A Feasibility Evaluation on Name-based Routing
Haesung Hwang1, Shingo Ata2, and Masayuki Murata1
1Graduate School of Information Science and Technology, Osaka University
1–5 Yamadaoka, Suita, Osaka 565–0871, Japan
2Graduate School of Engineering, Osaka City University
3–3–138 Sugimoto, Sumiyoshi-ku, Osaka 558–8585, Japan
Abstract. The IPv4 addressing scheme has been the standard for In-
ternet communication since it was established in the 1960s. However, the
enormous increase in Internet traffic usage has been leading in the past
to issues such as increased complexity of routing protocols, explosion in
routing table entries, provider-dependent addressing, and security prob-
lem, demonstrating the need for a redesign in advanced router technolo-
gies. The past proposals have limitations when it comes to establishing
the foundations of future-generation networks, which require more so-
phisticated routing protocols, like content-based routing. Furthermore,
those previous approaches were not conceived to fully utilize the ad-
vantages of TCAM, which is a type of memory capable of performing
high-speed lookups that is already implemented in high-end routers. In
this paper, we show that routing based on domain names is already a
feasible technology on the Network Layer and we evaluate the necessary
network and hardware resources needed to implement name-based rout-
ing strategies. We present a routing scheme and propose three methods
for equally balancing the routing information in the TCAM of multiple
routers. The results show that this routing scheme is scalable and that
the required number of routers is two orders of magnitude smaller than
the number of currently existing routers.
The Internet Protocol (IP) has been predominantly used for communication
among heterogeneous devices in the Internet. Despite its popular usage today,
problems arise because of significant differences in the traffic carried at the time
it was developed and current trends toward high-volume multimedia communi-
cation. In the following, some of the major issues are listed: (i) Routing protocols
are increasing in complexity and the routing table size is exploding (BGP Re-
ports, http://bgp.potaroo.net), (ii) The IP address acts simultaneously as an
‘ID’ that distinguishes different nodes and a ‘locator’ that specifies the location
of a node in the Internet, which makes the end node addresses depend on the
ISP (Internet Service Provider), (iii) sophisticated routing schemes such as those
based on names or content cannot be used. In this case, it is necessary to use
an overlay or the Domain Name System (DNS), which results in extra resource
consumption and propagation delay, and (iv) there are security problems such
as Distributed Denial-of-Service (DDoS) attacks.
Semantic Routing ，Content Addressable Network (CAN) ，and Loca-
tor/ID Separation Protocol (LISP)  partially resolve the problems mentioned
above, but most of those studies use overlay networks or routing based on agent
nodes, in both cases using IP addresses as before. On the other hand, next-
generation network projects, such as FIND, FP7, and AKARI aim at designing
a post-IP network architecture with one of the main trends being the effort to de-
sign a network that can perform routing using names and semantic information
instead of simply numbered addresses. One of the motivations for name-based
routing is to eliminate the need for DNS servers. Resolving Fully Qualified Do-
main Names (FQDNs) to IP addresses may require queries to multiple servers
which depends on the cache updating policy. Locating a specific host under fre-
quent changes of an IP address is difficult since DNS does not premise dynamic
updates. In addition, name-based routing can separate ID and locator without
using an IP address and provide an unlimited address space.
The final goal of our research is to propose a network architecture that uses
existing resources such as port numbers, applications, and types of informa-
tion. The mechanism for finding, querying, and fetching the desired information
can be performed within the routers themselves, which eliminates the process
of forwarding packets to specific data servers or domain name servers. As a
first step, we propose in this paper a routing scheme that achieves routing by
domain names utilizing the advantage of the router’s Ternary Content Address-
able Memory (TCAM) . Achieving name-based routing with TCAM has been
considered unrealistic in the past because it was presumed that a lot of hard-
ware resources are required. This paper is the first work to our knowledge that
suggests the feasibility of name-based routing by evaluating and estimating the
required TCAM resources.
The rest of this paper is organized as follows. Section 2 surveys existing stud-
ies that propose routing with names instead of the conventional IP addresses.
Section 3 presents the proposed system structure and routing scheme. Section 4
presents three methods to store the domain names in the TCAM of routers.
Section 5 evaluates the necessary network and hardware resources for the meth-
ods presented in Sect. 3 and Sect. 4. Section 6 concludes this paper by briefly
summarizing the main points and mentioning future work.
2 Related Work
In the Name-Based Routing Protocol (NBRP) , the IP address acts only as
a temporary routing tag and the Uniform Resource Locator (URL) is used for
identifying end nodes. When a user requests a certain content, a content router
(CR), which simultaneously acts as a conventional IP router as well as a name
server, returns the address of the server that has the content. However, although
that proposal can eliminate the multiple levels of redirection that is inherent to
DNS, the necessary memory size of the CR is evaluated by DRAM, which is
not optimal for high-speed packet forwarding because it can only perform exact
binary matches. Partial matches of the prefix should be able to correspond to the
network level of the packets in order to forward the packets rapidly to the next
hop. Furthermore, names are used only while the connection is being established
and not for routing the actual data packets. In , another name-based routing
scheme is proposed. They compare the performance of their proposal with IP
in terms of the creation, search, and update time of routing tables, as well
as the required memory. The biggest problem with the proposal mentioned in
that paper is the storage requirement, which is improved through caching and
aggregating domain names. However, the work is still in its initial phase and
the paper does not fully explain whether this is a feasible technology in terms
of hardware resources. Moreover, the algorithm is evaluated using only a single
router and does not state how the overall system should be designed nor do the
authors elaborate on details of the routing method.
In this paper, we focus on evaluating the general feasibility of routing using
domain names instead of IP addresses. This evaluation is done not only for the
network resources, but also for the hardware resources using TCAM.
3 Routing by Domain Names
This section describes the overall structure of the proposed system and the rout-
ing scheme with a purpose of listing the preliminaries and requirements for the
design of a new routing scheme.
Our proposed system has a hierarchical structure in order that local information
is routed in a closed network instead of being transmitted over a widespread
area. In addition, it is possible to take advantage of one specific characteristic
of domain names: they are already separated into different levels by dots. There
are some advantages to having many levels in a hierarchical structure: each level
consist of only few nodes, making the routing table of each node small. However,
there are also some disadvantages such as overhead for traversing multiple levels
and nodes in end-to-end communication. On the other hand, if there is only a
single level, routing information of a lot of nodes have to be handled by each
node, which makes the routing table of each node very large. In research on hi-
erarchical routing for P2P applications [7,8], two-level structures are commonly
considered. In addition, as the number of hierarchy levels increases the reduction
in routing table size is most effective when the structure has two or three lev-
els . Considering the maintenance cost, it is inefficient to have more than three
levels. Therefore, we chose the maximum number of levels of the hierarchy in the
system to be three. Since the number of FQDNs in different Top Level Domains
(TLDs) is not equal, not all TLDs have three underlying levels of routers. If the
?????????????????? ? ?????????
?????????????????? ? ??????????
?????????????????? ? ??????
Top level domain?
2nd level domain?
3rd level domain?
Fig.1. Hierarchical to real topology
router in the highest level can handle all of the routing information within this
TLD, then a single level is sufficient for that particular TLD.
The network topology inspired by the Abilene Network (http://internet2.
edu/network/) and the hierarchical topology used in this paper are shown in
Fig. 1. The Abilene Network is known to have similar characteristics to the
real Internet topology . It groups routers into three levels from the center
outwards to the edge: backbone, local gateway, and edge routers. The architec-
ture in this paper also groups routers into three levels: routers managing TLDs,
routers managing 2nd-level domain names, and routers managing 3rd-level do-
main names. We map this hierarchical topology to the Abilene-inspired topology
correspondingly. Since there are fewer than 300 kinds of TLDs, it is possible to
statically configure which router manages a TLD. For local gateway and edge
routers, we use dynamic configuration to write the routing information, which
is explained in detail in Sect. 4.
Domain name routing differs from IP routing in terms of the registration because
of the ID/locator separation. As IP serves as both an ID and locator, it simplifies
the aggregation of the downstream traffic because IP assignment is the same in
the hierarchical and in the Internet topology. However, domain names are not
always location dependent which makes a separate registration process necessary.
A node that intends to register its domain name injects a ‘registration message’
into the network and this message will eventually reach a router. The router that
returns a ‘destination unreachable’ message because it does not have the path
information is the place where the new domain name will be registered.
Each router has several TCAMs of predetermined size. In order to ensure
that there is always enough room for routing table updates, a threshold value is
used. After a certain threshold value is reached, the routing information must be
written in a new router. A router’s threshold of 0.75 means that the initial storage
available for existing domain names is 75% of the router’s memory capacity.
The remaining 25% is reserved for routing table updates. When a new FQDN
is registered, the utilization ratio of the router is checked. If it is below the
threshold, the domain name is registered in that router; otherwise, there are two
possibilities: split the original router’s routing table in half and divide it between
two routers or leave the original router in its current status and start storing new
domain names in a new router. These ‘new routers’ are designated as candidates
from the beginning and they are only used when a split table needs to be stored.
3.3Routing Table Exchange
Within an Autonomous System (AS) , one or more interior gateway pro-
tocols are used. An exterior gateway protocol such as BGP-4  is used to
route packets between ASs. In domain name routing, an AS can be regarded
as a set of routers that manage the routing information of one or more TLDs.
For example, a set of routers that reside in Japan and mainly have .jp as the
TLD can be regarded as an AS. In order for the routers to have the network
reachability information, it is necessary for the routers to exchange the initial
routing table that was created at the time of the domain name registration de-
scribed in Sect. 3.2. In the case of IP, routers exchange information about which
subnets the network can reach, such as 192.168.0.0/16, which means hosts from
192.168.0.0 to 192.168.255.255 can be reached. Similar behavior can be achieved
in domain name routing by exchanging information such as *.osaka-u.ac.jp,
which means that all domains having osaka-u.ac.jp as their suffix can be
reached. The exchange of routing tables is achieved by extending BGP-4.
3.4 Packet Forwarding
Routers forward packets on the basis of the information contained in the routing
table stored in its TCAM. An example of packet forwarding is shown in Fig. 2.
When the packets are forwarded from computer.ist.osaka-u.ac.jp to biz.
yahoo.com, the former sends packets to the router R1. This router’s routing
table consists of three parts for domain names in the upper level, the same level,
and the lower level. Since the packets are destined for a higher level than router
R1, they are forwarded to port P. Router R2 goes through a similar process
and sends the packets to port Q. Since router R3 is in the highest level of the
hierarchy, the packets are sent to port X to reach the router that has the more
specific address of the .com suffix. Router R4 refers to its routing table to send
the packets to port A and the packets finally reach R5, which has the direct
routing information for the destined domain name. The process is also done by
extending BGP-4, using character information instead of the IP address.
4 Balanced Distribution of Domain Names in Routers
Compared with the IP routing tables, domain name routing tables are larger
because of the variable length of domain names. Therefore, it is important to
balance the routing information among multiple routers. In this section, we pro-
pose three methods for equally distributing the routing information.
com.************** port P
net. ************** port P
jp. **************** port P
com.** port X
net.** port Y
jp.**** port A
jp.** port Z
net.** port V
com.* port A
biz yahoo com
com.******** port Q
net.********* port Q
jp.********** port Q
com.yahoo.* port A
Router R1’s Routing Table?
jp.ac.osaka-u.ist.** port A
jp.ac.osaka-u.***** port X
com.************* port Q
net. ************* port Q
jp. ************** port Q
jp.ac.************ port X
Router R2’s Routing Table?
jp.ac.osaka-u.*** port A
X X X
R R4 4?
R R2 2?
R R3 3?
R R1 1?
R R5 5?
? The routing tables are stored in TCAM?
Fig.2. Packet forwarding
4.1 Hash-based Distribution
The simplest of the three proposed distribution methods is hash-based distri-
bution. This has the best performance in terms of balanced distribution of the
routing table. An FQDN is made ‘flat’ by hashing it with SHA-1 where ‘flat’
means that it is no longer hierarchical. Since every FQDN is considered to have
equal status, we can obtain the total number of routers required for domain
name routing by multiplying the number of domain names by the number of bits
needed per entry and then dividing the result by the TCAM size in a router.
The advantage of this method is that the characteristics of SHA-1 allow a large
database to be equally distributed into a given number of groups. However, do-
main names that end with ccTLDs (country code TLDs) lose locality information
after hashing. In addition, when the nodes’ physical locations have no relation
to their IDs, it may lead to the problem of long stretch which is defined in 
as the ratio of the number of physical hops taken by the protocol to reach the
destination node from the source node and their shortest distance in terms of
physical hops. Therefore, although it is considered ideal in terms of achieving a
nearly equal distribution, but it is not realistic for implementation.
4.2 Hierarchical Longest Alphabet Matching
In order to store domain names in memory, hierarchical longest alphabet match-
ing represents domain names in ASCII code. The characters used in domain
names are 26 case insensitive letters of the English alphabet and the 10 numbers
from 0 to 9 plus hyphens and dots . Inspired by the longest prefix matching
scheme, longest alphabet matching is applicable when the domain names are
represented in binary format. In addition to the good searching speed, aggrega-
tion of multiple domain names with the same suffix is possible, which reduces
the occupied memory space. Hierarchical longest alphabet matching takes full
advantage of TCAM, which can represent a don’t care value ’*’ in each cell in
addition to 0 or 1. To calculate the total number of routers required, we first
compute the number of routers required in each TLD. The pseudocode for this
calculation is shown in Alg. 1.
Algorithm 1 Numbers of routers for Hierarchical longest alphabet matching
nL2← number of routers in 2nd level
nL3← number of routers in 3rd level
t ← threshold of TCAM utilization
entryl← 180 (bits per one entry)
routerc← 18 × 106× 10 × t
u ← entries with unique 2nd-level domain names
e ← entries in the TLD
if e × entryl≤ routercthen
RETURN nL2← 1, nL3← 0
else if u × entryl> routercthen
divide u into groups (dynamic configuration using 2nd-level domain name)
RETURN nL2← number of groups
for each group Func Calculate nL3
nL2← 1, Func Calculate nL3
Func Calculate nL3
divide e into groups (dynamic configuration using 3rd-level domain name)
RETURN nL3← number of groups
For each TLD, the first step is to determine whether all e domain name
entries would fit into a router. Each entry is multiplied by 180 bits (entryl)
and divided by the router’s available TCAM size (routerc). We assume that
one entry consumes 180 bits in order to make comparisons with the result in
Sect. 4.1, where 160 bits were the hashed value plus 20 bits of output port plus 4
parity bits. Again we define a router as having 10 TCAMs, each TCAM having
a capacity of 18 Mbits. Threshold t is explained in Sect. 3.2.
If e multiplied by entrylis less than routerc, then the total number of routers
required in this TLD is set to 1, and the process ends. Otherwise, u unique 2nd-
level domain names are written in the 2nd level of the hierarchy in the form of
TLD.uniq.*. The routers in the highest level are written with TLD information,
which we ignore for the moment. Here, u domain names are written dynami-
cally by referring to ASCII table. The basic idea is that names are grouped in
alphabetical order, and when a router overflows, it starts storing the domain
names in a new router. An additional idea is to take advantage of the prefix
values. Names are first divided into two groups: digits and letters. If the size
is still too large, the letters are divided again into smaller groups, a group of
110**** and a group of 111****. This process is repeated until every router is
well within the routerc. One example combination of the grouping is, hyphen
and digits (01*****), a to c (11000**), d to g (11001**), h to o (1101***), p
to q (111000*), r to s (111001*), t to w (11101**), and x to z (1111***).
When the partitioning of the 2nd-level domain names is over, the grouping
is done with 3rd-level domain names. However, since the maximum number of
levels in the hierarchy considered in this work is three, routers are written with
FQDNs instead of with the unique 3rd-level domain names. The number of
groups in each level corresponds to the number of routers. Therefore, the total
number of routers required is nL2+ nL3.
In this section, TLDs are distinguished by 9 bits because there are fewer than
300 TLDs. The hash function is applied to 2nd and 3rd-level domain names
separately. To calculate the total number of required routers, we first compute
the number of routers required in each TLD. The pseudocode is shown in Alg. 2.
Algorithm 2 Number of routers for Hybrid distribution
(Refer to Alg. 1 for the variables)
if e × entryl≤ routercthen
RETURN nL2← 1, nL3← 0
else if u × entryl≤ routercthen
RETURN nL2← 1
RETURN nL3← (e × entryl) / routerc
nL2← (u × entryl) / routerc
hash(2nd level domain name) % nL2(divide routers in 2nd-level into groups)
for each group calculate nL3
RETURN nL3← (egroup× entryl) / routerc
The process is the same as Alg. 1 until the list of u does not fit into a single
router. We obtain nL2 routers required in the 2nd level in the hierarchy by
multiplying u by entryland dividing it into routerc. In other words, the number
of groups based on the 2nd-level domain names is equal to hashing 2nd-level
domain names modulo nL2 to the most significant character. Then we obtain
groups that have nearly equal entries independent of the alphabets. The next
step is to calculate nL3routers required in the 3rd-level; these routers are located
under the corresponding routers in the 2nd level of the hierarchy. The router in
which each FQDN is stored depends on the hash value of the FQDN’s 2nd-level
domain names. To simplify the calculation of nL3, each entry is multiplied by
entryl and divided by routerc. This was done assuming that hashing has the
characteristic of equally distributing a large amount of data among groups.
5 Evaluation and Discussion
We evaluate the network and hardware resources required for domain name
routing with entries obtained from the ISC database in July 2008 .
5.1 Network Resources
Figure 3 shows the distribution of domain name entries in routers, arranged in
ascending order of the routers’ utilization ratio. The utilization is defined as the
number of entries in a router multiplied by 180 bits divided by the router’s mem-
ory size. The graph shows the results for hierarchical longest alphabet matching
and hybrid distribution for threshold values of 1, 0.75, and 0.5. The total num-
ber of required routers is smallest when the threshold is 1, letting the expected
number of routing hops to be minimum. However, since there is no space left for
updating routing tables, setting the threshold to 1 is not a realistic choice.
For each threshold, the required number of routers is 29.1% smaller on aver-
age for hybrid distribution than for hierarchical longest alphabet matching. In
addition, hybrid distribution is likely to make a better use of the memory space
because there are more routers with a utilization ratio close to the threshold.
The performance of both methods can be improved by aggregating those
entries in the routers that have a low utilization ratio. Since both algorithms
assign at least one router for a single TLD, approximately 150 TLDs out of
270 TLDs have a utilization ratio of less than 0.1%. Those entries are collected
and stored in only a small number of routers. That reduces the total number
of routers required to an average of 14.2% and 16.5% for hierarchical longest
alphabet matching and hybrid distribution, respectively. Therefore, statically
assigning routers in backbone routers, as proposed in Sect. 3.1, needs careful
consideration of the growth rate of domain names in each country.
Comparing these two algorithms, we can see that hybrid distribution needs
less routers, but hierarchical longest alphabet matching utilizes the advantage
of TCAM better. If TCAMs can handle DHT (Distributed Hash Table) data,
then hybrid distribution can also be considered to be a feasible method.
The number of routers required for different thresholds in routers are shown
in Fig. 4. Depending on the update rate of the routing table, it is possible to
estimate how the required number of routers will increase. The result shows
that the initial storage of existing domain names with a threshold value of 0.2
would require approximately 4,000 routers for both hierarchical longest alphabet
matching and hybrid distribution.
?? ???? ???? ???? ???? ????? ????? ????? ????? ?????
(a) Hierarchical longest alphabet matching
???????????? ????? ????? ?????
(b) Hybrid distribution
Fig.3. Distribution of domain name entries in routers for different thresholds
When hash-based distribution is used, the average number of entries among 4,096
routers is 145,976. We obtain the memory size of 26 Mbits by 180 bits (SHA-1
output (160 bits) + output port (16 bits) + parity bits (4 bits)) × 145,976.
Thus, having two 18-Mbit TCAMs is sufficient when there are 4,096 routers. In
other words, if there are ten 18-Mbit TCAMs in a router, a total of 597 routers
((145,796 entries × 180 bits × 4,096 routers) / (18 × 106bits × 10 TCAMs))
are required to store the existing domain names.
The number of routers required in hierarchical longest alphabet matching is
1,396, assuming that one entry consumes 180 bits. 7-bit ASCII code is sufficient
to distinguish the characters used in domain names. Furthermore, since there are
fewer than 300 TLDs, they can be differentiated using 9 bits. Whereas it is easy
to adjust the bit length in each entry in hash-based distribution because each
entry is hashed, this is not the case in hierarchical longest alphabet matching. To
satisfy the static length (180 bits) that we used to evaluate the required number
?????? ???????? ???? ??
Fig.4. Number of necessary routers required for different thresholds
of routers throughout this paper, 171 bits must be free for use, excluding the
9 bits reserved for the TLD. When one character uses 7 bits, approximately
25 characters can be written which only consists of 30% of the FQDNs .
To store 99% of the FQDNs, the TCAM must be able to store entries having
up to 50 characters. Although the search speed decreases when the lookup size
increases , designing the TCAM to suit longer bit lengths does not impose
much of a technological difficulty.
The number of routers required for hybrid distribution is 952. Hierarchical
longest alphabet matching and hybrid distribution require 2.4 and 1.7 times
as many routers as hash-based distribution, respectively. However, they both
use hierarchical structures, which have the advantage of restricting the local
information to be routed in a closed network instead of causing it to be forwarded
over a widespread area. Although hierarchical longest alphabet matching cannot
achieve a nearly equal distribution of domain names among the routers, because
it does not use a hashing function, it should reduce the burden of applying
hashing to every entry.
In [16,17], the authors use the work from  and estimate that the number
of routers deployed world-wide is approximately 228,260. Therefore, the router
numbers required for the methods proposed in this paper are realistic values
that indicate that a name-based routing architecture is feasible.
6 Conclusion and Future Work
In this paper, we used the domain name in routing as a substitute for the conven-
tional IP address. Through statistical evaluations of currently existing domain
names we showed that the proposed method is feasible in the Network Layer by
estimating the required network and hardware resources. Future work involves
evaluating the routing table updates by observing the evolution of the registered
12 Download full-text
domain name entries over time. This includes estimating the rate of increase of
entries as well as that of the number of newly added names. In addition, the
effect of eliminating the current domain name servers should be investigated.
Acknowledgement This research was supported by National Institute of In-
formation and Communications Technology (NICT) of Japan.
1. Inoue, K., Akashi, D., Koibuchi, M., Kawashima, H., Nishi, H.: Semantic Router
using Data Stream to Enrich Services. In: Proc. of CFI. (Jun 2008) 20–23
2. Ratnasamy, S., Francis, P., Handley, M., Karp, R., Schenker, S.:
Content-Addressable Network. In: Proc. of SIGCOMM. (Aug 2001) 161–172
3. Farinacci, D., Fuller, V., Oran, D.: Locator/ID Separation Protocol (LISP). ”draft-
farinacci-lisp-10.txt”, IETF Network Working Group, work in progress (Nov 2008)
4. Renesas Technology Corporation: Network Address Search Engine (9 M/18 M-
bit Full Ternary CAM) available at http://documentation.renesas.com/eng/
5. Gritter, M., Cheriton, D.R.: An Architecture for Content Routing Support in the
Internet. In: Proc. of USITS. (Mar 2001) 37–48
6. Shue, C.A., Gupta, M.: Packet Forwarding: Name-based Vs. Prefix-based. In:
Proc. of GI. (May 2007) 73–78
7. Garc´ es-Erice, L., Biersack, E.W., Felber, P.A., Ross, K.W., Urvoy-keller, G.: Hi-
erarchical Peer-to-Peer Systems. In: Proc. of Euro-Par. (Aug 2003) 643–657
8. Kersch, P., Szabo, R., Kis, Z.L., Erdei, M., Kov´ acs, B.: Self Organizing Ambient
Control Space: An Ambient Network Architecture for Dynamic Network Intercon-
nection. In: Proc. of DIN. (Sep 2005) 17–21
9. Lian, J., Naik, S., Agnew, G.B.: Optimal Solution of Total Routing Table Size for
Hierarchical Networks. In: Proc. of ISCC. (Jun 2004) 834–839
10. Li, L., Alderson, D., Willinger, W., Doyle, J.: A First-Principles Approach to
Understanding the Internet’s Router-level Topology. In: Proc. of SIGCOMM. (Aug
11. Hawkinson, J., Bates, T.: RFC 1930: Guidelines for Creation, Selection, and Reg-
istration of an Autonomous System (AS) (Mar 1996)
12. Rekhter, Y., Li, T.: RFC 1771: A Border Gateway Protocol 4 (BGP-4) (Mar 1995)
13. Lu, G.H., Jain, S., Chen, S., Zhang, Z.L.: Virtual Id Routing: A Scalable Routing
Framework with Support for Mobility and Routing Efficiency. In: Proc. of ACM
MobiArch. (2008) 79–84
14. Mockapetris, P.: RFC 1035: Domain names - Implementation and Specification
15. Internet Systems Consortium: available at https://www.isc.org
16. Yook, S.H., Jeong, H., Barab´ asi, A.L.: Modeling the Internet’s Large-scale Topol-
ogy. In: PNAS. (Oct 2002) 13382–13386
17. Lakhina, A., Byers, J.W., Crovella, M., Matta, I.: On the Geographic Location of
Internet Resources. IEEE JSAC 21(6) (Aug 2003) 934–948
18. Govindan, R., Tangmunarunkit, H.: Heuristics for Internet Map Discovery. In:
Proc. of INFOCOM. (Mar 2000)