Documente Academic
Documente Profesional
Documente Cultură
AbstractWith the exhausting of IPv4 addresses, the conduct a comparison study on packet delay and explore the
transition to IPv6 is imminent. In order to gain a deep reasons for the difference by dividing the as-level paths into
understanding of IPv6, this paper revisits several critical IPv6 three parts, i.e., domestic-as, middle-as and target-domestic-as.
performance metrics. Our extensive measurement shows that We find that the connectivity between middle-as and target-
packet delay and loss of IPv6 is similar to IPv4 when the AS-level domestic-as, and the length of middle-as are the dominant
paths are roughly the same. Specifically, when the link utilization factors that lead to the differences in RTT (Round-Trip Time)
exceeds a threshold, e.g., 0.83 in our study, variation of packet between IPv4 and IPv6. 2) We make a comparison between
delay presents a similar pattern with the variation of link IPv4 and IPv6 in packet-loss rate. The average packet-loss rate
utilization. If packet delay of a path is large, packet-loss rate of
of IPv6 is 0.25%, while the average packet-loss rate of IPv4 is
that path is more likely to fluctuate. In addition, we conduct an
analysis of packet reordering in IPv6 world. Few IPv6 probe
0.33%. We also investigate the relationship between packet
packets are out-of-order and the reordering rate is 2.310-6, delay and packet-loss rate. Packet-loss rate of a path is more
which is much lower than the average rate of 0.79% in IPv4 likely to fluctuate when packet delay of that path becomes
world. Our analysis consolidates an experimental basis for IPv6 large. 3) We conduct an analysis of packet reordering in IPv6
network operators and researchers. world. We find that average packet reordering rate of IPv6 is
2.310-6, while average packet reordering rate of IPv4 is
Keywordsnetwork measurement; IPv6; delay; packet loss; 0.79%. The reason for the difference is that fragmentation in
packet reordering; IPv6 is not encouraged and only the hosts can conduct packet
fragmentation in IPv6 network, which could reduce the
I. INTRODUCTION possibility of packet reordering.
Due to the predictable exhaustion of IPv4 addresses, The remainder of the paper is organized as follows. The
transition to IPv6 is imperative. To try and encourage adoption related work is presented in Section II. We describe the
of IPv6, more than 400 organizations and institutions methodologies used to measure IPv6 performance in Section
participated in the Word IPv6 Day on 8th June 2011 [1]. To III. Section IV presents our analysis results, including delay,
further promote the deployment and development of IPv6, packet loss rate and packet reordering. Our conclusions and
World IPv6 Launch was held on 6th June 2012 [2]. Google has future remarks are presented in Section V.
announced that over 4% of users access Google through IPv6
[3]. CAIDA has shown that from January 2013 to January II. RELATED WORK
2014, the number of IPv6 ASes increased by 80% and the
number of links between them increased by 63% [4]. All these Related work is summarized from two aspects. We begin
reveal that IPv6 network is developing rapidly. In addition, as with discussion on measurement of packet delay and packet
applications such as video conference, IPTV, VoIP and online loss. Then we show the studies that have been conducted to
games become more and more prevalent, keeping network characterize packet reordering in IPv4 world.
availability as well as good performance is very crucial for Wang et al. [5] compared IPv6 and IPv4 performance from
network operations. Investigating the performance of IPv6 the perspective of end users and found that it still has
network is the basis of guaranteeing the network with a good headroom to improve IPv6 performance, including increasing
performance. Therefore, many studies have been performed to its connectivity and reducing its packet-loss rate. Muniyappa
analyze IPv6 network performance, as well as figure out the et al. [6] carried out a simulation-based study comparing the
differences between IPv4 and IPv6 network [5-10]. However, use of the IPv4 versus IPv6 on simple campus networks and
our understanding of IPv6 network cannot catch up the speed found that performance of IPv4 and IPv6 are almost the same
of its development. A revisiting and multi-dimension and the difference is negligible. Shiwani et al. [7] measured
observation of current IPv6 network performance is needed. throughput, delay, and jitter on a test-bed implementation.
In this paper, we use OneProbe [16], a new TCP Zhou et al. [8] found that IPv6 paths have a higher delay and
(Transmission Control Protocol) method for reliable and loss than their IPv4 counterparts, mainly due to its poor
metric-rich path monitoring, to revisit IPv6 performance and performance in IPv6-in-IPv4 tunnels. Nikkah et al. [9] and
make a comparison with IPv4 from many aspects, such as Dhamdhere et al. [10] found that IPv6 delay is similar to IPv4
packet delay, packet loss and packet reordering, etc. Our main performance when the AS-level paths are identical. Czyz et al.
findings and contributions are summarized as follows. 1) We [11] concluded that IPv6 RTT is getting better according to a
APNIC 300
250 AFRNIC
ARIN
LACNIC 200
200 RIPE
100
150
8-7 8-8 8-9 8-10 8-11 8-12 8-13 8-14 0
Date 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
(b) RTT comparison betw een IPv4 and IPv6 Website
400
IPv6 Fig. 3. RTT distribution in APNIC
Round-trip Time (ms)
300 IPv4
has one more AS compared with IPv4 middle-as. When probe RIPENCC
400
packet begins to enter middle-as, IPv6 RTT is similar to IPv4
RTT. However, when probe packet enters target-domestic-as
from middle-as, the growth of IPv6 RTT is more obvious than 300
that of IPv4 RTT. After probe packet enters target-domestic-as,
the increase of IPv4 RTT is similar to that of IPv6 RTT. 200
Therefore, the path between middle-as and target-domestic-as
causes the difference between IPv6 RTT and IPv4 RTT. That is 100
to say, the connectivity between middle-as and target-
domestic-as in IPv4 network is better than that in IPv6 network,
0
resulting in that IPv4 RTT is smaller than IPv6 RTT for most 0 100 200 300 400 500 600
destination URLs. As shown in Fig. 3, IPv6 RTT of URL {7, IPv6 Round-trip Time
13} is smaller than IPv4 RTT. 1) For URL 7, the length of IPv6
middle-as (i.e., the number of ASes in middle-as) is the same Fig. 4. RTT distribution of IPv4/IPv6 in one polling cycle
as to that of IPv4 middle-as. However, IPv6 RTT makes an
increase of 120 milliseconds in middle-as, while IPv4 RTT We use Traceroute to explore AS-level paths between the
makes an increase of 150 milliseconds in middle-as. This is the probing client and the destination URLs. The total number of
reason for the smaller RTT of IPv6 compared with IPv4. 2) For ASes between the probing client and the destination URLs is
URL 13, the length of IPv6 middle-as is 1, while the length of denoted as AStotal. Note that there are some duplicated ASes in
IPv4 middle-as is 3. The increase of IPv4 RTT in middle-as is AStotal. The number of unique ASes in AStotal is represented as
much larger than IPv6 RTT. Thus, for this destination URL, ASunique. ASdiffer denotes the ratio of ASunique and AStotal, i.e.,
the length of middle-as causes the RTT difference of IPv4 and ASdiffer = ASunique/AStotal. We calculate the ASdiffer of the five
IPv6.
RIRs. The results are 91.7% (APNIC), 66.7% (RIPENCC), around 13 milliseconds except an abnormal point. When the
40% (LACNIC), 27.3% (AFRINIC), and 13.3% (ARIN). link utilization varies between 0.7 and 0.83, RTT begins to
Among the five RIRs, the ASdiffer of APNIC is highest, which fluctuate and generates some high values. Once the link
reveals that paths between the probing client and the utilization is 0.83 or above 0.83, RTT of N2 greatly fluctuates
destination URLs of APNIC are most abundant. This could from 20 to 80 milliseconds. In addition, when the link
explain why RTT distribution of APNIC is most dispersive. utilization exceeds 0.83, most points are distributed from 50 to
The ASdiffer of ARIN is lowest, which could explain that RTT of 60 milliseconds. As depicted in Fig. 8(b), the number of points
ARIN is clustered. in the right of the vertical line accounts for 49.4% of the total
points. RTT of N1 is relatively stable because the maximum
In addition, we conduct a comparison between IPv4 RTT link utilization of L1 is 0.8223, which is smaller than 0.83.
and IPv6 RTT without distinguishing regions. During the one According to the above analysis, we could conclude that,
week observation, the average RTT of IPv6 is 247.71 when the link utilization exceeds a threshold, e.g., 0.83 in our
milliseconds, while the average RTT of IPv4 is 244.48 case study, the RTT fluctuates greatly.
milliseconds. As depicted in Fig. 5, IPv6 RTT is relatively
stable over time, while IPv4 RTT fluctuates greatly. This can (a) RTT variation
be confirmed by the standard deviation, which is calculated 80
based on the average RTT of each hour. The standard deviation N
is 3.3.
40
260
20
ipv4
Round-trip Time (ms)
ipv6 0
250 8-7 8-8 8-9 8-10 8-11 8-12 8-13 8-14
Date
(b) Link utilization variation
1
240
0.8
Link Utilization
0.6
230
8-7 8-8 8-9 8-10 8-11 8-12 8-13 8-14 0.4
Date L
1
0.2
L
2
Fig. 5. RTT comparison between IPv4 and IPv6 0
8-7 8-8 8-9 8-10 8-11 8-12 8-13 8-14
Date
The more obvious fluctuation of IPv4 RTT may be caused
by overutilization of some links in IPv4 networks. In order to
investigate the relationship between delay variation and link Fig. 6. Relationship between RTT and link utilization
utilization, we probe the RTT of two backbone nodes of
CERNET2. These two nodes (N1 and N2) directly connect (a) N1
another node (N3) where we deploy the probing host. The link 24
23
labeled as L2. Then we calculate the link utilization for L1 and
22
L2. Note that the link utilization denotes the ratio between the
average traffic rate (in each five minutes) and the bandwidth of 21
the physical link. For the clarity of description, we calculate 20
and present the average link utilization of each five minutes.
19
As shown in Fig. 6(a) and Fig. 6(b), RTT of N2 shows a 0 0.2 0.4 0.6 0.8 1
Link Utilization
similar pattern with the utilization of L2. RTT of N2 is high and (b) N2
fluctuant when the utilization of L2 is large, while RTT of N2 is 80
low and stable when the utilization of L2 is small. However, not
Round-trip Time (ms)
0.015 AFRNIC
ARIN We utilize weighted moving average to compare the
0.01
LACNIC stability (and predictability) of IPv6 packet-loss rate with IPv4.
RIPE We first group the data into one-hour buckets. And then, we
0.005 compute the average packet-loss rate for each hour-scale
bucket. At last, we compute the moving average error for
0
8-7 8-8 8-9 8-10 8-11 8-12 8-13 8-14
different window sizes. Fig. 10 shows the moving average
Date error of IPv4 and IPv6 packet-loss rates when the window size
-3
x 10 (b) Packet loss comparison betw een IPv4 and IPv6 equals to 2. We find that moving average error of IPv6 packet-
6
IPv6
loss rate is almost below 50%, while moving average error of
IPv4 packet-loss rate is often higher than 50% and fluctuates
Packet Loss Rate
IPv4
4 greatly, which reveals that stability and prediction accuracy of
IPv6 packet-loss rate is much better than that of IPv4.
2
300
IPv4
0 IPv6
APNIC AFRNIC ARIN LACNIC RIPE
Regions 200
Error (%)