Internet2 NetFlow: Weekly Reports: Week of 20081222

  1. Introduction
  2. Bulk TCP
  3. Full Data Set

Introduction

You are looking at the weekly Abilene network usage report for the week of 20081222 produced from NetFlow records. The view of the whole network as a single traffic-relaying unit is presented. More formally, data from all interior circuits (those connecting two Abilene routers) were discarded while all the rest of the data were merged to create this view.

During this week, there were no missing data days.

The data are split into two sections: bulk TCP data and the full data set. A "bulk TCP" flow is defined as a TCP flow that transferred more than 10MB of data. The first section only concerns these data. The second section studies the overall traffic composition.

All the numbers in this report are hyperlinked to plots that show their history (e.g., clicking on the percentage of octets of NNTP traffic will bring up a time-series plot that shows the history of this parameter).

Bulk TCP

During this week, bulk TCP traffic comprised 49.81% of octets and 28.14% of packets of the full data set traffic.

The distribution of bulk TCP throughputs is the most important piece of data in this report. Cumulative distribution function plots (1-CDF vs. throughput in bits/second) in semi-log and log-log scales are as follows:
[Bulk TCP throughputs (semi-log scale).] [Bulk TCP throughputs (log-log scale).]

Distribution of the amount of data transferred (in semi-log and log-log scale, 1-CDF vs. total trasfer size in octets) is presented below. It should be recognized that NetFlow collection mechanism is always configured so that flows (in the accounting sense) cannot last longer than a certain period of time. Therefore, the distribution of transfer sizes is to a certain extent skewed in the upper part.
[Bulk TCP transfer sizes (semi-log scale)] [Bulk TCP transfer sizes (log-log scale).]

The distribution of durations of bulk TCP flows (in seconds) is as follows (you may notice the cut-off phenomenon mentioned above):

[Bulk TCP durations distribution.]

The following table shows actual values from the above distribution plots that correspond to characteristic values (such as median, 90%, max, etc.).

Table 1. Selected Points from Distribution Graphs (Bulk TCPs)

Percentile Throughput (b/s) Durations (s) Size (octets)
1 1.394M 1 10.08M
5 1.503M 7 10.50M
10 1.639M 14 11.10M
50 3.554M 58 18.74M
90 17.80M 59 59.78M
95 31.20M 59 90.45M
99 88.00M 59 200.1M
99.9 248.1M 59 601.9M
99.99 987.5M 59 3.636G
99.999 2.098G 60 7.065G
100 98.92G 64 38.21G

We compute average packet size of each flow by dividing the number of octets in a flow by the number of packets. Distribution of average sizes of packets belonging to bulk TCP flows is as follows:

Table 2. Packet Sizes (Bulk TCP)

Packet Size Packets
Small (<100B)1.22% 2.547G
Medium (100-1400B)8.03% 16.71G
Large (1401-1500B)90.11% 187.5G
Jumbo (>1500B)0.63% 1.319G
Total100.00% 208.1G

We show what applications transfer large amounts of data in the following table. Note that this is bulk TCP traffic only; full data set usage is presented in the next section.

Table 3. Aggregated Application Types (Bulk TCP)

Traffic Type OctetsPacketsFlows
Data Transfers27.88% 86.44T 28.74% 59.83G 36.75% 3.475M
Encrypted Traffic8.55% 26.51T 9.54% 19.85G 7.70% 727.9k
Measurement5.38% 16.67T 2.74% 5.712G 0.79% 74.90k
Advanced Apps4.60% 14.25T 4.72% 9.825G 6.79% 642.1k
File Sharing3.42% 10.61T 3.48% 7.250G 2.46% 232.4k
Misc0.54% 1.666T 0.60% 1.258G 1.00% 94.69k
Games0.18% 564.0G 0.19% 401.4M 0.21% 19.56k
Audio/Video0.07% 212.9G 0.08% 156.1M 0.16% 14.70k
Unidentified49.39% 153.1T 49.90% 103.8G 44.15% 4.176M
Total100.00% 310.0T 100.00% 208.1G 100.00% 9.458M

The following are the fastest 10 measurement flows with unique source and destination AS numbers (i.e., for any given pair of source and destination AS numbers, no more than one fastest flow is shown).

Table 4. Fastest Bulk TCP Measurement Flows with Unique AS Source and Destination

Throughput (b/s)Packet size (bytes)Duration (s)Src ASDest ASApplication type
7.401G900022SDSC [195]Abilene [11537]Iperf
2.265G900029DFN-IP service G-WiN [680]INDIANAGIGAPOP [19782]Iperf
1.045G900011ESNET [3428]Abilene [11537]Iperf
1.041G824410ESnet-West [292]Abilene [11537]Iperf
829.0M150012Brookhaven National Lab [43]Unknown [32361]Iperf
685.5M150017U Chicago [160]Unknown [32361]Iperf
408.3M150046GEORGE-MASON-UNIV [11279]Unknown [32361]Iperf
401.7M150022U Chicago [160]ESnet-East [291]Iperf
380.9M150025GEORGE-MASON-UNIV [11279]U Chicago [160]Iperf
236.4M150012U Chicago [160]Brookhaven National Lab [43]Iperf

The following are the fastest 10 non-measurement flows with unique source and destination AS numbers (i.e., for any given pair of source and destination AS numbers, no more than one fastest flow is shown). When unable to determine the application type, we give the source and destination port numbers.

Table 5. Fastest Bulk TCP Non-measurement Flows with Unique AS Source and Destination

Throughput (b/s)Packet size (bytes)Duration (s)Src ASDest ASApplication type
1.025G900020INDIANAGIGAPOP [19782]ORNL [50]60478 -> 5150
526.0M150060Unknown [36375]Boston U [111]54321 -> 39580
405.4M150015Brookhaven National Lab [43]Boston U [111]39506 -> 37009
399.2M150013JPL [127]Oregon State U [4201]Hotline
346.9M150011Network for Education and Research in Oregon [3701]Unknown [0]Rsync
320.4M150050Nat Lib Med [70]NCREN [81]50026 -> 47278
313.8M150011NOAA [6629]NIST-BOULDER [2648]FTP
310.9M150060Network for Education and Research in Oregon [3701]UT-Austin [18]Rsync
286.6M150013Boston U [111]Brookhaven National Lab [43]43217 -> 23420
273.2M142013Unknown [40127]U Minnesota [217]HTTP

We also compute the average concurrency of bulk TCP flows for the week (by adding durations of all captured flows and dividing the result by the by the duration of the week). This week's average number of concurrent bulk TCP flows: 745.0.

Full Data Set

In addition to bulk TCP flows data, we provide statistics that characterize the overall composition of the complete data set (everything that transited the Abilene network this week).

The following table describes what kinds of traffic went through the network (multiple applications are aggregated into classes):

Table 6. Aggregated Application Types (Full Data Set)

Type OctetsPackets
Data Transfers35.48% 220.9T 35.47% 262.3G
Encrypted Traffic7.36% 45.80T 8.42% 62.27G
Advanced Apps3.69% 22.94T 3.55% 26.23G
Measurement2.97% 18.47T 1.46% 10.81G
File Sharing2.59% 16.13T 2.31% 17.09G
Misc2.20% 13.69T 5.23% 38.70G
Audio/Video1.25% 7.796T 1.12% 8.315G
Games0.31% 1.923T 0.44% 3.281G
Unidentified44.15% 274.8T 42.00% 310.6G
Total100.00% 622.5T 100.00% 739.7G

This table is available additionally in the following more verbose version (no applications are aggregated into classes, but class composition is shown):

Table 7. Detailed Application Types (Full Data Set)

Traffic type OctetsPackets
Data Transfers
HTTP
FTP
Rsync
NNTP
---
30.95%
1.84%
1.76%
0.94%
---
192.6T
11.48T
10.93T
5.831T
---
31.43%
1.51%
1.61%
0.92%
---
232.4G
11.14G
11.93G
6.795G
Encrypted Traffic
SSH
HTTPS
IPsec ESP
IPsec AH
IPsec IKE
---
4.27%
2.57%
0.50%
0.01%
0.00%
---
26.61T
16.02T
3.083T
78.46G
7.483G
---
4.67%
3.11%
0.61%
0.02%
0.00%
---
34.54G
22.98G
4.531G
181.4M
34.09M
Advanced Apps
UNIDATA LDM
McIDAS
BBCP
BBFTP
GsiFTP
IBP
---
3.31%
0.23%
0.11%
0.03%
0.01%
0.00%
---
20.62T
1.447T
677.8G
158.8G
32.44G
3.691G
---
3.26%
0.17%
0.08%
0.03%
0.01%
0.00%
---
24.09G
1.243G
582.7M
204.5M
79.08M
31.30M
Measurement
Iperf
ICMP
IPMP
---
2.92%
0.05%
0.00%
---
18.15T
315.4G
0.000
---
1.11%
0.35%
0.00%
---
8.218G
2.599G
0.000
File Sharing
Audiogalaxy
Hotline
Shoutcast
BitTorrent
eDonkey2000
FastTrack
Gnutella
WinMX
Freenet
Carracho
Blubster
Neo-Modus
Direct Connect++
---
1.23%
0.77%
0.27%
0.21%
0.09%
0.01%
0.01%
0.01%
0.00%
0.00%
0.00%
0.00%
0.00%
---
7.649T
4.778T
1.709T
1.302T
536.9G
56.05G
53.37G
41.05G
4.951G
4.479G
1.233G
21.55M
5.751M
---
0.97%
0.60%
0.39%
0.23%
0.09%
0.01%
0.01%
0.01%
0.00%
0.00%
0.00%
0.00%
0.00%
---
7.159G
4.407G
2.910G
1.730G
647.0M
65.87M
97.70M
44.87M
5.538M
8.656M
13.79M
120.1k
39.80k
Misc
Mail
Port 0
Squid
DNS
AFS
X11
MS Windows
IRC
RTIP
NTP
Telnet
NFS
SNMP
SOCKS
IDENT
AOL AIM
RPC Portmapper
---
1.30%
0.28%
0.24%
0.22%
0.06%
0.03%
0.01%
0.01%
0.01%
0.01%
0.01%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
---
8.102T
1.760T
1.504T
1.366T
368.9G
211.1G
80.09G
68.10G
61.39G
59.63G
47.63G
25.78G
15.86G
14.42G
1.996G
1.828G
794.9M
---
2.58%
0.22%
0.35%
1.42%
0.08%
0.08%
0.17%
0.04%
0.08%
0.10%
0.07%
0.01%
0.02%
0.01%
0.00%
0.00%
0.00%
---
19.10G
1.599G
2.618G
10.52G
617.5M
577.6M
1.224G
331.4M
586.5M
764.2M
481.8M
46.40M
132.4M
46.43M
32.08M
3.318M
11.80M
Audio/Video
Any-Source Multicast
Real Player
Windows Media
H.323 Signaling
Backbone Radio
StreamWorks
Camarades webcams
Subset of VoIP
Single-Source Multicast
---
0.96%
0.26%
0.02%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
---
5.965T
1.643T
117.8G
27.56G
21.27G
10.86G
6.077G
3.800G
10.84M
---
0.73%
0.36%
0.02%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
---
5.416G
2.634G
156.7M
34.76M
33.44M
18.04M
10.47M
10.43M
8.000k
Games
DirectX
Battlenet
Half-Life
Spy Arcade
Quake
Starsiege Tribes
Asheron
---
0.18%
0.07%
0.03%
0.02%
0.01%
0.00%
0.00%
---
1.135T
433.0G
162.9G
104.1G
62.69G
13.43G
12.13G
---
0.21%
0.09%
0.11%
0.01%
0.02%
0.00%
0.00%
---
1.536G
681.4M
801.2M
104.4M
116.0M
21.41M
20.48M
Unidentified
Unidentified
---
44.15%
---
274.8T
---
42.00%
---
310.6G
Total
Total
---
100.00%
---
622.5T
---
100.00%
---
739.7G

The following table summarizes use of most popular IPv4 protocols:

Table 8. IP Protocols Distribution (Full Data set)

Protocols OctetsPackets
ICMP[1]0.05% 315.4G 0.35% 2.599G
IGMP[2]0.00% 42.39M 0.00% 1.236M
IP-ENCAP[4]0.01% 34.91G 0.01% 38.99M
TCP[6]91.30% 568.3T 86.75% 641.7G
UDP[17]6.71% 41.79T 11.13% 82.37G
IPv6[41]0.03% 163.5G 0.03% 228.9M
GRE[47]1.40% 8.714T 1.08% 8.008G
ESP[50]0.50% 3.083T 0.61% 4.531G
AX.25[93]0.00% 33.00k 0.00% 500.0
PIM[103]0.00% 4.112G 0.00% 34.55M
IPMP[169]0.00% 0.000 0.00% 0.000
Other0.01% 78.51G 0.02% 182.0M
Total100.00% 622.5T 100.00% 739.7G

We compute average packet size of each flow by dividing the number of octets in a flow by the number of packets. Distribution of (average) packet sizes is as follows:

Table 9. Packet Sizes (Full Data Set)

Packet Size Packets
Small (<100B)36.41% 269.3G
Medium (100-1400B)19.28% 142.6G
Large (1401-1500B)43.88% 324.5G
Jumbo (>1500B)0.43% 3.211G
Total100.00% 739.7G

We only track DSCP values for which special treatment was defined by Internet2 QoS working group (and the default of DSCP=0):

Table 10. Important DSCP Values (Full Data Set)

Type OctetsPackets
Best effort [DSCP=0]96.66% 601.7T 96.89% 716.7G
Scavenger [DSCP=8]0.07% 458.7G 0.08% 609.4M
EF [DSCP=46]0.00% 17.79G 0.01% 67.45M
Other3.26% 20.32T 3.02% 22.35G
Total100.00% 622.5T 100.00% 739.7G

We collect statistics about ECN-capable traffic:

Table 11. ECN-Capable Traffic

Type OctetsPackets
ECN-Capable0.45% 2.774T 0.28% 2.046G

To facilitate detection of emerging applications, we present statistics about frequently encountered unidentified port numbers (no distinction is made in this table between TCP and UDP):

Table 12. Frequent Unidentified Ports

Port OctetsPackets
600111.30% 8.116T 1.17% 8.652G
21280.77% 4.778T 0.80% 5.950G
200000.72% 4.500T 0.60% 4.454G
19350.63% 3.924T 1.02% 7.565G
150000.44% 2.756T 0.50% 3.673G