". ;)
De-peering is a fact of life, but Cogent takes something that other people consider a nuisance and turn it into a Real Problem. No other network has been "de-peered" for multiple days multiple times in the last several years. No other network has refused to provide some type of help (e.g. credits) for customers who were affected by the depeering. (Hell, Cogent offered more help to L3's customers than they did to their own - although many people say they did not honor those offers.) Etc., etc. Cogent claims they are good for the Internet as a whole because they keep prices down. That might be true for people who are only interested in price. Or for people who are interested in partial transit for cheap (same thing, really). But if you plan to single- home or otherwise _depend_ on Cogent, I would be hesitant. -- TTFN, patrick P.S. To be clear, Cogent has lots of peers and works very well for most destinations most of the time. However, is not necessarily what some people need from their provider.
On 3/8/06 8:57 AM, "Patrick W. Gilmore" <patrick@ianai.net> wrote:
On Mar 8, 2006, at 1:56 AM, alex@pilosoft.com wrote:
With regard to depeerings: they are a fact of life on the internet - and as a service provider, you should always have multiple transits, for this and other reasons. Yes, you obviously will have more risk of being caught in a depeering fight if you are buying from $low-price-leader-du-jour, because these are the ones more likely to be depeered by $big-boys for being "too-competitive". ;)
De-peering is a fact of life, but Cogent takes something that other people consider a nuisance and turn it into a Real Problem. No other network has been "de-peered" for multiple days multiple times in the last several years. No other network has refused to provide some type of help (e.g. credits) for customers who were affected by the depeering. (Hell, Cogent offered more help to L3's customers than they did to their own - although many people say they did not honor those offers.)
One way to look at this is that you are getting a very low price per mbps with Cogent. Therefore, when Cogent's CEO decides its in his best interest to partition for a week over a depeering situation, their customer's role is to suck it up. You get what you pay for, and in this case, that means mediocre to average transit with periodic partitioning. Frankly, for the price, that's pretty darn good. If your choice is between Cogent and some other provider, you are making a mistake. Cogent (and other low cost transit providers) can be part of a balanced stable of transit providers. Folks who single-home to Cogent deserve whatever Darwin delivers to them. Do what Peter Cohen said and run an RFP. Every competent network engineer should be able to write an Internet transit RFP. -- Daniel Golding
At 09:35 AM 3/8/2006, Daniel Golding wrote:
On 3/8/06 8:57 AM, "Patrick W. Gilmore" <patrick@ianai.net> wrote:
On Mar 8, 2006, at 1:56 AM, alex@pilosoft.com wrote:
With regard to depeerings: they are a fact of life on the internet - and as a service provider, you should always have multiple transits, for this and other reasons. Yes, you obviously will have more risk of being caught in a depeering fight if you are buying from $low-price-leader-du-jour, because these are the ones more likely to be depeered by $big-boys for being "too-competitive". ;)
De-peering is a fact of life, but Cogent takes something that other people consider a nuisance and turn it into a Real Problem. No other network has been "de-peered" for multiple days multiple times in the last several years. No other network has refused to provide some type of help (e.g. credits) for customers who were affected by the depeering. (Hell, Cogent offered more help to L3's customers than they did to their own - although many people say they did not honor those offers.)
One way to look at this is that you are getting a very low price per mbps with Cogent. Therefore, when Cogent's CEO decides its in his best interest to partition for a week over a depeering situation, their customer's role is to suck it up. You get what you pay for, and in this case, that means mediocre to average transit with periodic partitioning. Frankly, for the price, that's pretty darn good.
If your choice is between Cogent and some other provider, you are making a mistake. Cogent (and other low cost transit providers) can be part of a balanced stable of transit providers. Folks who single-home to Cogent deserve whatever Darwin delivers to them.
Do what Peter Cohen said and run an RFP. Every competent network engineer should be able to write an Internet transit RFP.
More of my point was geared toward what seemed like choosing a network mostly on an economic argument when there are more sophisticated ways to accomplish the same thing - and potentially get what you want, but with better access. For example: EXAMPLE METRO (A real metro, names changed to protect the innocent) 1Network-Sierra 2Network-Lima 3Network-Taco41 4<https://demo.renesys.com/mi/browse.do?net=174>Cogent Communications If this is what a metro area market looks like in terms of leadership, there is clear pressure put on the metro area pricing by Cogent. There is no reason to not exploit this fact. The response by the others may be to at least get to a range where you are more comfortable and able to get top tier transit vs. crap shooting. There are other factors, assume that they've been addressed and it's down to $ and performance. Yes, Cogent may be bigger than they "look", which is another argument to use in pricing negotiations for transit, IMHO. This won't fit everyone, but it should fit transit buyers ok. -M< -- Martin Hannigan (c) 617-388-2663 Renesys Corporation (w) 617-395-8574 Member of Technical Staff Network Operations hannigan@renesys.com
dan, all, On Wed, Mar 08, 2006 at 09:35:15AM -0500, Daniel Golding wrote:
If your choice is between Cogent and some other provider, you are making a mistake. Cogent (and other low cost transit providers) can be part of a balanced stable of transit providers. Folks who single-home to Cogent deserve whatever Darwin delivers to them.
we referred to this in the presentation about the depeering ( http://nanog.org/mtg-0510/underwood.html ), but just to update it: i see that that 258 ASes are relying on cogent for all or substantially all (>95%) of their connectivity to the Internet. many of these may be multi-homed and just never using their backup connection, but still, that's a lot of people trusting cogent for their day-in day-out connectivity. i'm not a cogent cheerleader or basher (or customer). just here to provide concrete, objective info. also, in response to a couple of questions that i've taken in private about the rankings: the good part about these network rankings that we've been doing is that they are objective, repeatable and difficult to falsify. this is in contrast to traffic stats, that are notoriously unreliable (read: false, lies) and difficult to confirm. there are certainly a couple of potential optimizations available in the future, though. we think that the current customer-base rankings may under-represent consumer broadband networks (such as comcast) , in part because these networks tend to generate substantially more traffic for each amount of prefix-space than do many other kinds of networks. we're thinking about ways to model that difference. the other reason we think that they may under-represent these networks is because many of them still use lots of different ASes--BellSouth is a good example here. although we could (and probably will) fix that with some AS clustering algorithms, the ASes in question could (and probably will) fix that by consolidating ASes, too. :-) the current rankings are accurate with respect to routing and are a reasonable approximation of some of the other metrics that people wish could be directly measured (traffic, market power, etc.). [people should also be aware that we're producing 5 different rankings that measure different kinds of things, so just focusing on the customer-base ranking may be a mistake]. t. -- _____________________________________________________________________ todd underwood +1 603 643 9300 x101 renesys corporation chief of operations & security todd@renesys.com http://www.renesys.com/blog
On Mar 8, 2006, at 9:35 AM, Daniel Golding wrote:
One way to look at this is that you are getting a very low price per mbps with Cogent. Therefore, when Cogent's CEO decides its in his best interest to partition for a week over a depeering situation, their customer's role is to suck it up. You get what you pay for, and in this case, that means mediocre to average transit with periodic partitioning. Frankly, for the price, that's pretty darn good.
My biggest complaint about Cogent 'Customer Service' is that I'm not a Cogent customer, I'm a Verio customer that was sold to Cogent. I'm still paying the higher Verio bandwidth price but getting the 'not as good' Cogent bandwidth. When Cogent decides to depeer is affects me and I would like credits. Either that or cancel my Verio priced contact and replace it with a Cogent priced contact. If they did that, I wouldn't mind the occasional depeering. Trying to explain that to my sales guy with impossible.
If your choice is between Cogent and some other provider, you are making a mistake. Cogent (and other low cost transit providers) can be part of a balanced stable of transit providers. Folks who single-home to Cogent deserve whatever Darwin delivers to them.
That is why I also have 3 providers, just in case 2 of them decide they don't like each other. Anyone out there running a RouteScience/Internap box on some Cogent + other provider bandwidth? How many routes get moved off/onto Cogent? -- Matthew S. Crocker Vice President Crocker Communications, Inc. Internet Division PO BOX 710 Greenfield, MA 01302-0710 http://www.crocker.com
At 08:57 AM 3/8/2006, Patrick W. Gilmore wrote:
On Mar 8, 2006, at 1:56 AM, alex@pilosoft.com wrote:
At certain cities, your experience will be worse - Cogent doesn't have peers with big boys in every city they are at - so you'll have more chance of being backhauled to sfo/iad than if you bought from $bigger- carrier.
It's not just cities, it's entire countries. Try being on a DSL line in France and getting to a Cogent web server in France.
How is this different from being a Comcast cable modem customer in New England, trying to connect to a web server also located in New England. Packets route through NYC if the user is lucky, but more often Chicago or Washington DC. In terms of mileage and latency, just how different is that from the DSL case in France you cite? Reality is "broadband" providers in some areas have sucky, or non-existant, peering. Do you blame that on the backhaul network, or on the "broadband" provider?
With regard to depeerings: they are a fact of life on the internet - and as a service provider, you should always have multiple transits, for this and other reasons. Yes, you obviously will have more risk of being caught in a depeering fight if you are buying from $low-price-leader-du-jour, because these are the ones more likely to be depeered by $big-boys for being "too-competitive". ;)
De-peering is a fact of life, but Cogent takes something that other people consider a nuisance and turn it into a Real Problem. No other network has been "de-peered" for multiple days multiple times in the last several years. No other network has refused to provide some type of help (e.g. credits) for customers who were affected by the depeering. (Hell, Cogent offered more help to L3's customers than they did to their own - although many people say they did not honor those offers.)
Etc., etc.
Cogent claims they are good for the Internet as a whole because they keep prices down. That might be true for people who are only interested in price. Or for people who are interested in partial transit for cheap (same thing, really). But if you plan to single- home or otherwise _depend_ on Cogent, I would be hesitant.
As others have said, cogent is OK as part of a transit mix, but not necessarily as a single homed provider. That said, they're far from the only network (including the biggest names/networks) that I would say that about. Everyone's networks have meltdowns at different times. Everybody seems to get into pissing matches.
-- TTFN, patrick
P.S. To be clear, Cogent has lots of peers and works very well for most destinations most of the time. However, is not necessarily what some people need from their provider.
On 3/8/06, Daniel Senie <dts@senie.com> wrote:
At 08:57 AM 3/8/2006, Patrick W. Gilmore wrote:
On Mar 8, 2006, at 1:56 AM, alex@pilosoft.com wrote:
It's not just cities, it's entire countries. Try being on a DSL line in France and getting to a Cogent web server in France.
How is this different from being a Comcast cable modem customer in New England, trying to connect to a web server also located in New England. Packets route through NYC if the user is lucky, but more often Chicago or Washington DC. In terms of mileage and latency, just how different is that from the DSL case in France you cite?
Boston, or New England in general, is somewhat of a oddity. It has a very high concentration of users but little to no peering in general. Back in the mid-late 90's when I was at BBN we tried to do private interconnects with folks, and did manage a few, but the majority of providers would rather backhaul to NYC. Our assumption was that (generically) most fiber networks considered NYC the edge of the North East and the hinterlands north would be considered spurs or regional rings (or whatever you would like to call them). -Steve -- -Steve
On Mar 8, 2006, at 11:43 AM, Daniel Senie wrote:
At 08:57 AM 3/8/2006, Patrick W. Gilmore wrote:
It's not just cities, it's entire countries. Try being on a DSL line in France and getting to a Cogent web server in France.
How is this different from being a Comcast cable modem customer in New England, trying to connect to a web server also located in New England. Packets route through NYC if the user is lucky, but more often Chicago or Washington DC. In terms of mileage and latency, just how different is that from the DSL case in France you cite?
Over twice as far. Does that matter to you? I don't know. But hopping an ocean twice to go down the street is not my idea of "optimal". Also, while I agree that more New England providers trade bits through DC than NYC, I would submit that most bits are traded through NYC or DC, not Chicago. Neither NYC or DC is really that far from Boston (he types sitting in Cambridge, MA). And that doesn't even address the difference between user <-> user and user <-> server interaction.
Reality is "broadband" providers in some areas have sucky, or non- existant, peering. Do you blame that on the backhaul network, or on the "broadband" provider?
Personally, I blame whomever I am paying. :) But you can blame whomever you like.
As others have said, cogent is OK as part of a transit mix, but not necessarily as a single homed provider. That said, they're far from the only network (including the biggest names/networks) that I would say that about. Everyone's networks have meltdowns at different times. Everybody seems to get into pissing matches.
There is a very large difference between a "meltdown" and a multi-day disconnection from major ASNs. Can you name any other network which has been disconnected from large swaths of the Internet for _DAYS_ this century? Can you name another network who has had such diconnectivity events and told their customers they would do _nothing_ for them, during or after? And in fact pointed their customers at the _other_ network to complain? Every network has problems. Good networks take responsibility for those problems. -- TTFN, patrick
On Tue, 7 Mar 2006, Drew Weaver wrote:
I am looking for user experiences for people who have purchased transit from cogent in the 300Mbps or up range as far as performance, stability, and any other measurable metric of quality you can come up with.
We have heard a lot of negatives about them, about their pricing model, about their network, about de-peering with Level 3, etc. What we really need is actual information.
Drew: Assuming you're looking for operational detail, I'll offer some of my experience with Cogent. Note that some of this is deduced based on visible symptoms, while some is closer to fact. When I decided to go with Cogent as one of my providers, I knew I'd get no more than I was paying for. I just didn't know how that would manifest itself. Since then I've found out a few ways they reduce costs: o few spare parts, such as using a parts depot in DC to cover Chicago o fragile peering, both technically & contractually o lack of well-published BGP knobs, such as communities to influence localpref... (they apparently exist, but are not well documented that I've been able to find) o deferred hardware upgrades, such as using old 15454 cards which most folks had replaced due to stability & performance problems o limited engineering coverage, requiring their NOC folks to drag in on-call folks while you wait On the positive side: o their NOC folks are responsive & willing to try to help o it does work most of the time to reach most places o it's cheap As others have said, Cogent is a useful piece of a transit arrangement as long as you go in with your eyes open. Let me know if you want more detail on any of my experiences with them. ________________________________________________________________________ Jay Ford, Network Engineering Group, Information Technology Services University of Iowa, Iowa City, IA 52242 email: jay-ford@uiowa.edu, phone: 319-335-5555, fax: 319-335-2951
On Wed, Mar 08, 2006 at 10:28:20AM -0600, Jay Ford wrote: [ .. snip .. ]
o lack of well-published BGP knobs, such as communities to influence localpref... (they apparently exist, but are not well documented that I've been able to find)
have you contacted support@cogentco.com for url to Cogent User Manual/Guide? It's all documented there, just letting you know. James
On Wed, 8 Mar 2006, James wrote:
On Wed, Mar 08, 2006 at 10:28:20AM -0600, Jay Ford wrote:
[ .. snip .. ]
o lack of well-published BGP knobs, such as communities to influence localpref... (they apparently exist, but are not well documented that I've been able to find)
have you contacted support@cogentco.com for url to Cogent User Manual/Guide? It's all documented there, just letting you know.
So it is. I stand corrected on that point. Thanks for the reminder. I think the rest of my rambling was valid. Jay
Drew Weaver wrote:
We have heard a lot of negatives about them, about their pricing model, about their network, about de-peering with Level 3, etc. What we really need is actual information.
Here's a good one about Cogent. 100BaseTX connection from us to a Cogent Cat3550 ("A"). "A" connects to Houston core router "B". "Normal" Cogent BGP setup: A announces a /32 on B, B announces full table, we announce our prefixes to A and nothing to B. A and B are (supposedly) directly connected. Our BGP session with B flapped numerous times this morning. Cogent NOCperson said that a flaky router in San Jose had problems, and had to be rebooted. As a result, "you couldn't reach your B peer". Yeah, that makes sense. Oh, and problems in Miami plus the problems in San Jose were causing problems in Boston and New York. Sure. YMMV, but not a confidence-inspiring answer IMO. pt
participants (13)
-
alex＠pilosoft.com -
Daniel Golding -
Daniel Senie -
Drew Weaver -
James -
Jay Ford -
Martin Hannigan -
Matthew Crocker -
Patrick W. Gilmore -
Pete Templin -
Peter Cohen -
Steve Meuse -
Todd Underwood