Toward Efficiencies in Canadian Internet Traffic Exchange
Toward Efficiencies in Canadian Internet Traffic Exchange
Toward Efficiencies in Canadian Internet Traffic Exchange
By increasing the richness and density of connections between Canadian networks, additional IXPs will increase the reliability of Internet access in Canada and its resilience to disaster and attack.
The gure above shows alternative paths to connect two Canadian Internet customers. The top option shows a short, efcient, direct path via a Canadian Internet exchange point, whereas the bottom path detours via a US ISP and Internet exchange point. Relative to comparable international peers, Canada is well behind in its provision of IXPs. Canada currently has two operational IXPs (in Toronto and Ottawa) and three in planning approximately one operational IXP per 17 million people, three to thirty times fewer than other similarly developed nations. A void of Canadian Internet strategy and policy has resulted in Canadian Internet users inheriting U.S. policy, costs, and aws rather than enjoying an environment deliberately crafted to Canadian benet. The provision of IXPs is not automatic: a network, or group of networks, must step up and take the lead in addressing the physical, managerial, and technical requirements. Moreover, the mere presence of an IXP is not sufcient to improve conditions; ISPs must actually make the effort to use them. Indeed, as many Canadian networks peer in London, UK, as in the existing exchange in Ottawa. Fortunately, IXPs typically cost less than $100,000 to establish, and return on investment can be seen in as little as a few days. This document proceeds as follows: In section 2, we present the relevant technical underpinnings and resulting incentives. In section 3, we explore the benets of increasing the number of IXPs in Canada. In section 4, we offer recommendations for the number, location, and structure of IXPs in Canada. In section 5, we ag possible challenges and offer specic recommendations.
Page 2
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
Page 3
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
Page 4
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
We have repeatedly observed data routed from one point in Canada to another point in Canada via an IXP in the United States. This occurrence is often known as tromboning because, like wind through the slide of a trombone, data ows to a distant intermediary, only to return to a destination near its origin. For an example, see the traceroute in Exhibit 3. Numerous, widespread Canadian IXPs could sharply reduce the prevalence of this wasteful practice. To conrm the scope of Canadian data routed via other countries, we analyzed traceroutes collected by the University of Toronto IXmaps Project and the University of Washington iPlane project from 2010 through the present.34 We identied 24,705 traceroutes between pairs of end users both located in Canada. Of these, 60% stayed entirely within Canada, whereas the remaining 40% were routed through one or more other countries. Most common by far was routing via the United States: fully 38% of traceroutes between two points in Canada were routed through the United States (occasionally in addition to at least one other country). It is difcult to conrm the degree to which the University of Toronto dataset is representative of Canadian Internet trafc as a whole, but these data nonetheless conrm that it is not unusual for Canadian trafc to be routed via the United States. Avoiding unnecessary round trips to and from other countries would help Canadian networks enjoy lower costs. If network operators are able to operate at lower costs, a portion of the savings will be retained by these networks as increased prots, and a portion will ow to consumers through lower prices and higher service quality. The allocation of savings depends on the relative elasticity of supply of Internet access and demand for Internet access. Where consumer Internet access is competitive and margins are low, most of the benets ow through to consumers. Avoiding unnecessary international detours would also greatly increase the performance of Canadian domestic trafc. The median round-trip time for the paths in our dataset that remained within Canada was 33.37ms, while that of the paths that passed outside of Canada was 48.21ms. Thus we can see that even with only two IXPs in Canada, a performance increase of more than 30% has already been achieved, and that gure can only increase with the construction of more IXPs and shorter paths.
Page 5
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
Improving reliability
By increasing connections between Canadian networks, additional IXPs will increase the reliability and resilience of Internet access in Canada. Without IXPs, networks typically nd it cost-effective to rely on just one or two links to key transit providers. In this sparsely interconnected network architecture, if one of a networks links fails, the network often suffers severely degraded performance. In contrast, a robust web of interconnected networks, with substantial local peering at IXPs, gives networks many ways to reach each other. Thus, if one link fails whether through operator error, equipment malfunction, maintenance, sabotage, or worse communications continue across the many remaining links unimpeded, and performance is not degraded. IXPs make it easy and costeffective for a network to have substantial excess capacity capacity that is easily and automatically invoked as needed.
Page 6
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
Recent cyber-attacks conrm the benets of IXPs in improving network reliability. During 2007 cyber-attacks, Estonia beneted from two domestic IXPs as well as transit and peering in diverse allied nations. During similar attacks in 2008, Georgia lacked those benets. Estonia weathered the attack easily, whereas the entire government of Georgia was forced to migrate onto Googles free hosted services until the attack subsided, months later.
4. Recommended approaches
Number and general location of IXPs needed
Canada currently has two operational IXPs, in Toronto and Ottawa, and three more in planning in Winnipeg, Montreal and Calgary approximately one operational IXP per 17 million people. By comparison, the United States has eighty-ve (one per 4 million), England twelve, Australia eleven, and New Zealand seven (one per 5 million, 2 million, and 0.5 million, respectively). See Exhibit 1. This comparison reveals Canadas laggard status in the provision of domestic IXPs. Based on experience in other countries, we believe that IXPs should be viable in most Canadian urban areas. An IXP is more likely to thrive where the population has sufcient income to support substantial Internet use. An IXP is more likely to thrive the further it is located from existing large IXPs (which serve as partial, albeit imperfect, substitutes). An IXP should be viable anywhere where three or more networks are willing to peer with each other. Exhibit 2 summarizes our preliminary review of prospects for IXPs in Canada, including recommended short-term and medium-term priorities for IXPs in key cities and regions. IXP locations were evaluated based on factors including population density, per capita income, and Internet usage. Because networks savings are typically proportional to the amount of data that can be exchanged at a given IXP, the benet will be greatest at IXPs in the largest Canadian cities. With Ottawa and Toronto already enjoying IXPs and Winnipeg and Calgary IXPs in planning as of fall 2012, it is natural to proceed to Edmonton, Halifax, Montral, Qubec City, Vancouver, and Windsor the proposed IXP locations detailed in the center panel of Exhibit 2. However, IXPs in all the listed cities and regions both short- and medium-term are likely to be cost-effective, so if local interest quickly arises in an IXP for a smaller city, the local support probably justies prioritizing effort there. We believe it is realistic for Canada to seek to establish seven to ten IXPs in total (ve to eight new IXPs) in the short run (within the next 18 months). Within four years, it is realistic for Canada to host twenty-ve IXPs, which would yield one IXP per 1.3 million people between the rates now seen in Australia and New Zealand.
Page 7
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
dominant network. In either case, networks hesitate to invest in connecting to a location in which they have limited condence.
Page 8
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
networks doing business in a given region, the networks usually realize that the greatest efciencies and benets are gained by keeping costs as low as possible. A networks biggest expense in participating at an IXP is the telecommunications infrastructure connecting the IXP and the networks users. A suitable choice of IXP location can reduce this cost. Usually, the best location turns out to be in the center of a dense urban environment where space is more expensive, but where the IXP is closer to the users who depend upon it (reducing the average cost to connect those users to the IXP). It is also desirable for an IXP to be served by competitive telecommunications providers because a competitive telecommunications marketplace will reduce the largest expense of IXP members.
Page 9
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
Page 10
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
operations. Attempts to nd suitable locations through open discussion are sometimes stymied by the awkwardness of the give-and-take often, a combination of networks wanting to appear generous yet also wanting to serve their self-interest. PCH has found that an expedient approach is voting through prioritized lists each network prepares separately. Each network is instructed to prepare an ordered list of the locations where it would be willing to connect to an IXP, from most preferred to least preferred. Usually, each networks single most-preferred location is a location of its own, of limited or no interest to others. But many networks often share several highly ranked locations. The stylized example below illustrates typical preferences. PCH has successfully used this method to select IXP locations more than one hundred times. Usually PCH styles the prioritized list as a homework assignment to be prepared by network staff separate and apart from other networks, a process that seems to yield thoughtful non-self-serving responses.
ISP As choices First choice, weighting multiplier: 4 Second choice, weighting multiplier: 3 Third choice, weighting multiplier: 2 Fourth choice, weighting multiplier: 1 ISP Bs choices ISP Cs choices ISP Cs point of presence 50 Central Avenue 1 Main Street 10 Broad Street ISP Ds choices ISP Ds datacenter 1 Main Street 50 Central Avenue 10 Broad Street
ISP As headquarters ISP Bs headquarters ISP Ds datacenter 10 Broad Street 50 Central Avenue 1 Main Street 50 Central Avenue 10 Broad Street
When the weighted votes are tabulated, we get the following results:
Weighted Votes 1 Main Street 50 Central Avenue ISP Ds datacenter 10 Broad Street ISP As headquarters ISP Bs headquarters ISP Cs point of presence 8 8 7 5 4 4 4
In this case, 1 Main Street and 50 Central Avenue are broadly acceptable, even though these locations were not the rst choice of any of the participating networks. ISP D's datacenter is a close runner-up because it is also used by ISP A. Following PCH's usual procedure, the workinggroup of participating networks would send representatives to the owners of the buildings at 1 Main Street and 50 Central Avenue. In discussions with those building-owners, the workinggroup representatives would explain the nature and requirements of an IXP, seeking proposals for space, power, and other resources the buildings would offer to obtain the IXP's tenancy. With two or more proposals in hand, the working-group would choose a location balancing the preferences of participating networks with the costs and other terms from building owners.
Page 11
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
6. Exhibits
Exhibit 1: IXPs by country (selected countries)
Country Argentina Australia Austria Belgium Brazil Canada China France Germany India Indonesia Italy Japan Kenya Mexico Nepal Netherlands New Zealand Norway Poland Russia Saudi Arabia South Korea Spain Sweden Switzerland Taiwan Uganda United Kingdom United States GDP (M CAD) 366,566 1,223,184 373,073 461,019 2,069,411 1,558,310 5,819,474 2,556,702 3,282,487 1,522,586 699,668 2,034,563 5,404,283 32,093 1,028,730 14,957 775,460 126,680 410,317 463,854 1,450,428 439,254 997,013 1,395,847 451,290 518,534 426,274 16,950 2,224,980 14,511,222 Population 40,091,359 22,705,020 8,404,252 10,918,405 190,732,694 34,581,000 1,210,193,422 65,821,885 81,751,602 1,210,193,422 237,556,363 60,626,442 127,950,000 38,610,097 112,336,538 28,584,975 16,695,800 4,315,800 4,968,200 38,186,860 142,914,136 27,136,977 48,988,833 46,125,154 9,440,588 7,866,500 23,188,078 31,800,000 62,435,709 312,200,000 Area (km2) 2,780,400 7,692,024 83,871 30,528 8,514,877 9,984,670 9,640,011 640,294 357,114 3,287,263 1,910,931 301,336 377,930 580,367 1,964,375 147,181 37,354 270,467 323,782 312,685 17,098,242 2,149,690 99,828 505,992 450,295 41,277 36,188 241,550 242,900 9,629,091 IXPs 1 11 2 2 19 2 3 15 14 7 7 7 16 2 0 1 5 7 2 5 14 0 4 6 12 3 4 1 12 85 Bandwidth (Mb/sec) 1,420 3,500 45,600 31,500 73,300 55,600 153,000 38,100 1,380,000 8,330 3,410 84,000 577,000 936 0 19 1,330,000 540 33,000 191,000 759,000 0 437,000 256,000 154,000 25,300 4,730 52 848,000 832,000 24.1 23.89 30.85 30.72 33.66 31.93 22.07 26.74 10.45 38.09 24.93 34.65 14.2 34.03 23.36 31.85 38.07 31.94 27.74 Broadband per 100 Pop/IXP (1,000s) 40,091 2,064 4,202 5,459 10,039 17,291 403,398 4,388 5,839 172,885 33,937 8,661 7,997 19,305 n/a 28,585 3,339 617 2,484 7,637 10,208 n/a 12,247 7,688 787 2,622 5,797 31,800 5,203 3,673
Page 12
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
High priority (8) Calgary Edmonton Halifax Montral Qubec City Vancouver Windsor Winnipeg
Page 13
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
Medium priority (15) Charlottetown Fredericton Iqaluit Kingston London Moncton Regina Saskatoon Sherbrooke St. John St. Johns Sudbury Thunder Bay Whitehorse Yellowknife
Page 14
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
Exhibit 3: Traceroutes from One Point in Canada to Another Point in Canada, passing through the US
The following typical tracroutes illustrate paths taken by trafc routed from one point in Canada to another point in Canada via the United States. In this rst example, the rst router on the path was 206.248.154.0, an IP address registered to Teksavvy Solutions (of Chatham, Ontario), from which the trafc was routed to Washington, DC, and New York, NY, en route to Athabasca University in Athabasca, Alberta. This 2009 example is drawn from the IXmaps Project corpus of traceroute archives, as discussed on page 9.
Host Name (no host name provided) 2110.ae0.bdr02.tor.packetow.ca ge-7-2-6.was12.ip4.tinet.net so-5-1-0.nyc22.ip4.tinet.net telus-gw.ip4.tinet.net edtnabxmgr01.bb.telus.com edtnabkddr02.bb.telus.com atbcab03-athu01.ab.tac.net urania-a.cs.athabascau.ca ren.pc.athabascau.ca IPv4 Address 206.248.154.0 69.196.136.34 77.67.68.125 89.149.187.53 77.67.68.42 154.11.10.141 154.11.5.36 216.123.198.147 131.232.193.100 131.232.31.232 Company Teksavvy Packetow Inteliquent Inteliquent Inteliquent Telus Telus Telus Athabasca University Athabasca University Country Canada Canada United States United States United States Canada Canada Canada Canada Canada
In this second example, utilizing IPv6, the rst router in the path belongs to PriorityColo, of Toronto. The trafc is sent to Hurricane Electric, a US network, in Toronto, which backhauls it to New York, where Bell Canada is willing to receive it, before returning it to Canada.
Host Name gi-2-16.dist01.tor1.ip6.prioritycolo.ca te-1-2.core01.tor1.ip6.prioritycolo.ca 2001:504:1a::34:112 10gigabitethernet4-1.core1.nyc4.he.net paix-ny.bell.ca 2001:4958:5:1::2 2001:4958:5:8::2 2001:4958:5:9::2 2001:4958:5:3::1 Company PriorityColo PriorityColo Hurricane Electric Hurricane Electric Bell Canada Bell Canada Bell Canada Bell Canada Bell Canada Country Canada Canada Canada United States United States Canada Canada Canada Canada
Page 15
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
New York
San Jose
ASN(s)
Chicago
Toronto
London
Ottawa
Seattle
Miami
Paris
Accelerated Connections Advanced Knowledge Networks Allstream Atria Bastionhost Beaneld Bell Aliant Bell Canada Cipherkey E-Gate Galaxybroadband Hydro One Hypnovista Internet Light and Power Mohawk MTO Telecom Neutral Data Centers Nexicom Primus Priority Colo Rogers Shaw SmarttNet Storm TekSavvy TeraGo Trends Videotron Total
21570
14453 6453 15290 26230 36119 21949 855 6539 25668 13657 14500 19752 10533 12059 14537 21548 33554 11666 6407 7788 30176 812 6327 30295 13319 5645 20161 25976 10678 5769 21 5 9 8 5 4 4 3 2 1 1 1 1 1 1 1 1
Page 16
Total
Oslo
15 1 1 2 2 3 1 1 2 1 1 1 1 1 1 5 5 1 2 8 1 2 2 2 1 4
Toward Efciencies in Canadian Internet Trafc Exchange ! Bill Woodcock & Benjamin Edelman
For an in-depth look at the interconnection practices of Internet networks, see Bill Woodcock and Dennis Weller, Internet Trafc Exchange: Market Developments and Policy Challenges. Organization for Economic Cooperation and Development; Directorate for Science, Technology and Industry; Committee for Information, Computer and Communication Policy; Working Party on Communication Infrastructures and Services Policy, DSTI-ICCP-CISP(2011)2, June 2011. Packet Clearing House, Internet Exchange Directory, available at https://pch.net/ixpdir. David J. Phillips, et al., The IXmaps Project, available at http://dev.ixmaps.ischool.utoronto.ca. EthanKatz-Bassett, et al., iPlane: An Information Plane for Distributed Services, available at http://iplane.cs.washington.edu. Ryan Singel, Whistle Blower Outs NSA Spy Room, Wired Magazine, April 7, 2006, available at http://www.wired.com/science/discoveries/news/2006/04/70619. James Bamford, The NSA Is Building the Countrys Biggest Spy Center, Wired Magazine, March 2012, http://www.wired.com/threatlevel/2012/03/ff_nsadatacenter/all/1. See, for example, the International Telecommunications Unions February 2012 proposal Internet Exchange Points (IXPs): Keeping Local Internet Trafc Local. Christian OFlaherty and Bill Woodcock, La Interconexin de Redes en Internet, January 8, 2011. Poltica Digital en Lnea. http://www.politicadigital.com.mx/? P=leernoticia&Article=21023&c=9.
2 3 4
Page 17