Professional Documents
Culture Documents
Abstract—The Internet has being growing rapidly with the of processing. In section 4,the evolution of Internet is
development of technique. Understanding the evolution of discussed. The conclusion of this paper would be presented in
Internet is essential to the Internet security. there has been some section 5.
studies on the evolution of Internet, which only analyze the
change of the Internet during a quiet short period. In this paper,
we have roughly analyzed the evolution of the Internet for last 20 II. RELATED WORKS
years and take days as the granularity of time. Some important During past several years, many researchers has analyzed
measures , such as average-distance and average-node-degree, the evolution of Internet. In history, the most cited literature is
have been analyzed. Our findings show that the growth of literature [7].The authors firstly focus on the node degree of
Internet is nearly binomial and the evolution of Internet can be AS, and conclude that the distribution of AS obey a power-law
divided into 3 different phases. Moreover, these findings can be distribution. Later, Zhou et al[8] further analyses the
used to model the evolution of Internet which would help us to distribution of degree, experimental results show that 0.1% of
better protect network security. nodes would take up over 80% of links, which means that the
node of Internet would like to connect to the nodes with large
Keywords—Internet; measurement; evolution; AS level
degree. At the same time,Tier1 network is extracted from
Internet. CAIDA[9] emphasis on the difference of three
I. INTRODUCTION (HEADING 1) different data source, including whois, skitter and RIPE, and
With the development of technique, Internet has being analyses their characteristics. The results show that the data
growing rapidly, having more than tenth in size in the past, from RIPE is more complete than others, and the
from 3000 Autonomous Systems in 1998 to 60000 characteristics of whois is quite different with that of others. In
Autonomous Systems in 2016.Since the fast changing of 2008,Gill et al.[10] mainly focus on the behavior of
Internet, it is necessary to have a clear understanding on the construction of text ISP results show that with the aim of
evolution of Internet. On the other hand, as the progress of ensuring its safety and stable, text ISP tend to connect to
attack technique, Internet is facing increasing challenges than nodes with small degree. This phenomenon means that
before. Hence, it is helpful to protect Internet security if we Internet may be flat in the future. Reference [1] analyses the
could understand the characteristics and the evolution of size of Internet data collected from 1998 to 2007,and conclude
Internet. that the number of AS exhibits an exponential increase before
2001,and growing linearly later. However, literature [3]
During the past, there are many studies focus on the shows that the number of AS is growing as exponentially.
analysis on the evolution of Internet[1-6].However, these
studies mainly analyze the evolution of Internet during a short
period. Since the RouteViews which collect the BGP data III. DATA SOURCE
have collecting data for over 20 years, we would like to A study of the evolution of Internet needs a frequent
analyze the evolution of Internet according to these data to get snapshots of the AS level Internet topology. Given that such
a better understood of Internet. And it is important for us to information is not available, we have to collect the data from
know the evolution of Internet if we want to create the other data source. The datasets we use and the way of filtering
evolution model for Internet. dirty data is introduced in this section.
In this paper, we attempt to measure and understand the We collect BGP data from BGP table dump obtained from
evolution of Internet ecosystem during the last twenty Route Views[11] for each day since November 1997.Note that
years(1997-2016).We mainly focus on some important not all AS path is useful, since BGP data may be false
characteristic of network, such as node degree, average AS sometimes. The situation of false data can be divided into
path. The growth of Internet is nearly binomial. Furthermore, three categories as follows:
three phases of Internet evolution is concluded. Firstly, new
1.Loops of Path. Loops of path means that one route has
added nodes tend to link to the center of Internet. At the
been received by a AS twice, which would cause the cycling
second phase, new added nodes would like to connect to the
of Internet traffic.e.g.AS path with 7652 8769 7652.
nodes located in the edge of Internet. Finally, Internet has
become stable for a long time. 2.Containing of Bogus Internet resource. The resource of
Internet here mentioned mainly is the unallocated AS number
The rest of this paper is organized as follows. In section
and the unallocated prefix.e.g.AS path with 458 785 65534 is
2,we would introduce the related work on the evolution of
Internet. The section 3 gives the source of data and the method
not useful in our analysis since the AS 65534,a private AS,
which is not available in the Internet.
3.Reduplicate AS path. There are many reduplicate AS
path, since the reduplication of AS in the path can restrict the
traffic translation, which would relief the traffic pass through
itself.e.g.AS path is 234 5631 667 667 899.
In order to get a nearly clean dataset for the analysis of the
Internet, we clear these dirty data from the collected BGP data.
Almost 5700 records are collected from November 8,1997 to Fig. 2. Correlation between AS number and interAS links
July 23,2016 for each day.
B. Evolution of node degree
IV. EXPREIMENTAL ANALYSIS In this section, we mainly analyze the evolution of node
degree in the Internet.Fig.3 shows the changing of the average
A. Grows and Trends of ASlevel Internet degree. As the figure shows, the changing of the average
The trends of AS level Internet mainly refer to the growing degree can be divided into three stages: firstly, the value of
of AS nodes and their connection between each other.Fig.1 average degree increase from November 1997 to 2003.The
shows that the number of ASes and the number of inter AS second phase is that the slowly decreasing of average degree
links in each snapshot. From Fig.1,a first observation is that, since 2003 until 2009.And at last, the node average degree
the increasing of the Internet is very fast during the past 20 increase to the largest value,5.It can be observed that the
years. Secondly, it appears that the growing trends of the AS growth of the firstly stage and the thirdly stage is similar.
number is the same as that of inter-AS links. And this situation
From the overall point of view, average degree of the
continues all the time. Then, the growing trends is regressed,
Internet is creasing all the way. It means that instead of
and the optimal fitting curves is switched to binomial for both
connecting to only one AS, more and more AS tend to connect
the number of nodes and edges.
to more ASs so that suddenly shutdown of some ASs may not
influence its normal work.
switched to be linearly.
-4
10
-5
10
-4 -3 -2 -1
10 10 10 10
Node Centrality x
0.25
0.10
We can see that the value of assortativity coefficient of the
Internet is negative all the time. It means that the nodes with
0.05
small degree prefer to link to the nodes with large degree than
0.00
0.0 0.2 0.4 0.6 0.8 1.0
the nodes with low degree. However, as shown in the figure,
Local Clustering lc
the value of assortativity coefficient is increasing slowly
through the whole time. This suggests that there are more low
Fig. 6. Evolution of node local cluster coefficient degree nodes prefer to connect the small degree nodes than
before, but the circumstance that Internet is disassortative
It is obviously that the distribution of node local cluster never changes all the time.
coefficient is stable as well as the distribution of node
centrality degree. But the difference is that the node local F. Evolution of standard network structure entropy
cluster is not strictly obey the power law distribution since that Standard network structure entropy(SNSE) is used to study
there is a suddenly decline in the distribution when the value the heterogeneity of the network. The evolution of SNSE is
of local clustering is 0.5. shown in Fig.7.It is obviously that the value of SNSE is
waving nearby a constant value,0.63,during the whole time.
Firstly, the value of average degree decrease as well as the
average cluster coefficient. The value of average distance
between each AS has increased since 2003 as well as the
assortativity coefficient of Internet. The reason may be the
increasing number of the stub network at the edge of the
Internet. At the third phase, the change of each measure has
become steady, which means that the Internet is becoming
more stable than before.
V. CONLUTION
During the past, there has been many studies focus on the
Fig. 9. Evolution of SNSE analysis of evolution of the Internet. However, the studies
mainly analyze the evolution of Internet during a short period.
Furthermore, the increasing rate of the SNSE is learned as In this paper, we measure the evolution of the AS-level
Fig.8 shows. It is obviously to see that the increasing of SNSE topology over the last 20 years in terms of the change of
is very small, which means that the structure of the Internet is measure. Our findings highlight some important trends: The
stable. growth of the Internet is nearly binomial when taking days as
the granularity of time. Based on the evolution of some
important measure, the evolution of Internet can be divided
into three different phases, which each phase has its special
characteristics. With this findings, we could then simulate the
evolution of Internet, and model Internet evolution which
could be used to protect the Internet safety better.
REFERENCES
[1] Dhamdhere A, Dovrolis C. Ten years in the evolution of the Internet
ecosystem[C]//Proceedings of the 8th ACM SIGCOMM conference on
Internet measurement. ACM, 2008: 183196.
Fig. 10. Evolution of increasing rate of SNSE [2] Oliveira R V, Zhang B, Zhang L. Observing the evolution of Internet AS
topology[C]//ACM SIGCOMM Computer Communication Review.
ACM, 2007, 37(4): 313324.
G. Summary of the evolution of Internet [3] Zhang G Q, Zhang G Q, Yang Q F, et al. Evolution of the Internet and
its cores[J]. New Journal of Physics, 2008, 10(12): 123027.
Analysis on evolution of each characteristic is independent
above, this section want to give a summary of the evolution of [4] Dhamdhere A, Dovrolis C. Twelve years in the evolution of the Internet
ecosystem[J]. IEEE/ACM Transactions on Networking (ToN), 2011,
Internet. It is clearly to see that the distribution of node degree 19(5): 14201433.
and local cluster coefficient is mainly obey the power-law [5] Soffer S N, Vazquez A. Network clustering coefficient without
distribution, and this phenomenon never changes. Also, the degreecorrelation biases[J]. Physical Review E, 2005, 71(5): 057101.
evolution of increasing rate of SNSE shows that the structure [6] Chang H, Willinger W. Difficulties measuring the Internet's ASlevel
of Internet is always stable. The growth of the AS numbers ecosystem[C]//2006 40th Annual Conference on Information Sciences
and inter-AS links is nearly binomial when taking days as the and Systems. IEEE, 2006: 14791483.
granularity of time. It may give us a new way to predict the [7] Faloutsos M, Faloutsos P, Faloutsos C. On powerlaw relationships of the
number of AS and links. Also, it can help us create evolution Internet topology[C]//ACM SIGCOMM computer communication
review. ACM, 1999, 29(4): 251262.
model of Internet which takes days as the granularity of time.
[8] Zhou S, Mondragón R J. The richclub phenomenon in the Internet
According to the analysis above, we think that Internet can topology[J]. IEEE Communications Letters, 2004, 8(3): 180182.
be divided into three phases. The first phase is before [9] Mahadevan P, Krioukov D, Fomenkov M, et al. The Internet ASlevel
2003.During the phase, the average degree of AS is increasing topology: three data sources and one definitive metric[J]. ACM
as well as the average cluster coefficient. However, at the SIGCOMM Computer Communication Review, 2006, 36(1): 1726.
same time, the average distance decreases. It means that the [10] Gill P, Arlitt M, Li Z, et al. The flattening Internet topology: Natural
evolution, unsightly barnacles or contrived collapse?[C]//International
new node may be added to the nearly center of Internet. The Conference on Passive and Active Network Measurement. Springer
second phase can be concluded from 2003 to 2009.During the Berlin Heidelberg, 2008: 110.
period, the trends of measures is different with the first phase. [11] Route Views Project Page[OL].http://www.routeviews.org.2005.