This is interesting. And it will be detrimental to network neutrality supporters. Cogent admits that while they were publicly complaining about other networks congesting links, they were using QoS to make the problem look worse.
One of the problems in "tech" is most people do not realize tone is important, not just substance. There was - still is! - congestion in many places where consumers have one or at most two choice of providers. Even in places where there are two providers, both are frequently congested. Instead of discussing the fact there is no functioning market, no choice for the average end user, and how to fix it, we will now spend a ton of time arguing whether anything is wrong at all because Cogent did this.
Wouldn't you rather be discussing whether 4 Mbps is really broadband? (Anyone else have flashbacks to "640K is enough for anyone!"?) Or how many people have more than one choice at 25 Mbps? Or whether a company with a terminating access monopoly can intentionally congest its edge to charge monopoly rents on the content providers their paying customers are trying to access? I know I would.
Instead, we'll be talking about how things are not really bad, Cogent just made it look bad on purpose. The subtlety of "it _IS_ bad, Cogent just shifted some of the burden from VoIP to streaming" is not something that plays well in a 30 second sound bite, or at congressional hearings.
It's enough to make one consider giving up the idea of having a functioning, useful Internet.
Network SLAs are usually on-net. Deciding how to queue packets down a congested link is certainly something many places have done for years, including when people did Random Early Discard(RED), Weighted RED or even more advanced AQM when there may be one-way congestion (Eg: cable/dsl uplink) at the home.
As a technical issue I always want to see congestion addressed promptly with either changes in the traffic pattern or network upgrades. If you have customers on a fixed monthly plan regardless of usage and your capital model doesn’t address that, or you hide the network costs in other ‘bundles’ it may become harder to do the accounting necessary to fund those upgrades. I do wish it were easier to get symmetric speeds on DOCSIS/xDSL technologies.
I noticed that Cogent has a Net Neutrality statement. If I understand what they disclosed on the M-Lab list it does not seem to jive with this. The second sentence seems like what they said they are doing, right?
"Cogent practices net neutrality. We do not prioritize packet transmissions on the basis of the content of the packet, the customer or network that is the source of the packet, or the customer or network that is the recipient of the packet.
It is Cogent's belief that both the customer and the Internet as a whole are best served if the application layer remains independent from the network. Innovation in the development of new applications is fueled by the individual's ability to reach as many people as possible without regard to complicated gating factors such as tiered pricing or bandwidth structures used by legacy service providers.
Applications proliferate in a free market economy which is the Internet today."
So they're de-prioritizing it on the basis of content, source, or recipient,
all the way to priority "NEVER".
That means they're prioritizing everything else ahead of it.
=> they're prioritizing "packet transmissions on the basis of the content of
the packet, the customer or network that is the source of the packet, or
the customer or network that is the recipient of the packet.", for all the
other packets,
The way I read it was that Cogent actually made things look artificially better for M-Labs while simultaneously making it much worse for one subset of their users and somewhat better for others.
I would suggest that if we get the educational process right, we should be able to explain that the point where you’re having to select traffic to prioritize is the point where your network is inadequate to the task at hand and should be upgraded.
I don’t see any reason we shouldn’t be able to use this article as a prime example of a provider doing the wrong thing instead of fixing the real problem — Congestion at exchange points.
If I were a Cogent customer I would like to have seen more transparency (an announcement at least). However, I don't see anything wrong with their practice of giving some customers "Silver" service and others "Bronze" service while reserving "Gold" for themselves. Even if applications like VoIP do not function well with a Bronze service level.
Now, a customer that was under the impression they were receiving equal treatment with other customers may not be happy to know they were receiving a lower class of service than expected. This is not a net neutrality matter, it's a matter of expectations and possibly false or deceptive advertising.
I would much rather see an environment where the customer gets to choose Gold, Silver, and Bronze levels of service for his or her traffic as opposed to an environment where the provider chooses fast/slow lane applications at their own discretion.
Personally I hope that such an environment never happens. Fast/slow lanes are pretty meaningless. Such service differentiation only has meaning when there’s persistent congestion and I’d rather that networks work out ways to scale past demand rather than throttle them.
Owen, should providers be able to over subscribe their networks? If so, at what tier level (tier 1, 2, 3, residential ISP)? Is it acceptable for a provider to permit frequent congestion if they choose to? Or should they be forced to take action that may (potentially) lead to increased customer rates or reduced customer bandwidth?
I do think that Cogent's customers likely expect to receive their full subscription rate, without congestion, nearly 100% of the time (at least within the Cogent network). This would mean that having congestion is a problem and QoS is not a solution to congestion. However, I don't think all customers of all IP transit providers have this expectation. For example, residential customers may be happy with "up to X Mbps" if the costs associated are 1/10th that of a "guaranteed X Mbps" service. This is essentially the difference between "Bronze" and "Silver" service levels. As long as market choice exists, I see no problem with a provider choosing to operate a slow, inconsistent, or unreliable network as long as the internet as a whole, being a piece of critical communications infrastructure, remains available and reliable. Effectively, this would mean that tier 1 and 2 transit providers (including Cogent) would need to be consistent and reliable. While regional transit providers and ISPs would be given much more flexibility. Regardless, I think letting transit providers/ISPs pick winners and losers is a losing strategy in the long term.
Tier levels are marketing terms - irrelevant to technical/operational discussions.
Every provider oversubscribes to some level, whether they're in the last mile serving residential users, or a carrier of carriers. It's just a question of what amount of oversubscription is acceptable, and what the risks are when customers blow that oversubscription model out of the water, either in the short term (streaming major sporting events, etc), or in the longer term (increased prevalence of streaming video, rich content, etc). Congestion due to short-term spikes is often seen as an acceptable risk. Congestion due to long-term shifts in customer network usage habits requires the oversubscription model to be re-worked, or the provider (and by extension... their customers) accepts a reputation of not dealing proactively with congestion.
Owen, should providers be able to over subscribe their networks? If so, at what tier level (tier 1, 2, 3, residential ISP)? Is it acceptable for a provider to permit frequent congestion if they choose to? Or should they be forced to take action that may (potentially) lead to increased customer rates or reduced customer bandwidth?
Oversubscribe, of course. Overrun with persistent congestion, no.
1. Yes.
2. Any.
3. No, not really.
4. Yes.
I do think that Cogent's customers likely expect to receive their full subscription rate, without congestion, nearly 100% of the time (at least within the Cogent network). This would mean that having congestion is a problem and QoS is not a solution to congestion. However, I don't think all customers of all IP transit providers have this expectation. For example, residential customers may be happy with "up to X Mbps" if the costs associated are 1/10th that of a "guaranteed X Mbps" service. This is essentially the difference between "Bronze" and "Silver" service levels. As long as market choice exists, I see no problem with a provider choosing to operate a slow, inconsistent, or unreliable network as long as the internet as a whole, being a piece of critical communications infrastructure, remains available and reliable. Effectively, this would mean that tier 1 and 2 transit providers (including Cogent) would need to be consistent and reliable. While regional transit providers and ISPs would be given much more flexibility. Regardless, I think letting transit providers/ISPs pick winners and losers is a losing strategy in the long term.
Cogent is not a residential ISP. They are a business to business provider. As such, it should be possible to assume that their customers reasonably understand what they are buying and if they do not, the rules of caveat emptor should apply.
Cogent’s failure to upgrade their peering (and their generally poor attitude towards peering overall) are one issue.
The bigger issue in this discussion is their lack of transparency in secretly prioritizing traffic in order to further an agenda. IMHO, such conduct is unethical at best.