"... In PAGE 10: ... According to the DNS name of the router from which a unreachable ICMP is sent, we identify the AS to which the router belongs. Table6 shows the location for the first loss due to routing failures. We observe that for recovery- 1 events, about 90% of routing failures occur within ISP1, while no routing failures are observed within ISP2 during recovery-2 events.... ..."

"... In PAGE 5: ...3 Which ASes leak bogon prefixes, and who sees them? By examining the origin AS in the AS path corresponding for each bogon prefix announcement, we observed 110 different ASes originated bogon prefixes. Table5 lists all 19 ASes that caused at least three origin AS-based events; the remaining 91 ASes only caused one or two origin AS-based events each, although some of these events were quite large: on March 25, 2004, AS 19962 leaked 87 bogon prefixes to Cogent, who passed these bogon routes to MIT. Table 5 shows that a few ASes are responsible for a large number of prefix-based and origin AS-based events.... In PAGE 5: ... Table 5 lists all 19 ASes that caused at least three origin AS-based events; the remaining 91 ASes only caused one or two origin AS-based events each, although some of these events were quite large: on March 25, 2004, AS 19962 leaked 87 bogon prefixes to Cogent, who passed these bogon routes to MIT. Table5 shows that a few ASes are responsible for a large number of prefix-based and origin AS-based events. Most of these events were typically visible at only one or two monitoring points, implying that, despite their frequency, these bogon route leaks are not visible at many places on the Internet.... ..."

"... In PAGE 5: ...3 Which ASes leak bogon pre xes, and who sees them? By examining the origin AS in the AS path corresponding for each bogon pre x announcement, we observed 110 different ASes originated bogon pre xes. Table5 lists all 19 ASes that caused at least three origin AS-based events; the remaining 91 ASes only caused one or two origin AS-based events each, although some of these events were quite large: on March 25, 2004, AS 19962 leaked 87 bogon pre xes to Cogent, who passed these bogon routes to MIT. Table 5 shows that a few ASes are responsible for a large number of pre x-based and origin AS-based events.... In PAGE 5: ... Table 5 lists all 19 ASes that caused at least three origin AS-based events; the remaining 91 ASes only caused one or two origin AS-based events each, although some of these events were quite large: on March 25, 2004, AS 19962 leaked 87 bogon pre xes to Cogent, who passed these bogon routes to MIT. Table5 shows that a few ASes are responsible for a large number of pre x-based and origin AS-based events. Most of these events were typically visible at only one or two monitoring points, implying that, despite their frequency, these bogon route leaks are not visible at many places on the Internet.... ..."

"... In PAGE 32: ... Indeed, Figure 1 and Figure 4 look very similar for this reason. Table8 examines the entry strategies of the firms in the largest number of counties. It... In PAGE 34: ... These firms have no reason to provide extensive networks comprised of many POPs when they also provide direct access lines to businesses. Table8 illustrates one important reason why most of the national firms do not expand beyond several hundred counties. Many national firms provide access to just under a third of the US population by having presence in approximately 100 counties (and this does not count neighborhood effects).... ..."

"... In PAGE 7: ... In this section, we an- alyze the effects of peering (particularly the Tier-1 peering) link failures on network reachability and traffic shift. Table7 presents the number of single-homed customers with and without the stub ASes for each Tier-1 AS, where single-homed refers to customers that can only reach only one Tier-1 AS through uphill paths. If all the physical peer- ing links between two Tier-1 ASes stop working, i.... ..."