Long-Term Traffic Statistics

• "Today's average residential broadband user consumes about 2 gigbytes of data per month, Kafka estimated, which costs the service provider about $1. As downloading feature films becomes more popular, they might consume an average of 9 gigabytes per month, costing carriers $4.50. The average IPTV user will likely consume about 224 gigabytes per month, he added, at a monthly cost to carriers of $112, a giant leap from the less than $5 attributed to Internet use. If that content were high-definition video, the average user would be consuming more than 1 terabyte per month at a cost to carriers of $560 per month." Telephony Online, March 7, 2006
• Projected comparative worldwide network demand (Gb/s) (America's Network, May 15, 1998):

  Year	Data	Voice
  1996	135	948
  1997	273	1,107
  1998	588	1,294
  1999	1,572	1,511
  2000	4,451	1,766
  2001	11,328	2,063
  2002	27,645	2,411

• Internet Still Growing Dramatically Says Internet Founder, August 15, 2001.

  Roberts' data shows traffic has been doubling every six months on average across core IP service providers' networks, or in other words, growing by four times annually. This is an even faster rate of traffic growth than the average of 2.8 times per year carriers had been experiencing across the core since the Internet began growing aggressively in 1997, spurred by mainstream interest in the web.

• A simple numerical comparison of traffic neglects "several important issues. One is that data links are shorter on average than voice links. Another is that utilization rates are much lower for data links than for voice links. Kerry Coffman and I considered this problem in our paper in the October 1998 issue of First Monday, and conluded that if one put in all the appropriate adjustment factors, then the bandwidth of long distance data networks in the U.S. exceeded that of voice networks already by the end of 1997, but that data traffic, measured in bytes, would exceed that of voice networks only around 2002. For the world as a whole, this transition is likely to occur a couple of years later." (Andrw Odlyzko, 1/31/00)

• From: vinton g. cerf (MCI)
  Sent: Monday, November 24, 1997 10:11 PM
  
  Host growth seems to be 85-100% per year; traffic in our backbone is now
  at 100% per year compounded annually.  Others report even higher values. 
  Last year traffic increased by 500%.
  

• Traffic statistics for voice and data; voice traffic is counted at 64 kb/s:

  network	when	volume/month	Gb/s	note
  NSFNET	end of 1994	15 TB/month	0.046	70% of total Internet traffic
  Internet backbone	end of 1994	20 TB/month	0.061	(see above, Andrew Odlyzko)
  AT&T frame relay network	late 1997	5.7 TB/day	0.53
  MCI backbone traffic	Nov. 1997	140 TB/week	1.8
  Total Internet traffic	late 1997	2,000 TB/month	6.1
  U.S. Internet backbone	late 1998	5,000-8,000 TB/month	15.4--24.7
  U.S. Internet backbone	late 2000	20,000-35,000 TB/month
  Other U.S. public data networks	late 1998	1,000 TB/month	3.1
  U.S. private line data	late 1998	4,000-7,000 TB/month	12.3--21.6
  U.S. local calls	1995	2,228 GDEM	271
  U.S. local phone calls	1997	2,683 GDEM	327
  U.S. local calls	1998	2,986 GDEM (119,440 TB/month)	271
  U.S. intrastate toll	1995	344 GDEM	42
  U.S. intrastate toll calls	1997	404 GDEM	49
  U.S. intrastate toll calls	1998	422 GDEM (16,880 TB/month)
  U.S. interstate toll	1995	451 GDEM	55
  U.S. interstate toll	1997	525 GDEM	64
  U.S. interstate toll	1998	555 GDEM (22,200 TB/month)
  U.S. switched access minutes	1996	468.8 GDEM	57.2
  U.S. international outbound	1997	22.6 GDEM	2.8
  U.S. international inbound	1997	9.1 GDEM	1.1
  World telephony	Nov. 1996		600	ITU Direction of Traffic 1996: Trends in international telephone tariffs, based on estimate of 640 m phone lines and 20 minutes/day use.

  GDEM: giga dial equipment minutes, convert to Gb/s by multiplying with 0.122 (60 * 64000 / (86400 * 365)); To convert from TB/month to Gb/s, multiply by 0.003042 (1000 * 8 / (365.25 / 12 * 24 * 3600)).

  Average international call is 3.8 minutes (Rothberg, 1997); Net2Phone call duration is 7.12 minutes. SLAC plots the distribution of their international calls and finds an average duration of 4.5 minutes.

• "The author estimates Netflix alone distributes 1,500 terabytes a day, which is impressive considering the Internet carries 2,000TB a day (by estimates cited in the article)." (NY Times, Sept. 23, 2002)
• "According to information collected by CTIA from U.S. wireless Relevant Products/Services from IBM carriers, total service revenues rose 22.6 percent for the six-month period ending in December 2001, up from $27.8 billion for the year-ago period. Total revenue for 2001 was $65 billion. Data revenue is showing some growth reaching $545 million in 2001, in just the third year of availability. By comparison, CTIA noted, voice revenue after three years (in 1985) was $482 million. Minutes of use during the survey period were up 75 percent from 2000, with consumers racking up more than 456 billion billable minutes in 2001, compared with 258.9 billion minutes in 2000. Larson attributed that increase to strong competition among operators that has resulted in falling prices. The survey shows that estimated subscriber numbers grew by 17 percent, with some 128.4 million wireless subscribers in December 2001, compared with 109 million subscribers at the end of 2000. Some 85 percent, or 101 million, of all U.S. customers now are using digital networks, which is up from the 70 percent digital penetration reported in December 2000." Wireless News, May 2002.
• "[...] estimated that Americans [in early 2000] are averaging 221 minutes per month on their cell phones this year, compared with 145 for Europeans. The FCC said cell phone use averages 204 minutes in Sweden, 186 in the United Kingdom, 182 in France, 137 in Spain, 132 in Finland and 123 in Germany." (FCC, http://www.teledotcom.com/news0800/news080400_2.html">tele.com)
• Unnamed Internet provider:

  week	Tbytes/week	annual growth
  11-27-95	30.462
  11-25-96	92.80	204.6%
  11-16-97	176.49	90.2%

• CAIDA Internet visualization (including Mbone)
• Abilene Traffic Stats, e.g., NYSERnet Buffalo, and NYSERnet NYC
• Cisco list of facts and statistics on traffic volume and other Internet issues
• US-UK link accounting, includes statistics on protocol type distribution. (The hours 01:00-06:00 are not charged; the link is a dual OC-3, i.e., 300 Mb/s.)
• Kazaa/Morpheus/FastTrack traffic percentages for Internet2, December 2001/January 2002 (Joe St. Sauver)
• Oregon Gigapop traffic characterization, Joe St Sauver, University of Oregon, 2001.
• Internet Traffic Flow Size Analysis, before and after Napster
• The size and growth rate of the Internet
• OWEN/NERO Bandwidth Audit, a nice survey of traffic measurements, general Internet architecture
• "Death of the Internet" stories
• Traffic engineering overview
• ProShare video traffic, introduces concepts such as busy hour and CCS
• Traffic computation (Erlang-B, Erlang-C, VoIP)
• World Wide Wait 60 Percent Shorter? (3/11/1998); graphed at Keynote Business 40
• Internet Performance Measurement and Analysis (IPMA) (routing instability)
• Selby Wellman, Cisco: 300% annual growth rate in Internet traffic.
• TTT (tele traffic tapper) for graphical monitoring of Ethernet traffic
• Universities
  • Princeton University
  • U. Toronto
  • UMass (6xT1 connection to the commodity internet and a T3 into the Internet 2. There are ~25K students and staff, and the connection carries traffic for a number of the surrounding colleges as well (Amherst, Smith, Mt. Holyoke, Hampshire), probably with anount ~15K students and staff.)
• A Survey of Internet Statistics / Metrics Activities by Tracie Monk (DynCorp) and K C. Claffy
• NLANR traffic measurements (also measurements)
• NLANR Passive Measurement and Analysis (packet counts, port statistics, packet lengths)

• Date: Fri, 26 May 2000 15:58:49 -0500
  From: "Steve Dispensa" 
  
  Here are some stats from an RMON probe on one of our broadband consumer
  Internet access networks.  It's sitting between our access network and
  our Internet transit connections, so it sees every packet coming from or
  going to the subscribers.  This represents about 7TB worth of data over
  the last few months.
  
  'Upstream' refers to traffic with a source IP address belonging to one
  of our customers' modems and a destination address somewhere else on the
  Internet.  'Downstream' refers to traffic with a source IP address
  belonging to an external Internet host and a destination IP address of
  one of our customers' modems.
  
  We only see DNS requests (<< 512 bytes), URLs (~40 bytes avg.  plus 40
  bytes of overhead in the request message, assuming it's sent as a single
  datagram), and TCP control segments (SYN, ACK, FIN.  .  .) on the
  upstream.  We have a lot of Napster users, too, which helps to pad the
  upstream numbers quite a bit.
  
  Packet Size Distribution
  All numbers are percentages
  
                Downstream      Upstream
                ----------      --------
  0    -  64         14.68         58.49
  65   -  127        13.87         29.73
  128  -  255         7.25          1.72
  256  -  511         6.44          3.98
  512  - 1023        13.59          3.37
  1024 - 1518        44.17          2.70
  
  Steve Dispensa
  Sr. Network Architect
  Sprint Broadband Wireless Group
  

• Simon Leinen, Feb. 2001:

  Those were bytes.  Here's a summary from the same transatlantic links
  (Eastbound direction only) which counts flows and packets too (new
  version of script appended):
  
  protocol.............flows..............packets...............bytes.........
  GRE               7071 ( 0.01 %)    268698 ( 0.02 %)     213346212 ( 0.04 %)
  ICMP           3473563 ( 6.09 %)  10420689 ( 0.94 %)    1083751656 ( 0.20 %)
  IGMP                 4 ( 0.00 %)         8 ( 0.00 %)          7264 ( 0.00 %)
  IP-other         11604 ( 0.02 %)   3724884 ( 0.34 %)     763601220 ( 0.14 %)
  IPINIP            4716 ( 0.01 %)     14148 ( 0.00 %)       2589084 ( 0.00 %)
  TCP-BGP         154665 ( 0.27 %)    154665 ( 0.01 %)      10053225 ( 0.00 %)
  TCP-FTP        1478600 ( 2.59 %)   7393000 ( 0.67 %)    2336188000 ( 0.42 %)
  TCP-FTPD        161850 ( 0.28 %)  69433650 ( 6.26 %)   49783927050 ( 8.96 %)
  TCP-Frag           285 ( 0.00 %)      3420 ( 0.00 %)        413820 ( 0.00 %)
  TCP-NNTP         70222 ( 0.12 %) 113338308 (10.22 %)   21307601904 ( 3.83 %)
  TCP-SMTP        968681 ( 1.70 %)  17436258 ( 1.57 %)    7689389778 ( 1.38 %)
  TCP-Telnet       75043 ( 0.13 %)   1876075 ( 0.17 %)     324560975 ( 0.06 %)
  TCP-WWW       24258155 (42.52 %) 315356015 (28.44 %)  235255587190 (42.34 %)
  TCP-X             6509 ( 0.01 %)   2512474 ( 0.23 %)     293959458 ( 0.05 %)
  TCP-other      7981277 (13.99 %) 415026404 (37.43 %)  215398703676 (38.76 %)
  UDP-DNS        8185178 (14.35 %)  16370356 ( 1.48 %)    2111775924 ( 0.38 %)
  UDP-Frag           444 ( 0.00 %)   1053168 ( 0.09 %)     767759472 ( 0.14 %)
  UDP-NTP        3676906 ( 6.44 %)   3676906 ( 0.33 %)     279444856 ( 0.05 %)
  UDP-TFTP            11 ( 0.00 %)        11 ( 0.00 %)           693 ( 0.00 %)
  UDP-other      6537172 (11.46 %) 130743440 (11.79 %)   18042594720 ( 3.25 %)
  
  So in terms of number of flows and packets, this particular
  transatlantic link does indeed have a higher share of UDP than what
  the 1998 "beast" paper observed.
  
  For a comparison, here are the totals from an access router at a
  random university:
  
  protocol.......flows................packets................bytes.........
  GRE          383 ( 0.00 %)       17235 ( 0.00 %)        3602115 ( 0.00 %)
  ICMP   101931237 ( 1.75 %)   305793711 ( 0.45 %)    37918420164 ( 0.11 %)
  IGMP       34662 ( 0.00 %)      901212 ( 0.00 %)       58578780 ( 0.00 %)
  IP       1406788 ( 0.02 %)    15474668 ( 0.02 %)     3528224304 ( 0.01 %)
  IPINIP      1297 ( 0.00 %)        1297 ( 0.00 %)         583650 ( 0.00 %)
  TCP   4361852662 (74.91 %) 63919315234 (93.39 %) 32455859980970 (96.82 %)
  UDP   1357265629 (23.31 %)  4201556174 ( 6.14 %)  1025284993546 ( 3.06 %)
  
  This matches the CAIDA values much better.  Maybe the transatlantic
  figures are biased because we run authoritative name servers for some
  ccTLDs - those can be expected to generate lots and lots of
  single-packet UDP port 53 flows.
  
  Note also that the "flow" concept used in the CAIDA work isn't exactly
  the same as Cisco NetFlow's, although the numbers may still be
  comparable.  In case someone wants to know, we use the default flow
  timeout values (30 minutes maximum lifetime or 1 minute(??) maximum
  inactivity).
  

• End-to-end service probability
• A discussion of routing needs, from a switch-router vendor's perspective
• FCC international telephone traffic development
• Georgetown University PBX traffic data
• "At some moment in the next 100 days, more than half of all business communications industry-wide will be data--and that's a crossover of landmark dimensions," Dorman [PacBell] said. He pegged national annual growth in data communications among businesses at 30%. source (no longer available)
• "In the summer of last year, Internet traffic from Sweden to the United States was about half the volume of voice traffic between those two countries. By the end of 1996, data and voice traffic volumes on the Sweden-to-U.S. route were equal, and now data volume is double that of voice. Meanwhile, voice still accounts for the vast majority of traffic to neighboring Finland and Denmark." tele.com 9/97
• "In Japan, there is now a total of 350 Mb/s of Internet connectivity to the US. The total fax/voice bandwidth was estimated by a senior KDD official to be 300 Mb/s.", Carl Malamud, note to mbone list, Feb. 10, 1997.
• Bell Atlantic study on the impact of ISPs on local phone service

  PacTel studied some of its telephone switches in detail and found that an average Internet surf was 20.8 minutes long, compared with 3.8 minutes for an average phone call. Ten percent of Internet calls were six hours or longer. To make matters worse, the peak hour for phone systems has now switched to 10 p.m. because of evening Internet use, throwing out the logistics of networks designed around pre- and post-lunch weekday calling peaks. PacTel said a study of one Silicon Valley telephone switch showed 16 percent of call attempts failed during peak evening hours because of Internet traffic, and 2.5 percent of lines used by Internet service companies absorbed 20 to 36 percent of the switch's capacity.

  Reuters, 10/30/96

  Pacific Bell says about 2.3 million people used its lines to dial up their Internet service provider last year, a number it projects will soar to 4.7 million by 2001. Last year, Internet traffic accounted for 27 percent of all residential phone use in its territory; by 2001 it will represent nearly half of all its residential traffic.

  Last year 19 percent of households in its service territory went online, compared to 14.5 percent at the next most affected phone company, Bell Atlantic. In contrast, only 5 percent of US West's customers dialed up to the Internet.

  The average Internet user ties up phone lines 45 minutes each day, compared to 22 minutes for voice callers, the white paper says. Nearly a third of Internet calls last three hours or more and 7 percent last more than 24 hours per session, in part because local calls to Internet service providers are free.

  As of January [1997], 62 of the 227 Pacific Bell offices that steer calls to Internet service providers were mildly or severely overloaded, the white paper says. That's nearly triple the number of offices that were congested only three months earlier. Typical symptoms are delays in getting dial tone and a rapid busy signal.

  SF Chronicle
• The Effect of Internet Use on the Nation's Telephone Network, a rebuttal
• Traffic engineering table, Erlang-B
• Bell Labs Internet Traffic Research
• Newsletter with traffic information and list of most popular web sites
• "According to the International Recording Media Association, 4.4 billion CDs will be manufactured in 2001, while the Recording Industry Association of America says almost 30,000 new CD titles will be released next year." (While a new music economy waits to be born, how fast will the sun set on the compact disc?, 12/10/00)
• "As near as I can tell, most large well-run ISPs have roughly (Trans-Pacific RTT ms * max interface line rate) of packet buffer on each core router interface, in order to handle transient conditions." (Ran Atkinson, end2end mailing lists)

Last updated by Henning Schulzrinne