Network Neutrality: Classification of Arguments For and Against
This page aims to distinguish different arguments and reasoning in the debate around network neutrality, or control over traffic transmission on digital networks. The page was created to disentangle the many arguments, because the people arguing for and against network neutrality use multiple definitions of the term and mix together many arguments on different levels. The purpose of this page is not to air polemics, but to elucidate the various points made for and against various forms of network neutrality.
The document treats network neutrality is a business practice, and therefore does not cover related topics such as copyright enforcement, censorship, the move of processing and data to remote servers (often called "into the cloud"), policies of mobile providers toward content and applictions, or surveillance. Essentially, the document covers a public issue that started as a set of economic concerns and has been invested by debaters with moral concerns.
Varieties of intervention in Internet traffic
Much confusion among the public, and many instances of debaters talking past each other, are caused by the possibility of differentiating among Internet traffic in many different ways.
In all these cases, an Internet service provider monitors traffic at its routers and chooses to drop certain packets entirely or to send them at a slower frequency. When traffic uses TCP, dropping selected packets near the beginning of a session causes the sender to throttle back the speed of transmission, because TCP assumes that the lack of acknowledgment indicates a congested route.
A router is normally expected to check only crude indicators of traffic such as IP address and port (characteristics stored in Internet packets at the Internet Protocol level). Deep packet inspection is a technique permitted by faster and more sophisticated routers, letting them look at more fine-grained elements of the session and even at application-layer data.
Differentiation by service (port, protocol)
Services that use a lot of bandwidth in a sustained manner can have trouble cohabiting the lines with other traffic. Voice over IP used to fall into this category, although no the protocols have reduced its demands and the bandwidth in most places has increased to the point where it is not a problem, and streaming video has replaced it as a concern. Streaming media such as voice and video are the high-bandwidth transfers that concern both users and ISPs, because any delays except the most minor ones introduce noticeable problems such as jitter, hanging, and indecipherable voice content. Non-streaming services simply take longer to transmit material, but users often don't notice.
By detecting streams and giving them higher bandwidth, ISPs can provide a more desirable service. Differentiation can be accomplished by directing the streams onto different lines reserved for high-priority traffic or by privileging the packets over other traffic. Customers have traditionally reserved high-priority lines and paid extra. Using the same lines would allow an ISP more fine control and more options in offering enhanced services and charging higher revenue for them.
Potentially, if specialized new services become popular (telemedicine and long-distance education are often cited), the ISP can provide the high bandwidth needed for the services and charge the users on one or both ends. If the provider of the service is charged, the charge would presumably be reflected in service fees.
On the other hand, some ISPs have seen VoIP and video streaming as competition, particularly when the ISP is offered by a company that also offers telephone service of cable TV. Thus, these ISPs have been known to ban VoIP or video streaming and take steps to detect and block them.
Although most Internet services have traditionally been offered to individual end-users for free (or more likely, with embedded advertising to pay for them), businesses have paid for Software as a Service for some time, and there is a trend toward end-user services charging subscription fees, such as some journalistic sites. The popularity of services running on mobile devices provides another opportunity for user fees to be introduced on the Internet. Differentiation by ISPs (especially mobile companies) among services allow them to charge for a service and perhaps share the revenue with the site providing the service.
From the beginning of the network neutrality debate, phone and cable companies have claimed that charging for Internet service--usually, charging the provider rather than the end-user--is necessary to provide the funds to bring higher bandwidth to areas that are currently underserved. Differentiated service is seen as a way to promote innovation in networks. In addition to allowing phone and cable companies to differentiate themselves--one perhaps being seen as a friendly site for games, another for telemedicine, etc.--it can also help Internet sites to differentiate themselves on the application layer. For instance, one Internet site can offer faster access to its content by paying for higher bandwidth.
Services such as email, the web, and DNS traffic traditionally use different ports to make it easy to route traffic to the right program, although this is not universally true and some programs deliberately use ports meant for others (especially port 80, which is usually open, at least for incoming traffic, because it is assigned to the Web).
A router can therefore check the port number for a preliminary indication of the traffic being carried. Because programs can use other programs' ports and the port numbers indicate nothing about the actual traffic going between services, deep packet inspection is often used to determine the type of service in use. Thus, although peer-to-peer file transfers could be considered a distinct service, the programs do not use well-known ports and must be detected in other ways.
For instance, many ISPs want to detect Voice over IP for a variety of reasons: to give its packets a higher privilege because it's a streaming service and therefore subject ot jitter, to block the service because the ISP considers it a competitor to its own voice service, or to offer surveillance to law enforcement. Voice over IP tends to use many ports, often including port 80. Therefore, it is usually detected by building into the router a sophisticated understanding of its protocols. A protocol includes various standard forms of data at the start or end of a message (headers and footers), as well as a strict set of messages exchanged to start, end, or manage a session.
Conditions for success
- ISPs must be able to distinguish the traffic they wish to privilege or downgrade, with few false positives (for instance, few instances of dropping or downgrading legitimate traffic). Current deep packet inspection systems have proven very adept at making the distinction, but they are continuously challenged by attempts to bypass them as well as new services that haven't yet been fingerprinted.
- ISPs must not abuse their use of preferential treatment or anger so many customers that they face legal penalties or lose customers.
- If ISPs discriminate against a hostile user, the user must be unable to find work-arounds that defeat discrimination.
- The preferential treatment must make enough of a difference in bandwidth use to be worth the cost.
Differentiation by participants (IP address)
Differentiation is usually provided for pay or on some other contractual basis, so traffic needs to be tied to the organization that has contracted for the service. Organizations often contract for a particular Quality of Service measured in throughput. The ISP is not required to demonstrate a particular response time or other element of the user experience, because such measures can be subjective and are affected by many elements of the environment outside the performance of the network. But some aspects of quality, such as the number of packets sent, received, and dropped over a period of time, are easy to measure.
Most people and institutions have one or more contiguous IP addresses assigned by their ISP. (A few institutions with historic roots in the Internet were assigned IP addresses directly.) Thus, the institution can be identified by its IP address. Many large organizations also contract with content delivery networks such as Akamai to cache content close to the users and deliver it from servers owned by the content delivery network.
Conditions for success
- The ISPs must persuade a large number of sites to pay for the preferential treatment. If ISPs try to widen the gap between ordinary ("best-effort") and preferential treatment by deliberately starving ordinary transmissions, they could produce a backlash and legal repercussions.
- The ISPs must earn enough revenue from preferential treatment to pay for a substantial amount of new lines and facilities, and must invest the money this way instead of in other expenditures. Otherwise, the extra revenue just serves as a kind of tax on Internet service with no benefits to the public.
- Culling substantial profits will probably require new form of Internet service: applications and social uses that don't exist yet or are in their infancy.
Differentiation by volume or density of traffic
The number of bytes transmitted corresponds fairly closely to the costs to the ISP. No ISP wants to meter service to the end-user, but the ISP must estimate and aggregate the expected usage among all users in order to provision lines and routers and make deals with other ISPs carrying traffic. Although some ISPs rely on natural network behavior to control the volume of traffic (there is no need for a cap when there's little congestion, and all transmissions will slow down when congestion strikes) caps on transmissions have been common for quite some time on some networks, notably cable Internet service and wireless ISPs. Many ISPs distinguish between "home" and "business" service, requiring customers to pay more for the "business" service if they use more than a certain amount of bandwidth during a month.
Home users are unlikely to strain the network unless they regularly download large amounts of material, and the usual culprits are music and videos. Other individual transfers can be large--such as a major PowerPoint presentation or a Linux distribution--but users are unlikely to conduct enough transfers of such material to make a noticeable dent on the ISP's bandwidth. The stress on the network seem to come with music and video, particularly when peer-to-peer file sharing is used. This usage is also normally illegal because most legitimate copyright holders have other ways to distribute music and video.
Differentiation comes into play when an ISP decides to react dynamically in real time to large data transfers instead of simply telling the customer later to upgrade to a more expensive service. Some ISPs cut off transfers above a certain size or drop packets selectively to degrade the speed of the transfer. The sophisticated techniques mentioned under Differentiation by service (port, protocol) can also be used to identify peer-to-peer behavior.
Conditions for success
These are similar to the conditions for success listed under Differentiation by service (port, protocol).
Ban on "servers"
Although the Internet is designed around a model of communicating peers, most applications use a client/server model. Content is stored in a large central system (the server) and parcelled out to simpler programs (clients) running on the computers of individual users.
For instance, even though email appears to be a communication between peers, each side actually communicates with a server that intermediates between them.
Servers generally handle more traffic than clients, so ISPs who provision resources for home use sometimes include a ban on servers in their terms of service.
In the early 2000s, a peer-to-peer model arose that doesn't fit into the client/server model. Virtually no services run as pure transfers between peers; most require some server or a set of super-peers for connection. But these systems at some point ask the end-users to disconnect from the servers or super-peers and exchange data directly. Most ISPs define this peer behavior as running a server, because it involves higher data transfers than most client programs engage in. The ISPs also justify the ban on the basis that most such peer-to-peer transfers are unauthorized transfers of copyrighted material.
Conditions for success
These are similar to the conditions for success listed under Differentiation by service (port, protocol).
Every ISP combats unsolicited transmissions (spam, phishing, and virus transmission) by checking the content of messages. Suspicious traffic can be dropped or flagged to warn the user. Some security experts would like ISPs to check outgoing traffic as well, so that when an individual user's machine gets infected and starts sending out spam or malicious traffic, the ISP can block it and cut down the spread of infections.
Because at least 95% of all email is estimated now to be spam, end-users clearly benefit from filtering at the ISP level. However, disruptions from this filtering are well-known. Legitimate email can be trapped in error (false positives), and all the users of an ISP may be blocked over a dispute in which a black hole operator believes the ISP is not being strict enough to terminate a malicious user.
Checks for dangerous traffic are usually done at the application layer--for instance, a dedicated system through which all email is run to check for spam and other undesired content.
Conditions for success
- ISPs must be able to distinguish the undesired traffic from legitimate traffic with few enough false positives and false negatives to keep customers.
Objections to differentiation
- Differentiation introduces more variability into a network whose behavior is already unpredictable. Network administrators trying to allocate routing equipment and resources will find their plans frustrated by traffic shaping upstream by ISPs. The secrecy generally maintained by ISPs about the means they use to detect and shape different types of traffic make planning even harder.
- Deep packet inspection introduces more delay overall, although the delay should be absorbed by the ISP doing the inspection by installing faster routers.
- Providing preferred delivery to customers who pay extra is a policy that rests on, and reinforces, an assumption that valuable services can afford to pay for their delivery. Privilege is given to services provided by well-established and well-endowed organizations, services who can draw substantial funding, or services that can persuade new users to pay a premium.
- Services that people expect to be free of charge are more likely to charge money in the race for preferential treatment by ISPs.
- Whereas adequate response time (with occasional outages or blips in speed) was taken for granted as part of standard Internet service, it will now require extra payments. Some services that might have been started in a low-cost environment may never be started or may not develop an audience because the innovators cannot afford to provide the response time to which users are accustomed. This concern is related to the first concern in this list. There are questions, however, as to whether preferential delivery can create such large disparities in response time that new services will really be hampered.
- The newly commercialized environment may dry up new services at just the moment when ISPs rely on the growth of new services to make preferential treatment economically successful. As explained in an article by Andrew Odlyzko, current offerings (he focuses on video) will not be enough.
- Promising a particular Quality of Service is difficult because activity further downstream (among the other organizations carrying traffic from the ISP to other parts of the Internet) has effects on packet transmission. Differentiation on one end, in particular, has effects on the other end. Thus, ISPs may find it hard to provide reasonable contracts for differentiated service that can be monitored and enforced.
- Because ISPs, as owners of the lines, can decide whom to contract with for preferential service, they get to "pick winners and losers" and will effectively have undue influence over other businesses as well as forms of social interaction.
- ISPs might use their power to censor sites (by downgrading their service) or turn social and political debate to topics of their preference.
- Most of the ISPs discussing differential services own the "last mile" or "middle mile" in the chain of networks connecting end users, but are making decisions that affect other network providers. This argument has to deal with the history of the Internet as a voluntary association of many providers, each of whom has the right to make individual decisions about how to run its network.
- Because many techniques used by ISPs can be defeated ("routed around") by customers who know that the techniques are in use, and because some ISPs don't want their customers know that preference is given to certain types of traffic, the ISP is seen as making decisions for the customer and sometimes as dishonest.
- Complaints about discrimination could be dealt with through laws on competition and monopoly, on an individual basis, by such bodies as the Federal Communications Commission or Federal Trade Commission. Objections to this course of action include:
- ISPs will be tempted constantly to test the limits of tolerance for discriminatory behavior, and the details of each case will make it hard to develop precedents.
- Enforcement of anti-trust laws varies widely from one time period to another.
- "Truth in advertising" laws and regulations could be invoked to require the ISPs to tell customers what kinds of differentiation are in use. See for instance David Farber's offering in the Communications of the ACM's Point/Counterpoint Network Neutrality Nuances. Objections include:
- Some forms of differentiation are used to combat fraud and illegal activity, and depend on secrecy to be effective because the actors will work around any known surveillance.
- With limited competition, customers cannot force ISPs to change behavior even if the customers know of and disapprove of differentiation.
- Most new uses of networks start with a tiny cohort of users and then spread if they prove valuable. But if a new use runs up against an ISP ban, both the ISP and the majority of its customers are unlikely to see any reason to lift the ban.
- The FCC could require ISPs to act as common carriers toward services above them. A federal appeals court ruled in April 2010 that the FCC (in a suit brought by Comcast) lacked the authority to require this neutrality. The suit concerned blocking service through traffic analysis, but applies equally to any form of differentiation chosen by an ISP. The FCC could attempt to reassert authority by reclassifying Internet service, but many people believe Congress must or should be brought in to resolve the question through law. Essentially, common carrier status would require every bit to be treated equally, also known as "first-in-first-out" treatment, with possible exceptions for classes of traffic that the government has a reason to suppress. Objections include:
- Economic grounds, because it would disrupt the business plans cited earlier
- Technical grounds, on the belief that regulations tend to overreach their objectives and, in this case, would prevent beneficial forms of traffic shaping
- Moral grounds, on the basis that ISPs should be allowed to contract in any manner they desire with their customers and that regulation would constrain customer choice as well as ISP choice. The argument is that customers will choose competitors if ISPs abuse their control.
- Congress could institute "structural separation" between companies that offer transport and companies who offer services. This would remove the incentive for companies offering transport to play favorites and privilege some transmissions against others either to promote particular application-layer services or to suppress services they see as competitive threats to their own services. Objections include:
- Because people currently are willing to pay more for applications than for transport, separating them might remove the incentives for transport companies to invest in more lines, new routers, and other improvements.
- Structural separation would be an even more wrenching economic change than the historic break-up of AT&T, and would thus either be politically unfeasible or cause long-term disruption to telecommunications and the Internet.
- Competition could provide alternatives to customers, who could then leave ISPs who try to control their choices.
Types of Internet control not covered in this document
Several types of intervention in Internet traffic are not covered in this document because they are related only in indirect ways to the central ISP business issues expressed in the network neutrality debate.
ISPs do not take on the task of detecting and stopping the unauthorized transfer of copyrighted material, such as P2P file sharing of videos and music recordings. However, copyright holders often demand cooperation from ISPs in stopping transfers or catching the parties involved, and use techniques related to the ones described in this paper, notably detecting transfers and using IP addresses to identify the party at one end. Furthermore, ISPs often cite the use of high-volume data transfers and P2P file sharing for illegal purposes as justification for traffic control.
Many of the techniques used to detect various types of transfers for business purposes can also be used for censorship, along with even more sophisticated techniques. For instance, governments often block access to IP addresses associated with news sources they disapprove of, or services used to proxy and hide data transfers.
The techniques in this document are often used by ISPs to support efforts by law enforcement to detect criminals or other people they want to track for any reason.