Geoff Huston

Traffic in Network

Geoff Huston
0 You have liked this article 0 times.
0 has been assigned by IANA to APNIC on 19 January 2010, for use for as public unicast space for further address allocations and assignments. Before starting to allocate and assign from this range, APNIC has undertaken a study into

Authors of this report are Geoff Huston and George Michaelson, APNIC. The report was originally published here:



The address plan for IPv4 has a reservation for “Private Use” address space. This reservation, comprising of 3 distinct address blocks, namely, and, is intended for use in private contexts where the devices that are addressed in this manner do not have any requirement to be visible to the public Internet. However, it has long been recognised that other addresses have also been used in private contexts, Some of these uses are entirely informal and local, while other uses have been a little more systematic. One recent study of this form noted that the address block of network 1, or, was “ widely used as private address space in large organizations whose needs exceed those provided for by RFC 1918” [1]. has been assigned by IANA to APNIC on 19 January 2010, for use for as public unicast space for further address allocations and assignments. Before APNIC commences  allocations and assignments, APNIC has undertaken a study into The particular question under investigation here is the extent to which addresses in network are an “attractor” for unwanted traffic. In particular, is there a significant level of “leakage” of supposedly private use traffic directed to addresses in that “leak” into the public Internet?

While network was an unallocated and unadvertised network any such traffic directed to an address in this block that “leaked” into the public Internet would follow a “default” routing path to the point where there was a “default-free” routing element, where the traffic would be discarded.  As soon as any such address in was advertised as reachable into the public Internet, instead of being discarded as the boundary of the default-free zone (DFZ), the packets would be passed onto the destination rather than being discarded.

On 27 January 2010 the RIPE NCC, in collaboration with APNIC, announced 4 prefixes in, namely,, and They noticed an immediate effect in terms of a jump in traffic to the announcement point that saturated the 10Mbps port (Figure 1) [2]. The announcements of and were dropped by the RIPE NCC on 2 February, and the announcements of and were dropped on 9 March 2010. The RIPE NCC performed some packet capture of the traffic being sent to these prefixes, and their results are described in [2].


Figure 1 – Traffic load at the RIPE NCC announcement point of more specifics [2].



It appears that there is a significant amount of traffic that is being directed to addresses in network 1. This traffic is likely to be a combination of leakage of traffic from private use domains, potential “leakage” from mis-configured equipment and a certain amount of scanning activity. Following the work undertaken by the RIPE NCC, we commenced further investigation of the traffic in network 1 in February 2010.

The primary objective of this work was to quantify the extent to which networks in attract “pollution” or “unwanted” traffic.

A related objective was to investigate the distribution of traffic directed to addresses in to determine if there are particular addresses that are “hot spots” for attracting unwanted traffic, and quantify the additional traffic that such addresses attract.

It is also the intent to use this study to inform potential recipients of addresses in this address block of the anticipated profile of unwanted traffic that may be directed to such addresses.

In February 2010 we solicited assistance from potential collaborators to perform a comprehensive packet capture for the entirety of Collaborative experiments have been undertaken with Merit and with YouTube as an outcome of this exercise, and this report summarizes the initial findings of these studies. We would like to acknowledge their assistance here, as their generous support and flexible responses to our requirements have been vital in assembling this report.


AS237 Announcement of

AS237 (Merit) announced network from 22 February 2010 until 1 March 2010. A single system was used as the packet capture device and the configuration was entirely passive. The system did not respond to any packets that it received in The system performed a full packet capture of all packets received.

The analysis reported here is based on the 6 day period from 0000 23 February 2010 UTC -6 through to 2400 28 February 2010 UTC -6.

Traffic Profile

Figure 2 shows the traffic received by the collection point for this 96 hour period on a second-by-second basis.


Figure 2 – Traffic received by AS237

The traffic logs show that the traffic sent to is a relatively steady 160Mbps for the period. There is a slight element of a 24 hour diurnal cycle in the data, but it is not a pronounced cycle in terms of total received traffic levels.

There is a regular interval of reduced traffic in these logs, evidently corresponding to the file cycle interval for the packet capture system. Other incidents of short term reduced traffic incidents to be based around packet loss by the packet capture system.

There are short bursts of between 1 and 30 seconds of elevated traffic levels  There are 20 or so incidents of  burst traffic levels of between 200Mbps and 300Mbps. There is a 3 second isolated burst at 860Mbps and a 10 second burst at 750Mbps in this period.

The profile of traffic in each protocol is shown in the following collection of figures.


Figure 3 – UDP Traffic received by AS237



Figure 4 – TCP Traffic received by AS237



Figure 5 – ICMP Traffic received by AS237



Figure 6 – Other Traffic received by AS237


Packet  Profile

Figure 7 shows the received packet rate for the period. Of note is the pronounced diurnal pattern in the data, particularly in the UDP data. The peaks are not as pronounced, indicating that the peaks in traffic rate are due to a burst of larger UDP packets.


Figure 7 – Received Packet rate

In terms of protocol distribution as measured by bytes of traffic the distribution according to protocol is shown in Table I.


Protocol Proportion of Traffic Proportion of Packets
UDP 88.1% 76.9%
TCP 9.8% 19.8%
ICMP 1.6% 2.5%
Other 0.5% 0.7%

Table I – Distribution of traffic by Protocol

The distribution of packet sizes by protocol is shown in Figure 8. TCP packet sizes were consistently in the range 69-70 bytes, indicative of a TCP SYN packet header of 16 bytes of framing, 20 of IP, 20 of TCP, and 13-14 bytes of TCP options.


Figure 8 – Distribution of Packet Sizes by Protocol


Distribution of Traffic across Network 1 - /16s

The traffic in network 1 is not evenly distributed across all destinations. Figure 9 shows the distribution of traffic within 1/8 divided up into address blocks of a /16 in size. Note that this is a log scale of traffic levels, and the levels reported here are an average for the entire collection period.


Figure 9 – Traffic levels per /16

Figure 10 provides an average of the traffic across the 6 day period. The figure shows both the average level of traffic and the peak 60 level of traffic for each /16.  The RIPE NCC advertisements of


Figure 10 – Traffic levels per /16

A similar picture can be drawn of the incoming packet rate to the /16s in network 1. It appears from a comparison of the two data series that the elevated traffic levels in the low numbered address block are due to certain addresses receiving larger packets, as distinct from simply seeing elevated traffic volumes.


Figure 11 – Packet levels per /16

It appears that from this data there are significantly higher traffic levels sent to addresses in,,

Those /16s with an average traffic level of more than 150Kbps for the 6 day period are listed here in Table II.

/16 Address Prefix Average Traffic 60 second peak 79,981 kbps 353,839 kbps 12,564 kbps 45,752 kbps 12,010 kbps 127,997 kbps 8,816 kbps 32,038 kbps 3,320 kbps 8,668 kbps 261 kbps 2,388 kbps 258 kbps 2,519 kbps 238 kbps 3,650 kbps 212 kbps 2,816 kbps 188 kbps 3,351 kbps 176 kbps 1,805 kbps 160 kbps 788 kbps 153 kbps 2012 kbps 151 kbps 1429 kbps

Table II – Distribution of traffic by /16s


A “Control Point” Measurement

Is this level of traffic being passed into network “normal”? What’s a “normal” expectation in terms of traffic?

To provide answer to this question we advertised a smaller prefix, namely, under similar conditions to the advertisement of This prefix was drawn from the unallocated address pool managed by APNIC. This advertisement of this address block was undertaken in collaboration with  

YouTube, and was advertised by AS36351 on 19 March 2010. The traffic levels attracted by this advertisement was collected by 4 passive collectors  in a load-sharing configuration. The aggregate traffic levels, per /16 is shown for the 24 hour period. It is evident that there is a distinct correlation across the /16s of an average traffic level of 10Kbps. This is shown in Figure 13.


Figure 12 – traffic levels per /16



Figure 13 – average traffic levels per /16


Distribution of Traffic  by /24s

A similar analysis can be performed at the granularity of /24s. Figure 14 shows the profile of traffic to each /24 in the address block


Figure 14 – Traffic levels per /24 in

One /24 within this address block dominates all others, namely The traffic profile for this particular /24 is shown in Figure 15. There is a steady load of between 80 to 100Mbps of traffic directed to this prefix. With the predominate volume being directed to the single address


Figure 15 – Traffic levels for

In order to provide an overview of the traffic sent into each of the /24s in, Figure 16 shows the distribution of traffic levels by number of /24s. divided into “bins” of 50bps increments. Together with a cumulative count.


Figure 16 – Distribution of traffic in per /24

The distribution has some similarity to an exponential decay function (which would be the case if the log of the distribution was linear). Figure 17 shows the same data presented using a log scale for the count of /24’s in each average traffic level bin.


Figure 17 – Distribution of traffic in per /24 (log scale)

In this distribution 98% of the /24s, or 64,252 /24s received an average traffic load less than 132 bytes per second (or an average of 1 packets per second).  The following table shows the distribution of /24s by average packet rates, using an average packet size of 132 bytes (as measured in the packet capture for

Average Pkt per sec. Average bps Number of /24s Cumulative Total % of total
<= 1 <= 1056bps 64252 64252 98.0%
1 .. 2 <= 2112bps 916 65168 99.4%
2 .. 3 <= 3168bps 195 65303 99.6%
>= 3 > 3168bps 233 65536 100%

Table III – Distribution of traffic by /24s


Control Point Comparison

A similar exercise has been undertaken for traffic in The distribution of traffic per /24 is shown in Figures 18 and 19.


Figure 18 – Distribution of traffic in per /24


Figure 19 – Distribution of traffic in per /24 (log scale) 

Average Pkt per sec. Average bps Number of /24s Cumulative Total % of total
<= 1 <= 776bps 4092 4092 99.9%
1 .. 2 <= 1552bps 1 4093 99.9%
2 .. 3 <= 2328bps 1 4094 99.9%
3 .. 4 <= 3104bps 1 4095 99.9%
>= 3 > 3104bps 1 4096 100%

Table IV – Distribution of traffic by /24s in

In this case 4087 /24s (or 99.8% of the /24s) received less than 258 bits per second, or the equivalent of 1 packet every 3 seconds (the average packet size for the packet capture was 97 bytes)


Correlation of the Data

The experiment of passive listening of incoming packets addressed to network was repeated for a further 4 hours on 21 March 2010. This was undertaken by YouTube, and the prefix was originated by AS36351, with incoming traffic collected using four systems and a load balancing front end. The profile of traffic gathered in this interval is shown in Figure 20. (The data anomaly at 22:40 was the result of the restarting of the packet collection system on one of the four data collection systems.)


Figure 20 - traffic to AS36351

The profile of traffic for this 6 hour period is comparable to the profile of the original AS237 announcement. The traffic level is some 150Mbps. With the predominate volume comprising of UDP packets. The measurement of traffic in the busiest /24,, is similar to the profile gathered in AS237, shown in Figure 21.


Figure 21 - traffic to AS36351

The profile of traffic distribution per /16 is shown in Figure 22. Compared to Figure 9, there is a comparable traffic pattern, with 5 /16’s showing incoming traffic levels higher than 1Mbps, and all other /16s showing traffic levels of less than 100Kbps (Figure 23).


Figure 22 – Traffic levels per /16 – AS35361 


Figure 23 – Traffic levels per /16 – AS35361

Those /16s with average traffic of more than 150Kbps for the period are listed in Table V.

/16 Address Prefix Average Traffic
Average Traffic
(AS237) 86,757 kbps 79,981 kbps 19,714 kbps 12,564 kbps 10,241 kbps 8,816 kbps 3,656 kbps 3,320 kbps 3,611 kbps 12,010 kbps 452 kbps 238 kbps 390 kbps 115 kbps 337 kbps 87 kbps 206 kbps 92 kbps 192 kbps 160 kbps 178 kbps 145 kbps 169 kbps 117 kbps 168 kbps 261 kbps 151 kbps 92 kbps

Table V – Distribution of traffic by /16s - /16s with > 150kbps measured by AS36351

A comparison of the /16s with average traffic levels of more than 150kbps as measured in the extended AS237 experiment with average traffic level measured in the shorter AS35361 experiment is shown in Table VI. The set of the 5 highest /16s (,, and are common to both data sets. Of the remainder, and are common.

/16 Address Prefix Average Traffic
Average Traffic
(AS35361) 79,981 kbps 86,757 kbps 12,564 kbps 19,714 kbps 12,010 kbps 3,611 kbps 8,816 kbps 10,241 kbps 3,320 kbps 3,656 kbps 261 kbps 168 kbps 258 kbps 143 kbps 238 kbps 453 kbps 212 kbps 118 kbps 188 kbps 106 kbps 176 kbps 96 kbps 160 kbps 192 kbps 153 kbps 100 kbps 151 kbps 105 kbps

Table VI – Distribution of traffic by /16s – comparison of AS237 and AS36351 /16s

A comparison of the two data sets is shown in Figure 24.


Figure 24 – Traffic levels per /16 – Comparison of AS237 with AS35361

The data from this second data collection experiment correlates well with the extended data set gathered from AS237.  There is a good correlation between the two data sets for the overall majority of the traffic. Of the 150Mbps of the incoming traffic directed to, 122bps, or some 80% of the traffic is directed to addresses in 5 of these /16s, namely, ,, and There is not as close correlation with traffic levels in the other /16s. This could be due to the short time period of the AS35361 experiment (4 hours) compared with the 6 day AS237 data collection experiment, with the consequent limitation that the AS35361 data was not exposed to a complete 24 hour traffic cycle.


Traffic Directed to

Network currently attracts an average of some 140Mbps - 160Mbps of incoming traffic, as a continuous sustained traffic level.

Traffic appears to be a combination of scans that pass across part or the entirety of the addresses in this block and streams of UDP traffic addresses to particular individual addresses in the block.

The traffic in is not evenly distributed. The majority of the traffic is directed at, and the covering /24, receives some 90 – 100Mbps of traffic as a continuous load, with isolated peaks of 1 second intervals in excess of 800Mbps.

Using a control point of to establish a “normal” background traffic level, it appears that 90% of the /24s in receive less than 1 average packet per second of incoming traffic (the 90% percentile value of the traffic distribution per /24 is at the level of 1 average sized packet every 3 seconds).

No /24 in has an average traffic level that is less than the average /24 load observed in

Using a benchmark threshold level of 1 average-sized packet per second, or 1.056 Kbps, 98.8% of all /24s in receive less than this threshold, and 1,344 /24s that receive a higher rate.

There are 428 /24s that receive more than 2.112kbps of incoming traffic, or more than 2 average-sized packets per second.

There are 233 /24s that receive more than 3,168kbps of incoming traffic, or more than 3 average-sized packets per second.  These /24s are located within 14 /16s within This list of /24s that received in excess of 3 average-sized packets per second, and their enclosing /16s, are provided in Appendix I of this report.



  1. It is recommended that the following /24s be withheld from general allocation by APNIC, on the basis that each of these /24s individual attract more than 1 Mbps of unsolicited traffic:
             8Mbps        79 Mbps       11 Mbps       12 Mbps     3 Mbps

    These /24s should be marked as allocated to APNIC R&D to allow further experimentation in the long term nature of unsolicited background traffic to be conducted by APNIC in collaboration with interested researchers and the operational community.

If further investigation reveals that the traffic to any of these /24s abates to a normal background level in the future, or if there is a viable form of mediation that makes any of these network prefixes useable in a general fashion on the public Internet, then these addresses would be returned to the APNIC unallocated address pool at that time.

It is recommended that the following /16s be temporarily marked as reserved and withheld from general allocation by APNIC:

These /16s should be marked as allocated to APNIC R&D to allow further short term experimentation in the distribution of unsolicited background traffic to these addresses to be conducted by APNIC in collaboration with interested researchers and the operational community. The experimentation would be conducted in the period April – October 2010, and a followup report on recommended longer term reservations from this set of addresses be provided at the end of this period.



This work has benefited greatly from the generous assistance from our collaborators in this exercise. We wish to thank the folk from the RIPE NCC, Merit, AARNet, YouTube and Google for both their patience and enthusiasm to provide us with the necessary facilities to undertake this work. Particular thanks are due to Nathan Hickson of YouTube and Steve Padgett of Google for their generous assistance.



[1]     Leo Vegoda, “Awkward /8 Assignments”, Internet Protocol Journal, September 2007. ( )

[2]     “Pollution in 1/8”, RIPE Labs Report, RIPE NCC, 3 February 2010. ( )


Appendix I

Top 233 /24s – Ordered by Address

The following is a “map” of the /24s in with the nominated high traffic points of more than 3 packets per second per /24 highlighted.



Authors of this report are Geoff Huston and George Michaelson, APNIC. The report has originally be published here:

0 You have liked this article 0 times.

You may also like

View more

About the author

Geoff Huston AM is the Chief Scientist at APNIC, where he undertakes research on topics associated with Internet infrastructure, IP technologies, and address distribution policies. From 1995 to 2005, Geoff was the Chief Internet Scientist at Telstra, where he provided a leading role in the construction and further development of Telstra's Internet service offerings, both in Australia and as part of Telstra's global operations. Prior to Telstra, Mr Huston worked at the Australian National University, where he led the initial construction of the Internet in Australia in the late 1980s as the Technical Manager of the Australian Academic and Research Network. He has authored a number of books dealing with IP technology, as well as numerous papers and columns. He was a member of the Internet Architecture Board from 1999 until 2005 and served as its Executive Director from 2001 to 2005. He is an active member of the Internet Engineering Task Force, where he currently chairs two Working Groups. He served on the Board of Trustees of the Internet Society from 1992 until 2001 and served a term as Chair of the Board in 1999. He has served on the Board of the Public Internet Registry and also on the Executive Council of APNIC. He chaired the Internet Engineering and Planning Group from 1992 until 2005.

Comments 0