Regulating content on the Internet: A new technological perspective

B. Promoting Access to Content

So far, we have focused on how technology might be used to restrict access to content. We will now look at whether current technologies could be used to promote user access to, or awareness of, particular types of content. As with content restriction, any technological solution to content promotion relies on the ability to identify the content to be promoted. Apart from human inspection of content, the available approaches to content identification for the purposes of promoting content are the same as those available for restricting content: matching the content against a pre-identified list of IP Addresses, Domain Names, Character Strings, Protocols, or Digital Signatures. Any attempts to use these content identification methods to promote specific types of content would be subject to the challenges, weaknesses, evasion techniques and countermeasures described above in connection with content blocking and filtering, and would be subject to the caveats outlined in Section C on The Limits of Technology, below.

What is Being Done Today?

Technology is being used in limited ways to promote content on the Internet today, primarily as part of a business strategy to attract and hold user attention. While, as we have seen, a number of countries are attempting to use technology to restrict access to specific content, we are currently not aware of any states that have implemented technological measures to try to promote specific types of content. The following sections will provide examples of how technology is being used by individuals, businesses, and other organizations for content promotion. None of these approaches will guarantee that users are exposed to any specific content, but they may increase the likelihood.

Search Engine Optimization; Sponsored Links and Ad Words

Keywords and geo-location are among the primary mechanisms used by search engines to provide relevant results for search terms. Savvy Web developers give careful consideration to Web site text, including text in alt tags (text which describes graphics, audio and video for users who cannot view these media types), in order to increase the likelihood that their sites will appear in search results. Content providers can leverage search engine companies' efforts to provide relevant results to users by optimizing their sites to increase their rankings when users search on specific terms.

Purchases of targeted "ad words" and "sponsored links" use a combination of keyword matching and IP location detection to raise the profile of sites of interest in a particular region, when a user searches on specified keywords. So, for example, a search on google.ca for "automobile repair" conducted by a user with an IP address in Ottawa, will not only deliver search results prioritized by keyword relevance, but will place ads for Ottawa auto repair shops in a sidebar on the results page. The costs of purchasing "sponsored links" or "ad words" are determined by market forces (buyers bid to secure the keywords they desire) and can vary greatly, ranging from pennies to $80 or more per click depending on what the keywords are and who is competing for them.

Local Content Servers

Large media, software, and retail companies frequently use local content hosting and media streaming servers to improve download times, thereby making their content more appealing and accessible to users. For a fee, companies such as Akamai will host mirror sites for these companies on proxy servers, and have systems that will direct the user's request for content to the nearest server in order to minimize download times (unless the nearest server is very busy, in which case the request will be redirected to the server which will provided the fastest download). The cost of hosting and delivering content from mirror sites is generally based on the amount of data downloaded by users of the site in a month.

Social Media Marketing

Many organizations and individuals are also furthering their marketing and communications goals by making use of social networking sites such as Facebook, YouTube, Twitter, and Second Life. Social media marketing campaigns often focus on building brand awareness and visibility, and encouraging feedback, participation, and a sense of ownership. Politicians are beginning to make very successful use of social media networking sites. In the current US Democratic Presidential Primary, for example, both of the two final candidates have extensive articles on Wikipedia; both candidates (and/or their supporters) have posted videos on YouTube; and both have official Facebook pages. At the time of writing, a YouTube search on "Barack Obama" returned 94,200 videos, the vast majority posted by Obama supporters, although 928 were posted by BarackObamaDotCom, the Web site of the Barack Obama campaign. (Obama appears to have been more successful in using YouTube than Clinton. Not only were there more than ten times as many "Barack Obama" videos as "Hillary Clinton" videos, but almost all of the former seem to have been posted by Obama supporters, while a large percentage of the latter seem also to have been posted by Obama supporters.) Similarly, both Obama and Clinton have Facebook pages that allow users to click a link to "become a supporter." The Obama's page shows him having 782,737 Facebook supporters, while Clinton's shows her having 148,421 supporters.

 
Screenshots of the Facebook sites for Barack Obama and Hillary Rodham Clinton

In much the same way, the entertainment industry is using social media sites to promote their properties. The screenshot to the left shows one of the many trailers and previews for the motion picture Juno available on YouTube. As of the date of writing, this trailer had been viewed 2,724,759 times, and the posting had solicited 4,830 text comments and 20 video responses from users. Juno also has a Facebook page, and over 52,000 Facebook users had clicked the link on that page to "become a fan" as of the date of this writing.

When a user clicks the link to "become a supporter" on a politician's Facebook page or to "become a fan" on a movie's Facebook page, a story appears in the mini-feed for that user's own Facebook page, serving as an automated form of viral marketing to that user's Facebook friends.

Screenshots of Facebook pages for the Canadian movie Juno (L) and Facebook page of user who has clicked the "Become a Fan" link for Juno (R). A "mini-feed story" announcing that the user has become a fan of Juno appears on her Facebook page, and will be sent to all of the user's friends who have subscribed to her mini-feed.

Content Insertion

It is also possible for a local ISP to insert content into Web pages being downloaded by its users. While technically possible, such efforts have been met with huge resistance and have many non-technical implications. In December 2007, Rogers Cablesystems experimented with inserting its own content, such as customer service notifications, into the Google homepage.⁶¹ This resulted in an immediate outcry from users, who charged that Rogers was violating the principle of net neutrality. (While not a technical issue, content insertion also raises questions of potential copyright infringement and violation of author's moral rights. A full discussion of these non-technological issues, however, is outside the scope of this report.)

Screenshot of Google homepage with notification at top inserted by Rogers

How Might these Technologies be Used to Promote Canadian Content?

In looking at how the technological approaches mentioned above might be used to promote content, we will focus on the type of content that the CRTC has historically been concerned with promoting: that is, high-production-value audio and video content intended for television or radio distribution. This content has traditionally been "certified" as Canadian by the Canadian Audio-Visual Certification Office (CAVCO), which processes applications for certification and determines whether a production meets the criteria to qualify as a "Canadian" production. For the purposes of this discussion, we will ignore the growing body of user-generated content posted by Canadians on social media sites.⁶²

It would be possible to extend the CAVCO certification process to online productions, allowing the producers to submit the IP addresses, domain names, or URLs for specific Web content, in order to have it certified as Canadian. With a registry of certified Canadian online content, it would be possible for the Canadian government to consider a variety of programs designed to increase the visibility and accessibility of this content for Canadian Internet users.

Agreements with Search Engine Providers

Through negotiated agreements, it may be possible for the Canadian government to increase Canadians' awareness of Canadian content by contracting with search engine providers to use a registry of certified Canadian online content to display sponsored links or ad words in the results for relevant searches. A program to certify online Canadian content and to promote that content through sponsored links or ad words would fit with the business model of the search engine providers, and would be consistent with existing practices and user expectations. This could be a powerful tool for bringing relevant content to the attention of users who are most likely to be interested in it. However, it is difficult to imagine exactly what keywords would be most appropriate to trigger the sponsored links or ad words for Canadian content. It is unlikely, for example, that many people will actually search for terms such as "Canadian movie" or "Canadian music" — and if they did so, it is likely that sites for Canadian movies or Canadian music would be already prominent in the natural search results. Furthermore, as noted elsewhere in this report, search engines already use geo-location technologies to respond to search queries with locally relevant content.

It might be possible to purchase ad words or sponsored links for keywords for content that is "similar to" specific Canadian content in some way and would appeal to the same audience. (For example, if the Canadian TV series "Da Vinci's Inquest" arguably appeals to the same audience as the American series "CSI", it might make sense to promote "Da Vinci" by purchasing ad words or sponsored links that will appear when users search for "CSI"). However, this kind of audience matching would be time consuming and labour intensive, with significant cost implications.

While it is beyond the scope of this report to analyse the costs and complexities of implementing a program for certifying Canadian Internet content, it is clear that the volume of content eligible to be considered could vastly exceed the content currently eligible for certification under the CAVCO provisions. (One need only consider the volume of Internet content as it compares to traditional media content to conclude that the costs of a program for certifying Canadian Internet content would be significantly higher than the current program for traditional media.) We should also note that there would be obvious cost implications if the Canadian government entered into a commercial relation with search providers. Any estimate of these costs at this point would be entirely speculative since, as noted above, costs vary wildly depending both on the popularity of the keyword and on the number of clicks it receives. (An interesting side note, in this regard, is that "success" in the use of sponsored links or ad words is generally measured in the "click through rate" (CTR), that is, the number of "click throughs" an ad word receives.⁶³ However, "success" can be costly, as sponsors only pay for their ad words or sponsored links when they are clicked. Sponsors need to carefully consider the potential average return on investment (ROI) for each click, and be confident that the ROI will exceed the cost of the paying for the ad words or sponsored links.)

While we are not aware of any state that is using search engine results to promote specific content, it would be theoretically possible for the Canadian government to pass a law requiring search engines to display "certified Canadian content" as sponsored links. In the area of restricting search results, various governments, including not only China, but a number of other countries around the world, impose legal requirements on search engine companies to remove particular search results in the localized versions of their search engines. For example, Germany requires removal of results for neo-Nazi sites. It is beyond the scope of this report to assess on what basis the government could establish a policy requiring search engine companies to display sponsored links for designated Canadian content. It is worth noting, however, that the most popular search engine companies are not based in Canada, and it would be up to the search engine companies to decide whether from a business perspective would customize their localized search engines (i.e., the ".ca" versions) in this way or abandon the market. While markets such as China and Germany are large enough to justify the expense of customizing the search results for those countries, it is an open question whether Canada is a large enough market to induce search engine companies to introduce features not in use elsewhere in the world. (And, presumably, up to the law courts to determine whether this violated any constitutional freedom of speech provisions.) The introduction of such a law also would not obviate the problems, discussed above, of determining which keywords should trigger the display of specific "certified Canadian content". Furthermore, if Canadians found that the ".ca" versions of search engines were providing results that were not useful or relevant to their search terms, they could always choose to use the ".com" version of the search engine instead.

Local Content Servers

Making use of services that stream high quality audio and video content from local servers in order to reduce the download times and enhance the user experience is another way of promoting high production-value content to Canadian Internet users. The Canadian government could consider a program to acquire space on local servers for certified Canadian content from services provided by companies such as Akamai. There would be of course costs associated with such a program. As mentioned above, the costs of hosting content on local caching servers provided by these companies are a function of the amount of data transferred. Accordingly, it is impossible for us to estimate the cost of contracting for such a service for certified Canadian content without knowing the amount of content that would be hosted on the server and having an estimate of the demand for that content. However, a rough rule of thumb is that these services currently charge approximately $1,000 per terabyte of data transferred each month.⁶⁴ To give a sense of how much data this is: one terabyte of data would be the equivalent of a 90 minute movie streamed to 5,000 users or a 5 minute audio clip streamed to 100,000 users.⁶⁵ (If the same 90 minute movie were viewed by 100,000 users, it would involve the transfer of 20 terabytes of data.)

Social Media Marketing; Search Engine Optimization

Canadian content producers need to understand how to take advantage of the opportunities provided by social media marketing and how to optimize their sites to improve their rankings in search engine results. The Canadian government could develop a program to increase producers' awareness and understanding of these opportunities. Learning to harness the power of online social marketing as well as to optimize online search results are perhaps the most significant ways in which technology can be used to promote Canadians' awareness of specific kinds of Internet content. The key to the success of these techniques is that they attract users' attention to particular kinds of content implicitly during in the course of their normal Web use, rather than intervening in ways that force or expect users to alter their preferred patterns of Webbased behaviour. For example, discovering that one's Facebook friends are interested in a particular Canadian TV drama series would be more effective in raising user awareness than potentially annoying the user by inserting an ad for that TV series prominently on his or her google.ca homepage, or promoting the series on a special Canadian content site that the user never visits.

C. The Limits of Technology

Up to this point, we have examined the current approaches to content identification, and how these approaches can be used to restrict access to content on the Internet. We have also explored the weaknesses and challenges associated with each of these approaches. And we have looked at how technology is being used to promote content on the Internet, and how the Canadian government might leverage these approaches to promote Canadian content. In this section we will discuss what technology cannot do today, and hope to address some of the common misconceptions about what technology is currently capable of.

Perhaps the most important thing to understand about all current technological approaches to identifying and controlling data transiting the Internet today is that they all rely on comparing data against pre-identified criteria. These criteria, for example, include lists of IP addresses or domain names of Web sites, strings of bytes in URLs or data files, protocols, waveforms of sound files, or patterns of network traffic. The researchers found no technology currently deployed or likely to be available in the near future that can intelligently determine the semantic content of a piece of data it has not encountered before. Targeted tests on the Internet called "captchas" which most humans can solve easily but current computers cannot, take advantage of this technological limitation. The captchas prevalent on Web sites such as Hotmail or Yahoo are used to prevent automated systems called "bots" from creating thousands of fake email accounts from which to send spam. Initially, since bots could only interpret alphanumeric data, these sites interposed a graphic displaying a distorted word. A user wishing to create a new email account was asked to type the displayed word into a text field. Since bots cannot read (they simply recognize the bits representing text characters), the sites knew that only a human could enter the text displayed in the graphic. However, for many years image scanning software has been capable of translating pictures of text into computer text. Using a variation on this software, the bot developers were able to defeat the first wave of captchas. The captcha creators responded first by distorting the text even further. However, this created a problem for human users of these systems. Humans could not read the extremely distorted words. Captcha creators then began to combine the captcha graphics with a question that the user would have to understand and respond to correctly. So for example a graphic showing three words might be displayed, along with the question: What is the second word in the captcha phrase "seva ada ibi igis ayese"?⁶⁶ Or a captcha might display a simple arithmetic expression with the question "What is the answer to the arithmetic question in the captcha?" Unless a bot has been programmed to expect the specific question and to respond to it correctly, it will fail the test, thereby confirming it is most likely not a human but a machine.

Screenshot of captcha from the Web site locr.com. The captcha prevents the creation of user accounts by "bots" by asking a question about the information presented on a map. The user must understand the question, interpret the graphic, and respond correctly in order to create an account.

A Game of Whack-a-Mole

In discussing the limits of today's technology, it is useful to look at the areas in which society and the technology community have expended the greatest efforts to detect and restrict access to content: child pornography, email spam, and malicious software programs (viruses, worms, and denial of service attacks). Despite a sincere desire, commitment on the part of all involved, and the expenditure of millions of dollars, the sad truth is that the efforts to stem the tide of these types of content have had limited success.

With respect to child pornography, the best method of restricting access that has been found to date is to block the IP addresses of known child pornography sites, an approach that has been singularly unsuccessful in restricting access because as soon as the pornographers know that their Web site is on a block list, they simply change the IP address.⁶⁷

Email spam is the bane of most email users, and hundreds of millions of dollars have been spent developing and deploying spam filtering technologies. Spam filters have become better and better, over the past ten years, and they now filter out hundreds of spam emails every day from the email of most Canadian email users. But with every advance in spam filtering technology, the spammers devise new evasion techniques to which the filtering companies must adapt. If IP blocking is, as we were told by Telus CIO Kevin Salvadori, "a constant game of whack-a-mole," attempts to filter out email spam is similarly a never-ending game, with regularly updated spam filtering software attempting to eliminate spam, and the spammers devising new evasion techniques which pop up even before the new filters have been released.

If email spam is the bane of email users, computer viruses, worms, and malware are feared by network administrators and computer users, alike. Virus detection software that sits on the user's computer, and devices that perform so-called deep packet inspection that sit on the ISP network, attempt to detect and block these malicious software programs. Experience and improved algorithms have improved the blocking ability of detection software and devices; however, rather like the flu vaccine, the detection and blocking software must be updated regularly to deal with the latest attack. And just like the flu virus, as soon as these malicious programs encounter resistance, they rapidly mutate to avoid detection. Indeed, just as with spam, the perpetrators have often devised new ways of purveying their malicious software, ready to be released as soon as they know the previous attack has been blocked.

Deep Packet Inspection: How Deep is Deep?

There are a lot of misconceptions about the meaning of the term "deep packet inspection" (DPI). Unlike "shallow packet inspection," that only parses the headers of IP packets and compares the results against a rule set defined by the system administrator, "DPI engines parse the entire IP packet, and make forwarding decisions by means of a rule-based logic that is based upon signature or regular expression matching. That is, they compare the data within a packet payload to a database of predefined attack signatures (a string of bytes). Additionally, statistical or historical algorithms may supplement static pattern matching."⁶⁸ The primary use of DPI devices on ISP networks has been to manage traffic and prevent attacks by viruses, worms, and denial of service attacks.

A number of vendors (e.g., Narus, Ellayoca, Sandvine, and others) sell systems designed to detect patterns of behaviour in network traffic or the digital signature of specific types of files transiting the network either to protect network integrity (against, for example, viruses or denial of service attacks) or to manage and regulate Internet traffic (by, for example, limiting the volume of peer-to-peer traffic). Narus Inc. informs us that one of the world's largest carriers uses its devices as part of a system designed to detect and identify "anomalous IP traffic" (e.g., viruses, worms, and denial of service attacks). When anomalous traffic is detected, a copy is streamed to other applications for processing and a decision as to what action should be taken. For the purposes of maintaining network health, automated analysis by inspection of packet headers can usually be performed quickly and action initiated in time to prevent damage to the network.

Virus detection is done inside the email systems of most ISPs, corporations and even on a large percentage of client systems. Any binary file or script is scanned and compared to a database identifying characteristics of known malicious or undesirable files. If a file contains a matching signature, it is deleted or repaired.

In an excellent paper entitled The Perils of Deep Packet Inspection that discusses DPI in the context of the evolution of firewall technology, Thomas Porter compares header inspection (shallow packet analysis) with DPI:

He goes on to say:

Porter's reference to detection of data that fits within a single packet is an important one. Devices that perform DPI are subject to two important limitations:

1. they can only detect pre-defined strings of bytes or pre-defined patterns; and
2. they can only detect content within a single packet.

As discussed earlier in this report, since the content of any data file will be divided arbitrarily into packets as it transits the IP network, pre-defined strings may be split across packets and, therefore, not found by DPI devices.

Porter goes on to describe DPI technology as complex and immature, leaving networks open to exploitation. Although his paper was written in 2005, and devices have certainly improved since then, the recent Internet Evolution study of carrier-grade peer-to-peer filtering devices discussed earlier in this report⁷¹ indicates that, for most devices, even the relatively straightforward task of detecting protocols remains a challenge.

Aside from the cost and network performance issues, discussed above, data encryption is one of the biggest challenges to DPI. Although the Internet Evolution study cited earlier found that the best P2P filtering devices were able to detect the protocol in peer-to-peer traffic generated by most common file sharing applications, even when the protocols had been encrypted, this does not mean that they were decrypting the packets and reading the content. P2P falls into a category of traffic (such as denial of service attacks) that are unusual, in that the traffic often exhibits unique patterns of behaviour, and these patterns can be detected even when the DPI device cannot actually read the bits in the data packet. What these devices are doing is detecting traffic behaviour patterns. The contents of the packets are not being read. For P2P traffic the DPI device does not need to have any knowledge of the contents of the packet in order to identify the protocol being used. This is not, however, the case if one wishes to identify byte strings in ordinary content files (text, graphics, audio, video) being transferred over the Internet. As the President of Narus advised us, encryption would defeat any attempt to identify the contents of a data packet using DPI.

In the context of attempting to block the exchange of copyright protected content, another interesting development is a new peer-to-peer file sharing technology known as the Owner Free Filing system (OFF). Unlike earlier forms of P2P, the OFF system does not involve entire files being stored on one computer and transferred directly from one user's computer to that of another. Instead, random bits of data are stored on various computers throughout the system in 128 KB blocks, and can used as components to reconstruct requested files. Because no-one is actually storing "the file" and because the blocks of data can actually be used as components of numerous files, it is impossible to say that any single block belongs to a specific copyrighted piece of content.

The Wikipedia article on OFF states:

With an OFF system, a packet would contain only one of these random data blocks, making any attempt to match it against pre-defined criteria meaningless, and making it essentially impossible for a DPI system to identify any semantic content within a packet.

The Limits to Traffic Shaping

Because of news stories about ISPs using network devices to throttle peer-to-peer (P2P) traffic, the question has arisen as to whether it would be possible for ISPs to improve the quality of service (QoS) or speed up the delivery of "certified Canadian content" over the Internet.

QoS and speed are essentially the same thing. It all amounts to traffic shaping. There really is no such thing as a "fast lane" on the Internet. There is simply, more or less bandwidth. The more bandwidth available, the less time it takes for traffic to transit the network. A good analogy is a physical highway. Suppose there are three lanes on the highway. One is a carpool lane, and can only be used by vehicles with at least three passengers. If there aren't very many cars on the road, all the traffic gets through at the speed limit. No-one goes faster than anyone else. As traffic increases, the cars in the carpool lane may still be able to go the speed limit, while the cars in the regular lanes have to slow down because of congestion. At really busy times, however, even the traffic in the carpool lane may slow to a crawl.

The network devices currently being used by ISPs are capable of throttling or prioritizing traffic based on protocol. To carry the analogy one step further, the devices can distinguish between cars with three passengers and cars with less than three passengers. They do not distinguish, however, based on who is in the car. Hence, P2P traffic (i.e., traffic that uses the ports normally used by P2P) can be restricted to a certain queue (or portion of the bandwidth). So, for example, it is not BitTorrent, specifically, that's being throttled, or P2P content coming from particular providers, but all traffic using the P2P protocols. Conversely, traffic that uses certain other protocols (such as the ones that are used by mainframe computers or the ones that are used by videoconferencing or VOIP) can be put in other queues that have more bandwidth available to them. This is not prioritizing specific content, but specific protocols. The traffic isn't really even being prioritized: a portion of the bandwidth is "reserved" for certain queues based on protocol.

This all relates to the way ISPs have traditionally engineered their networks, which is based on "leaky bucket" provisioning. That is to say, ISPs have tried to build their networks so that, at peak, they will only be at 75% capacity. The remaining 25% is then available for them to ensure sufficient bandwidth for traffic that is highly time sensitive. In this regard, it is important to realize that any rules that allocate a larger proportion of the bandwidth to specific traffic, by necessary corollary, reduce the bandwidth for all other traffic. Returning to our physical highway analogy, a carpool lane is reserved for vehicles with three or more passengers, even if there are very few of these vehicles. Meanwhile, all of the other vehicles must share the remaining lanes, even if this causes a traffic jam. By reserving bandwidth for certain protocols, ISPs would be, by default, providing preferential treatment to certain types of traffic at the expense of other traffic.

Network devices must be programmed directly to implement traffic management rules, and they are not designed to refer to external sources (e.g., updatable lists) to find the rule sets. The rule sets also tend to be quite limited. Altering these systems to interact with a list of certified content or to recognize a watermark and move the content to a different queue would be a non-trivial engineering task. Also, each device manufacturer implements QoS or traffic shaping differently, so to implement a consistent nation-wide system would mean either to do this engineering for each of the systems used by the different ISPs or to require all Canadian ISPs to reconfigure their networks using identical network devices. (For QoS to be effective, the traffic must be prioritized at every point along the line.)

Finally, the trend is for ISPs, including the telcos, to move to ethernet networks, which are not designed for QoS.

For all of these reasons, the notion of somehow prioritizing certified Canadian content (if that content is using the same protocols as other non-certified content), is simply not feasible at this time. The best way of improving download times is to mirror the content on local caching servers. When the user requests that content from the content provider, geo-location redirects the request to the closest server, ensuring the minimum network distance between the user and the content, and thus the fastest possible download speeds.

---

⁶¹ http://blog.wired.com/27bstroke6/2007/12/canadian-isps-p.html

⁶² Although it would be possible for creators of user-generated content to apply to have their content certified as Canadian in order to take advantage of any programs put in place to promote Canadian content, it is unlikely that many would be willing to undertake the time and expense involved in doing so.

⁶³ I.e., the number of users who find the link interesting and relevant enough to induce them to click on it.

⁶⁴ Base rates can be as much as $2,000 per terabyte, but prices go down with volume.

⁶⁵ Assume a bit rate of 300 kilobits per second (a common transfer speed).

⁶⁶ Captcha from the contact page for the Open Web Vancouver site http://www.openwebvancouver.ca/contact.

⁶⁷ See discussion, above.

⁶⁸ Thomas Porter, The Perils of Deep Packet Inspection, 2005
http://www.securityfocus.com/infocus/1817.

⁶⁹ Ibid.

⁷⁰ Ibid.

⁷¹ Internet Evolution, "Peer-to-Peer Filtering: Ready for Internet Prime Time? ", op. cit.

⁷² http://en.wikipedia.org/wiki/Owner_free_filing_system.