23.02.2014 Views

IPv6 Evolution, Stability and Deployment

IPv6 Evolution, Stability and Deployment

IPv6 Evolution, Stability and Deployment

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

2011 19th IEEE International Conference on Network Protocols<br />

<strong>IPv6</strong> <strong>Evolution</strong>, <strong>Stability</strong> <strong>and</strong> <strong>Deployment</strong><br />

Xiaoke Jiang ⋆ , Jun Bi † , Yangyang Wang ⋆ , Zhijie He ⋆ , Wei Zhang ⋆ , Hongchen Tian ⋆<br />

{ ⋆ Dept. of Computer Science, † Network Research Center}, Tsinghua University, Beijing, 100084, P.R.C<br />

Email: ⋆ {justok, wyy, hzj, zw, tianhc}@netarchlab.tsinghua.edu.cn, † junbi@tsinghua.edu.cn<br />

Abstract—Our subject focuses on <strong>IPv6</strong> network, which develops<br />

for more than 10 years. How <strong>IPv6</strong> evolve in those years? Is<br />

<strong>IPv6</strong> network mature enough to undertake the load produced by<br />

users? Can we find some principles to guide <strong>IPv6</strong> deployment,<br />

which make the whole network more robust <strong>and</strong> efficiency?<br />

This paper tries to answer these questions with in-depth<br />

statistics. Good news is that network is growing at a speed of<br />

O(d 2 ) (d is time) after 2006, moreover, network itself <strong>and</strong> its<br />

routing system become more <strong>and</strong> more stable. And we explore<br />

special properties of this preliminary network, We find that<br />

distribution of AS degree follows "Power-Law Distribution", but<br />

AS-level topology cannot be described as "Small-World Model"<br />

properly. We also propose a method to define the importance<br />

of AS <strong>and</strong> give a simple principle of <strong>IPv6</strong> deployment. We even<br />

build "6Stats Project"[1] to provide data which help deploy <strong>IPv6</strong>.<br />

Index Terms—<strong>IPv6</strong>, Routing, <strong>Deployment</strong>, Network Measurement<br />

I. INTRODUCTION<br />

<strong>IPv6</strong> experienced a slow development for a long time since<br />

1999, but now it goes into a rapid growth stage, which is<br />

different from IPv4[4].<br />

But most of research on <strong>IPv6</strong> uses a poor dataset, which may<br />

be small or outdated. What’s more, most of existing research<br />

follows what we have done on IPv4. As a newborn technology,<br />

the special properties haven’t get enough attention. So we<br />

calculate a lot of indexes to show its tomography, based on<br />

which we study on principles of <strong>IPv6</strong> deployment using graph<br />

theory.<br />

In this paper, we use whole <strong>IPv6</strong> BGP RIB data collect by<br />

RouteViews Project from May 3/2003 to Oct 31/2010, Taking<br />

account of limitation of monitors <strong>and</strong> this methodology itself,<br />

we mostly calculate one index result instance using data of<br />

whole month.<br />

II. EVOLUTION<br />

A. Growing at a Speed of O(d 2 )<br />

Fig 1a gives amount of prefixes, ASes, links. If "d" represents<br />

time gap between statistics date <strong>and</strong> starting date,<br />

network grew with the speed of O(d) before 2006[3], And<br />

after [4] argues the speed is O(e d ). But we suppose O(d 2 ) is<br />

more accurate.<br />

We draw asymptotic lines of those results between Jan/2006<br />

<strong>and</strong> Oct/2010 with quadratic polynomial <strong>and</strong> exponential. For<br />

link line(Highest, Green),<br />

y = 0.0025x 2 − 196.36x + 10 6 (R 2 = 0.9927)<br />

is better than y = 3×10 −11 e 0.0008x (R 2 = 0.9445) , for 0.99<br />

is bigger than 0.94. As to AS(Lowest, Red) <strong>and</strong> prefix line, we<br />

8000<br />

7000<br />

6000<br />

5000<br />

4000<br />

3000<br />

2000<br />

1000<br />

AS#<br />

Link#<br />

Prefix#<br />

0<br />

2003 2004 2005 2006 2007 2008 2009 2010 2011<br />

(a) Amount of Entities<br />

35<br />

30<br />

25<br />

20<br />

15<br />

10<br />

Prefix<br />

PATH axes x1y1<br />

Coef.(y2 axis)<br />

5<br />

2003 2004 2005 2006 2007 2008 2009 2010 2011 0.11<br />

(b) Avg Life, Cluster Coef<br />

Fig. 1: <strong>Evolution</strong> of <strong>IPv6</strong> Network<br />

get similar result. So O(d 2 ) is better than O(e d ) to describe<br />

network size. This matches people’s intuition: changing from<br />

O(d) to O(d 2 ) is smoother than to O(e d ).<br />

B. Special Properties of Preliminary Network<br />

According to our statistics, amount of <strong>IPv6</strong> prefix <strong>and</strong> ASes<br />

is just 1% of those of IPv4, we suppose <strong>IPv6</strong> is not ready to<br />

undertake all the traffic people produce daily.<br />

And we also measure the average AS_PATH length from<br />

monitors to other AS. We find an abnormal phenomenon.<br />

Small-World Model infer that length increase slowly while<br />

amount of node increase, which is positive correlation. But<br />

from 2003 to 2010, amount of AS increase(Fig 1a) while<br />

average length decrease! Fig 3a shows this trend 1 . That is<br />

negative correlation. So Small-World Model cannot describe<br />

<strong>IPv6</strong> AS-level topology. But Log-log CCDF line in Fig 3b is<br />

nearly liner, which proves that AS degree meets Power-Law<br />

Distribution.<br />

We suppose the phenomenon is result of small ASes <strong>and</strong><br />

links set, which leading to no "shortcuts" for packet forwarding<br />

between AS pairs. So packet has to follow a long path<br />

without choice.<br />

A. Key Indexes<br />

III. STABILITY OF IPV6 NETWORK<br />

Fig 1b shows average prefix life, path life <strong>and</strong> cluster<br />

coefficient in one month. Life here means appearing days per<br />

months. Path life obviously becomes longer. Upper bound of<br />

prefix life doesn’t change clearly, but lower bound increases.<br />

As for cluster coefficient, it shook more fiercely before 2006<br />

than after that.<br />

1 here path is AS_PATH from other ASes to RouteViews monitors. So result<br />

here is not average path length of whole network, but it does show the trend.<br />

0.17<br />

0.16<br />

0.15<br />

0.14<br />

0.13<br />

0.12<br />

978-1-4577-1393-4/11/$26.00 ©2011 IEEE 123


1<br />

5.8<br />

5.6<br />

5.4<br />

5.2<br />

5<br />

4.8<br />

4.6<br />

4.4<br />

3.5<br />

3<br />

2.5<br />

2<br />

1.5<br />

month<br />

week<br />

day<br />

2003 2004 2005 2006 2007 2008 2009 2010 2011<br />

Fig. 2: Path# for AS Pair in one Month/Week/Day<br />

Average Length of AS PATH<br />

1<br />

0.9<br />

0.8<br />

0.7<br />

0.6<br />

1 10 100 1000<br />

1<br />

CDF<br />

Log-Log CCDF (x2-y2)<br />

0.1<br />

0.01<br />

similar index, clustering coefficient[2] (short as coef in the<br />

following), which reflects node trend of clustering, or "friends<br />

of friends".<br />

Pick 3 types of ASes first: ASes with most neighbors,<br />

biggest local coef <strong>and</strong> smallest local coef, excluding coef equal<br />

0 or 1 in order to clear effect of single link <strong>and</strong> small total joint<br />

group. Then recalculate average coef after removing AS in the<br />

picked list one by one. We find that removal of ASes with<br />

most neighbors cause greatest effect on the new average coef.<br />

This proves that kind of ASes own more important position<br />

in AS-level topology, for they contribute more to network<br />

performance.<br />

So we suppose that there does exist strategies which help to<br />

increase the robust <strong>and</strong> efficient of network when deploy <strong>IPv6</strong>.<br />

For example, connecting to AS "important" neighbors, which<br />

leads average path to other ASes short <strong>and</strong> that AS forwarding<br />

efficiency(but the whole network forwards may not be the most<br />

efficiency). And other deeply strategies can be explored.<br />

4.2<br />

4<br />

3.8<br />

2003 2004 2005 2006 2007 2008 2009 2010 2011<br />

(a) average path length<br />

0.5<br />

0.4<br />

0 100 200 300 400 500 600 700 800 900 0.001<br />

(b) AS Degree CDF <strong>and</strong> log-log CCDF<br />

Fig. 3: Special Phenomenons<br />

We also analyze appearance <strong>and</strong> disappearance of path, links<br />

<strong>and</strong> AS, prefix length <strong>and</strong> origin AS of prefix, all indexes show<br />

trend of becoming stable. In particular, status after 2006 is<br />

more stable than before that, Fig 3a is an evidence.<br />

B. Instability of Routing System<br />

There are 3 lines in Fig 2, which represent average amount<br />

of paths for AS pairs(Source AS <strong>and</strong> destination AS) calculated<br />

in different intervals: Month, Week <strong>and</strong> Day. The value<br />

decreases over time, especially the month line, decreasing from<br />

3 in 2003 to 1.5 in 2010.<br />

We also explore the similarities of set of AS, path <strong>and</strong> links<br />

between sequential months, <strong>and</strong> the similarities increase over<br />

time. All those prove that network itself <strong>and</strong> its routing system<br />

are becoming stable. And year 2006 is prove to be milestone<br />

of stability, too.<br />

A. Environment Faced<br />

IV. DISCUSSION OF DEPLOYMENT<br />

Fig 3b is AS degree CDF <strong>and</strong> log-log CCDF graph 2 , which<br />

shows the environment ISP facing to deploy <strong>IPv6</strong>. 44% of<br />

ASes only have one degree. CDF grows very fast at start-up,<br />

more than 80% of ASes have degree small than 6. It grows<br />

very slow at later stage, less than 8% of ASes have degree<br />

more than 16, but maximal degree is more than 800.<br />

B. Importance of AS<br />

We suppose that the best way to measure importance of<br />

AS is its impact to the average shortest path length, but this<br />

is really hard under the circumstance of complicated BGP<br />

configuration <strong>and</strong> business relationship. So we pick another<br />

2 Based on data in Mar/2010, others show similar properties<br />

V. CONCLUSIONS<br />

In this paper, our conclusions include:<br />

• Scale of <strong>IPv6</strong>, growing at a speed of O(d 2 ), not O(e d )<br />

as some paper argues. Growing doesn’t follow Moore’s Law,<br />

but does go into rapid spread stage.<br />

• <strong>IPv6</strong> network itself <strong>and</strong> its Routing System are becoming<br />

more <strong>and</strong> more stable.<br />

• Year 2006 is a milestone of network. After 2006, network<br />

is more mature <strong>and</strong> stable <strong>and</strong> develops faster. We suppose<br />

the reason is that network change gradually from a scientific<br />

research network to production network, for example, applications<br />

of CNGI-CERNET2 in China, <strong>IPv6</strong> site of Google <strong>and</strong><br />

<strong>IPv6</strong> commercial service of Comcast.<br />

• AS-level topology cannot described as Small-World<br />

method, which reflects properties of its preliminary stage.<br />

• AS with more neighbors take more importance position<br />

in the topology. Connecting with importance AS is a good<br />

strategy of establishing new link, which should get special care<br />

when nearly half of ASes access the Internet by single Link.<br />

And our future work will explore more special properties, a<br />

more accurate model to define importance of AS <strong>and</strong> give<br />

more deeply principles to guide <strong>IPv6</strong> deployment.<br />

VI. ACKNOWLEDGMENTS<br />

Supported by National Science Foundation of China under<br />

Grant 61073172. Program for New Century Excellent<br />

Talents in University. Specialized Research Fund for the<br />

Doctoral Program of Higher Education of China under Grant<br />

200800030034,<br />

REFERENCES<br />

[1] 6Stats Project. http://202.112.49.243/Web6/index/.<br />

[2] Cluster Coefficient. http://en.wikipedia.org/wiki/Clustering_coefficient.<br />

[3] W. Eddy. Basic properties of the ipv6 as-level topology. ACM SIGMET-<br />

RICS Performance Evaluation Review, 36(3):50–57, 2008.<br />

[4] G. Zhang, B. Quoitin, <strong>and</strong> S. Zhou. Phase changes in the evolution of<br />

the ipv4 <strong>and</strong> ipv6 as-level internet topologies. Computer Comunications<br />

(COMCOM), 34(5):649–657, 2011.<br />

124

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!