COMM647/Bandwidth/Notes

From Driscollwiki

Driving questions

Is the pursuit of bandwidth in competition with freedom?
In what circumstances will users be willing to trade bandwidth for freedom?

The problem

In each of the major debates regarding internet governance (network neutrality, free speech, universal access, privacy, the "kill switch"), all of the voiced stakeholders argue from two shared assumptions regarding the future of the internet. They all imagine the internet to be always-on and as high-bandwidth as possible.

One result of these often unspoken assumptions is that owners of higher-bandwidth infrastructures such as wireline cable or "4g" wireless are afforded considerable attention while comparatively lower-bandwidth or non-persistent solutions remain either unexplored or dismissed. Further complicating this arrangement are two undesirable economic trends: increasing consolidation among owners of high-bandwidth infrastructures (e.g. recent announcement that AT&T intends to acquire T-Mobile) and vertical integration within the media production industries (e.g. anticipated merger of NBC/Universal with Comcast). These corporate actors cannot enter into good faith negotiations regarding the public good as the outcome may run counter to their core interests.

In addition, events of the last few years have revealed entrenched stakeholders in the high-bandwidth industries actively working against the interests of their users.

AT&T assisted the NSA in unlawful wire-tapping
ISPs cooperate with intellectual property extremist organizations and monitor user traffic
Comcast disrupts legitimate user traffic, lies about it
Comcast blocks public from attending public hearing regarding its bad behavior
ISPs cooperate with repressive governments and take down network during elections, demonstrations, etc.

Calling attention to the always-on, high-bandwidth assumption opens space for new voices and solutions in these debates that may de-center the entrenched market leaders.

There are (at least) two reasonable objections to considering lower-bandwidth solutions. The first is a prevailing vision among libertarian technologists of the internet as a network in which all traffic is treated equally and all nodes are equally accessible to all users. The second is a lack of technical detail in the discussion of internet bandwidth in practice - most evident in recurring depictions of the internet as a single, undifferentiated cloud rather than a complex network of networks.

http://en.wikipedia.org/wiki/File:Handbuch_Webdesign-InternetDatenaustausch.jpg

Outline

I. Description of the leading positions on the "net neutrality" debate in the U.S. with special attention to the persistence of the high-bandwidth assumption. How did the pluralistic vision of the internet as a "matrix" or "network of networks" give way to a homogeneous imaginary dominated by a handful of large-scale, high-bandwidth infrastructures?

II. Investigation of the term "bandwidth". What is the technical definition and implementation of internet "bandwidth"? How is it measured? How do users experience "speed" online? Is bandwidth an accurate measure of network speed? When are lower-bandwidth systems "faster" than higher-bandwidth systems?

III. Comparison of the technical meaning of "bandwidth" with the practical use of bandwidth among several categories of internet services (e.g. email, microblogging, discussion boards, transit maps, online news, streaming video). This part of the project will analyze the bandwidth required by each service to disaggregate data desired/requested by users (the message or signal) from unrequested data tacked on by the service provider (advertisements, links to other sites, unrelated images, data-mining scripts and cookies, or noise). I hypothesize that there is significant variation in the efficient use of bandwidth to transmit desired information both among and within service categories.

IV. A handful of highly-visible services (Twitter, YouTube, Facebook, Netflix, NYT) seem to provide the dominant examples used in internet governance debates. This fourth section will investigate the assumption that all services must be equally accessible to all users at all times. Are there local or hyper-local services that are of interest primarily to geographically proximate populations? If so, what are the costs and benefits of exposing these services to the entire internet? To what extent does the preservation of highly-centralized services such as Facebook foreclose the possibility for a more pluralistic internet?

V. Finally, this paper will return to the net neutrality debate and suggest new positions based on the information gathered above. One fear that persists among net neutrality advocates is that the internet will be split into "fast and slow" lanes. How might the experience of speed in "store and forward" systems such as Fidonet or the UUCP-based USENET inform the preservation and development of future internets?

I. The always-on, high-bandwidth assumption

Open with narrative account of Susan Crawford's presentation to SFC NYC in Feb 2011

Because wireline cable provides the most bandwidth, questions of regulation and internet governance should begin from the assumption that cable will be the dominant infrastructure for all network activities
Note that she is responding to dark fiber, e.g. Google experiment in the Bay area

Outline some positions in the debate around network management and neutrality:

Susan Crawford
Tim Wu
Save the Internet
Comcast

The current problem

Who owns the cable network today?
Revise Crawford's assertion: as long as we imagine the internet to be a service that is always-on at the highest-bandwidth possible, we are stuck with the wireline cable infrastructure

II. Bandwidth in discourse

Bandwidth, like processor speed and storage capacity, is a variable characteristic of the personal computing experience. For families and individual users, bandwidth is a feature of one's service provider. "Broadband" wireline service is "faster" than "dial-up" service. "3g" wireless service is "slower" than "4g" service. In promotional materials for commercial internet service, "bandwidth" is frequently discussed out of context from specific content or services. "Blazing fast speed!" does not refer to the time it will take your child to respond to an email. And when services are taken into account, "bandwidth" usually refers to the speed with which some desired data is delivered to the user; e.g. "YouTube videos load in a snap!"

Numerous factors contribute to the perceived speed of internet services. In addition to the bandwidth of your home internet connection, a YouTube video will load faster or slower depending on the available bandwidth at YouTube's data center as well as your local computer's capacity to decompress and playback the video. To some extent, these are accounted for in the term "latency".

For example, if I need to transmit 20kb of plain text (roughly the size of this document) to Google Docs, a bitrate of 5Mbit/s doesn't matter as much as the time it takes the Google servers to respond. Similarly, bandwidth is a two-way street. Cable and DSL services tend to have a large gap between "downstream" and "upstream" capacity. In other words, there is more bandwidth available to watch a video than there is to transmit your own.

When Comcast "throttled" BitTorrent on its network, most users were puzzled by their failed connections and slow speed. It wasn't until a journalist spoke to a team of network engineers that users discovered that their own provider was responsible for the service interruption. The users imagined that the internet service they paid for each month would available 24/7 with the maximum bandwidth possible. In contrast, their ISP took a more systemic view that a few users were taking up more than their fair share of network resources.

For many, news coverage of the Comcast situation revealed for the first time that cable users share bandwidth with their neighbors. Although it is obscured by cable TV, the technical reality of the wireline cable network is that it is essentially a high-bandwidth party-line. This revelation provokes some interesting questions about the deployment of cable internet services. Do all neighborhoods require the same bandwidth? Can neighborhood groups organize bandwidth use among themselves? If a given neighborhood requires or desires an unusual amount of bandwidth, can they collectively request it from Comcast?

Major ISPs seem reluctant to highlight bandwidth interdependence among neighbors on the cable network. Instead, subscriptions are organized by individual household. This arrangement leads to inefficiencies in dense areas such as apartment buildings where countless wi-fi networks overlap. Can we imagine a future in which bandwidth is organized on a neighborhood scale like trash pickup or street cleaning?

TODO The comcast example points to a larger problem of network plurality

III. Bandwidth in practice

In spite of the always-on assumption, typical bandwidth use is sporadic. When a user clicks a link in a webpage, for example, a request is sent to a computer elsewhere on the work. If that server is found, it will attempt to process and fulfill that request. Data is briefly transmitted between our "client" and the "server" and then the line is quiet while the user reads the email, looks at the picture, or watches the video.

Bandwidth is measured with "bitrate", the number of bits that can be transmitted per second. Today, wireline coaxial cable (of the kind used by Comcast and Time Warner Cable) provides for the highest possible bitrate of the available infrastructures. For network engineers, efficiency in such a network refers to maximizing the amount of useful information transmitted in the smalled amount of time by minimizing redundancy and correcting for errors introduced by electrical noise. The engineer's primary concern is that all of the data sent from the server is received by the client.

The user's experience of this transaction is quite different. Often, the server will send considerably more data than is requested by the user. This data might even include undesirable information such as advertisements or data-mining scripts. The user assesses efficiency according to a goal that is often external to the network:

How long do I have to wait before I can read this news article?
How long does it take for this video to buffer?
How much of my data plan is being used up to read this blog post?
How much does my online game "lag"?

Through a systematic analysis of internet services, I hope to evaluate a new measure for bandwidth that will take into account the qualitative needs of the internet user. Bitrate fails to account for the ratio of requested information to unrequested information. In my approach, we will take into consideration the efficient concentration of desired information.

Concentration can vary among the various interfaces to a single service (e.g. reading an article on the Sports Illustrated website vs. reading the same article with the SI iPad app), between services of a similar type (e.g. YouTube vs. Hulu), and among different categories of services (e.g. LA Times website versus World of Warcraft).

For example, receiving a tweet by SMS is a highly concentrated use of bandwidth. All of the data transmitted from the server to my handset is either required for the transmission or part of the information I (presumably) requested. The ratio in 7-byte characters is 160:160; there's nothing more included than what was requested. Receiving a tweet via Web interface, however, has a much lower concentration. Here, the ratio is 160:n+160 where n = all of the images, scripts, html, css, etc. required to render the page.

If key debates about internet freedom turn on questions of bandwidth, then we need more sophisticated means to measure and compare the bandwidth usage of various services. Comparing the concentration of desired information across services offers one new way to talk about internet use. Netflix may be a high-bandwidth service, but it is also a highly-concentrated service. The LA Times website may be a comparatively lower-bandwidth service but very little of the information transmitted to users is information that they specifically requested.

IV. Distribution of bandwidth among nodes

It is rare to see a website list hours of operation. The web is assumed to be always-on and always available. Servers don't shut down at the end of the work day, even if the majority of their users only visit during a certain window of time. This arrangement is logical for sites like Hotmail or Yahoo who have global audiences in many time zones. Is it necessary for the entire web? In which circumstances might nodes benefit from limited hours of operation?

When we talk about limiting access to internet services, it is often in the context of filtering, censorship, or intellectual-property restrictions. Users outside of the UK can only request a subset of BBC content. Users in China cannot make requests of Facebook's servers. Users who imagine the internet as an extra-geographic space lament this fracturing by location of origin. But it bears asking, are there cases in which restricting access to geographically-proximate users is desirable?

Bulletin-board systems (BBS) from the 1980s and 1990s provide a valuable example of rich network services that were regionally bound and intermittently available. BBSes were primarily accessible through the telephone network and often run out of the homes of hobbyists. In many cases, only one user could access the BBS at a time. Users developed habits of logging on for short periods of time once or twice per day to check for updates. Thousands of BBSes were available to interested users but distance systems would require a long-distance call. As a result, users tended to only access BBSes that were within a few dozen miles from their homes.

In the mid-1980s, Fidonet provided a means to interconnect BBSes. In the middle of the night when local users were unlikely to call, the Fidonet software would bundle up the day's messages and make a single long-distance call to another system. They two systems would exchange their bundles and hang up. This simple system enabled users of a local BBS to communicate with users on a distant BBS without having to pay for a long distance call.

Fidonet represented a global network of networks run out of the homes of hobbyists. Despite this global reach, however, each Fidonet Node retained its local character. The system was still located in a user's home. The administrator was still geographically proximate to his or her users. Presumably, some users made little use of the global Fidonet discussions, preferring instead to focus on local interactions with the fellow users of a particular node.

Fidonet also succeeds in resisting government shutdown ("kill switch") in a way that highly-centralized services do not. No one node in Fidonet knows all of the other nodes in the network. As a result, government repression would need to occur at the site of each individual node. As with most network compromises, the "kill switch" is a matter of social organization rather than technology.

For this section, I am drawing first on an interview I conducted last year with the founder of Fidonet. I'd like to flesh it out with at least one more interview and am trying to make contact with folks in the L.A. area who contributed to Santa Monica PEN, a local network for SM residents. Depending on how things go for the next week or two, this aspect of the paper might be pushed back until summer.

In the meantime, it might also be useful to recount the popular networks deployed in Egypt by hobbyists, activists, and amateur radio operators in the wake of ISPs being shutdown. There were 4 main providers of always-on, high-bandwidth internet service. Landlines were not shut off and radio operators can transmit independently of these networks.

V. Rethinking assumptions of bandwidth use

Sec. of State Clinton's public speeches on internet freedom respond to the shutdown of information services by repressive governments during demonstrations and elections. The recent events in Egypt, Tunisia, and elsewhere support the notion that communication infrastructures are important features of an active democracy. These examples do not, however, suggest that the network need be always-on nor particularly high-bandwidth. Facebook, Twitter, SMS, and MMS are all intermittent, low-bandwith network services.

One important feature of the recent uprisings is that rich accounts in text, still images, audio, and video were available beyond the local area. Always-on, high-bandwidth internet is not a requirement for this kind of transit to occur. In store-and-forward systems, local networks communicate with each other at appointed times. If a video is of sufficient value to an activist group, is it necessary that they be able to view it immediately? What is the maximum time it should take to transmit electronically? One hour? One day? One week? If the video is so valuable as to inspire demonstrators to risk life and limb, is one week too long to wait?

Always-on, high-bandwidth networks are wonderful technologies and one day we may all have access to one but they they are merely subnets of the internet. It is in the interest of entrenched stakeholders to assume that the only possible future of the internet is always-on and high-bandwidth. By highlighting this assumption and demanding more efficient bandwidth use from service providers, we may retain the benefits associated with network pluralism.

Resources

On internet freedom, governance, and policy

Video of Susan Crawford speaking to the Students for Free Culture conference, NYU, NYC, 2011.

In this short presentation, Crawford states that because of the high bandwidth capacity of wireline coaxial cable, we are stuck dealing with the cable companies
She goes on to describe various approaches to regulating cable
This is the event that sparked my thinking for this whole project - it sounds like this political approach places the nationwide cable network at the center of the internet and seems to sacrifice network plurality once and for all

Morozov, E. (2011). The Net Delusion. New York: Public Affairs.

Morozov's goal is the burst the bubble of techno-centric approaches to toppling repressive governments. He provides some valuable insight into the costs that are associated with highly-centralized internet services
Morozov questions a lot of assumptions built into the "internet freedom" debate but he doesn't investigate the questions of bandwidth or equal access that I hope to open up in this paper

Wu, T. (2011). The Master Switch. New York: Knopf.

This book is also useful for its last chapter in which Wu writes, "our appetite for bandwidth - the new black gold - is insatiable"
I'm wondering if there are other fossil fuel metaphors that might be applicable here. Is a low concentration site a "bandwidth hog" or a "data guzzler"?
From a policy perspective, Wu argues for a "Separation Principle" in which government regulates these industries such that network service providers and content producers are kept distinct, I feel that this point of view would be strengthened with a more nuanced approach to the use of "bandwidth" but I'm not sure how they fit together yet
I'm also relying on Tim Wu's articles on net neutrality: http://timwu.org/network_neutrality.html

Zittrain, J. (2007). The Generative Internet. Harvard Law Review, Vol. 119, No. 7, May, pp. 1974-2040.

Zittrain identifies a trade-off between security and freedom in internet technologies
He warns of a "divorce" in which there is a safe, surveilled walled garden and a free wild west
Zittrain pursues a middle path in which certain tech and social means are used to limit the effects of malware so that people will not retreat into apps and walled gardens
He provides some useful language for discussing network pluralism: The internet is both a single "logical network" and a network of "heterogeneous networks... function[ing] independently"

On network pluralism

Rheingold, H. (1991). Virtual Communities. Cambridge: MIT Pres.

In this paper, I'm using this book as a historical artifact
Rheingold uses "The Net" to describe a network of networks that is inclusive of Fidonet, BBSes, Minitel, and a variety of other services alongside the internet derived from ARPANET

On users' experience of "speed"

Eriksen, T. H. (2001). Tyranny of the Moment: Fast and Slow Time in the Information Age. London: Pluto Press.

Eriksen tackles the experience of time in a moment just before widespread availability of "broadband" internet into the home
This is a useful book for how it frames anxiety over internet "speed" and "efficiency" from the point of view of users rather than network engineers
A more recent article on this topic appears in the magazine Technology Review, "The Slow-motion internet". It's useful to compare the shift in expectations from 2001 to 2011 evidence in these two articles.

== Driving questions ==

Is the pursuit of bandwidth in competition with freedom?
In what circumstances will users be willing to trade bandwidth for freedom?

The problem

In each of the major debates regarding internet governance (network neutrality, free speech, universal access, privacy, the "kill switch"), all of the voiced stakeholders argue from two shared assumptions regarding the future of the internet. They all imagine the internet to be always-on and as high-bandwidth as possible.

One result of these often unspoken assumptions is that owners of higher-bandwidth infrastructures such as wireline cable or "4g" wireless are afforded considerable attention while comparatively lower-bandwidth or non-persistent solutions remain either unexplored or dismissed. Further complicating this arrangement are two undesirable economic trends: increasing consolidation among owners of high-bandwidth infrastructures (e.g. recent announcement that AT&T intends to acquire T-Mobile) and vertical integration within the media production industries (e.g. anticipated merger of NBC/Universal with Comcast). These corporate actors cannot enter into good faith negotiations regarding the public good as the outcome may run counter to their core interests.

In addition, events of the last few years have revealed entrenched stakeholders in the high-bandwidth industries actively working against the interests of their users.

AT&T assisted the NSA in unlawful wire-tapping
ISPs cooperate with intellectual property extremist organizations and monitor user traffic
Comcast disrupts legitimate user traffic, lies about it
Comcast blocks public from attending public hearing regarding its bad behavior
ISPs cooperate with repressive governments and take down network during elections, demonstrations, etc.

Calling attention to the always-on, high-bandwidth assumption opens space for new voices and solutions in these debates that may de-center the entrenched market leaders.

There are (at least) two reasonable objections to considering lower-bandwidth solutions. The first is a prevailing vision among libertarian technologists of the internet as a network in which all traffic is treated equally and all nodes are equally accessible to all users. The second is a lack of technical detail in the discussion of internet bandwidth in practice - most evident in recurring depictions of the internet as a single, undifferentiated cloud rather than a complex network of networks.

http://en.wikipedia.org/wiki/File:Handbuch_Webdesign-InternetDatenaustausch.jpg

Outline

I. Description of the leading positions on the "net neutrality" debate in the U.S. with special attention to the persistence of the high-bandwidth assumption. How did the pluralistic vision of the internet as a "matrix" or "network of networks" give way to a homogeneous imaginary dominated by a handful of large-scale, high-bandwidth infrastructures?

II. Investigation of the term "bandwidth". What is the technical definition and implementation of internet "bandwidth"? How is it measured? How do users experience "speed" online? Is bandwidth an accurate measure of network speed? When are lower-bandwidth systems "faster" than higher-bandwidth systems?

III. Comparison of the technical meaning of "bandwidth" with the practical use of bandwidth among several categories of internet services (e.g. email, microblogging, discussion boards, transit maps, online news, streaming video). This part of the project will analyze the bandwidth required by each service to disaggregate data desired/requested by users (the message or signal) from unrequested data tacked on by the service provider (advertisements, links to other sites, unrelated images, data-mining scripts and cookies, or noise). I hypothesize that there is significant variation in the efficient use of bandwidth to transmit desired information both among and within service categories.

IV. A handful of highly-visible services (Twitter, YouTube, Facebook, Netflix, NYT) seem to provide the dominant examples used in internet governance debates. This fourth section will investigate the assumption that all services must be equally accessible to all users at all times. Are there local or hyper-local services that are of interest primarily to geographically proximate populations? If so, what are the costs and benefits of exposing these services to the entire internet? To what extent does the preservation of highly-centralized services such as Facebook foreclose the possibility for a more pluralistic internet?

V. Finally, this paper will return to the net neutrality debate and suggest new positions based on the information gathered above. One fear that persists among net neutrality advocates is that the internet will be split into "fast and slow" lanes. How might the experience of speed in "store and forward" systems such as Fidonet or the UUCP-based USENET inform the preservation and development of future internets?

I. The always-on, high-bandwidth assumption

Open with narrative account of Susan Crawford's presentation to SFC NYC in Feb 2011

Because wireline cable provides the most bandwidth, questions of regulation and internet governance should begin from the assumption that cable will be the dominant infrastructure for all network activities
Note that she is responding to dark fiber, e.g. Google experiment in the Bay area

Outline some positions in the debate around network management and neutrality:

Susan Crawford
Tim Wu
Save the Internet
Comcast

The current problem

Who owns the cable network today?
Revise Crawford's assertion: as long as we imagine the internet to be a service that is always-on at the highest-bandwidth possible, we are stuck with the wireline cable infrastructure

II. Bandwidth in discourse

Bandwidth, like processor speed and storage capacity, is a variable characteristic of the personal computing experience. For families and individual users, bandwidth is a feature of one's service provider. "Broadband" wireline service is "faster" than "dial-up" service. "3g" wireless service is "slower" than "4g" service. In promotional materials for commercial internet service, "bandwidth" is frequently discussed out of context from specific content or services. "Blazing fast speed!" does not refer to the time it will take your child to respond to an email. And when services are taken into account, "bandwidth" usually refers to the speed with which some desired data is delivered to the user; e.g. "YouTube videos load in a snap!"

Numerous factors contribute to the perceived speed of internet services. In addition to the bandwidth of your home internet connection, a YouTube video will load faster or slower depending on the available bandwidth at YouTube's data center as well as your local computer's capacity to decompress and playback the video. To some extent, these are accounted for in the term "latency".

For example, if I need to transmit 20kb of plain text (roughly the size of this document) to Google Docs, a bitrate of 5Mbit/s doesn't matter as much as the time it takes the Google servers to respond. Similarly, bandwidth is a two-way street. Cable and DSL services tend to have a large gap between "downstream" and "upstream" capacity. In other words, there is more bandwidth available to watch a video than there is to transmit your own.

When Comcast "throttled" BitTorrent on its network, most users were puzzled by their failed connections and slow speed. It wasn't until a journalist spoke to a team of network engineers that users discovered that their own provider was responsible for the service interruption. The users imagined that the internet service they paid for each month would available 24/7 with the maximum bandwidth possible. In contrast, their ISP took a more systemic view that a few users were taking up more than their fair share of network resources.

For many, news coverage of the Comcast situation revealed for the first time that cable users share bandwidth with their neighbors. Although it is obscured by cable TV, the technical reality of the wireline cable network is that it is essentially a high-bandwidth party-line. This revelation provokes some interesting questions about the deployment of cable internet services. Do all neighborhoods require the same bandwidth? Can neighborhood groups organize bandwidth use among themselves? If a given neighborhood requires or desires an unusual amount of bandwidth, can they collectively request it from Comcast?

Major ISPs seem reluctant to highlight bandwidth interdependence among neighbors on the cable network. Instead, subscriptions are organized by individual household. This arrangement leads to inefficiencies in dense areas such as apartment buildings where countless wi-fi networks overlap. Can we imagine a future in which bandwidth is organized on a neighborhood scale like trash pickup or street cleaning?

TODO The comcast example points to a larger problem of network plurality

III. Bandwidth in practice

In spite of the always-on assumption, typical bandwidth use is sporadic. When a user clicks a link in a webpage, for example, a request is sent to a computer elsewhere on the work. If that server is found, it will attempt to process and fulfill that request. Data is briefly transmitted between our "client" and the "server" and then the line is quiet while the user reads the email, looks at the picture, or watches the video.

Bandwidth is measured with "bitrate", the number of bits that can be transmitted per second. Today, wireline coaxial cable (of the kind used by Comcast and Time Warner Cable) provides for the highest possible bitrate of the available infrastructures. For network engineers, efficiency in such a network refers to maximizing the amount of useful information transmitted in the smalled amount of time by minimizing redundancy and correcting for errors introduced by electrical noise. The engineer's primary concern is that all of the data sent from the server is received by the client.

The user's experience of this transaction is quite different. Often, the server will send considerably more data than is requested by the user. This data might even include undesirable information such as advertisements or data-mining scripts. The user assesses efficiency according to a goal that is often external to the network:

How long do I have to wait before I can read this news article?
How long does it take for this video to buffer?
How much of my data plan is being used up to read this blog post?
How much does my online game "lag"?