|
At the opening of NANOG 53, Kevin McElearney of Comcast commented that within Comcast Regional Area Networks (CRANs), the company is regularly pushing 40+ Gbps of traffic out to the global Internet. This is a massive amount of traffic and in many cases, it’s more traffic than entire countries around the world push out to the Internet.
It got me thinking about just how much traffic there is on the Internet and the rate at which that traffic will grow over the coming years.
Around the world, ISPs and networks interconnect with each other at specially designed data centers called Internet Exchange points (IXes). An IX is a large data center with access to multiple fiber optic pathways to the surrounding area, a large layer 2 peering switch (typically Ethernet-based) and a number of ISPs and networks that want to exchange traffic with one another.
One of the world’s most popular Internet exchanges—MAE-East—was located in Virginia and served as the East Coast hub for Internet traffic back in the day. While I was at NANOG, I had drinks with Steve Feldman, who was sharing some stories of his experience working at MAE-East. Steve was commenting on some of the upgrades that were being performed at MAE-East, saying “When we installed the Gigaswitches in the exchange, they had 100 megabit per second (Mbps) FDDI ports. We were significantly future proofing ourselves. This would be plenty of capacity for a very long time.”
So this got me thinking about exactly how much traffic can we anticipate on the Internet in the next 20 years? Knowing that Internet traffic grows at a rate similar to or ahead of Moore’s Law, I assumed that traffic at the East Coast IX was approximately 100 Mbps in 1996 and ran the model out through the year 2030.
Yes, that’s right: macro scale growth of a single 100 Mbps Internet Exchange serving the US East Coast to over 650+ Tbps of traffic alone!
Checking this data indicates that there is approximately 102 Gbps of traffic being exchanged on the US East Coast across the facilities that have replaced the MAE-East exchange. I believe this number to be approximately correct. Given that and the projection of Moore’s Law, this means we can expect to be exchanging traffic volumes of over 650 Tbps in the year 2030.
There are some exchanges around the world today that are pushing over 1 Tbps in traffic and even this amount of traffic pushes the limits of the capabilities of today’s networking hardware. We have quite a bit of work to do if we’re going to be ready to handle 650+ Tbps of peering traffic!
Sponsored byVerisign
Sponsored byWhoisXML API
Sponsored byVerisign
Sponsored byCSC
Sponsored byIPv4.Global
Sponsored byDNIB.com
Sponsored byRadix
Is the US walking behind or something? :)
Just checking http://www.ams-ix.net/statistics/ that is an average of 750Gigs/s in+out (thus depends if one combines that for more effect); that would thus be a factor 20 of Comcasts’s “CRAN” things or 7 times more than what MAE-East does. And note that those are averages over a year, the peak was double that, in both directions and of course that excludes all the private/direct peering which does not touch that platform. I would not be surprised that that would explode those numbers even more.
Though, of course, AMS-IX carries quite a few more traffic than just for the inhabitants of .nl and caters for quite a bit of the rest of Europe too.
Anyway, with lots of kidding aside as these views are very limited and does not disclose even 5% of the traffic, maybe….
Could you extrapolate the data from AMS-IX, as it would be interesting to take that into your calculation. Traffic levels back to 2007 are published here: http://www.ams-ix.net/historical-traffic-data/ I am fairly sure that if you ask the AMS-IX folks that they can provide possibly even older data sets from before that date. 0.3% today is already IPv6 ;)
Jeroen, Thanks for the thoughtful comment. Its not really a case of 40 Gb/ps, or even 100 Gb/ps. The data I extrapolated out only factored the US East Coast exchange points in the Washington DC area. There are many other large regional peering fabrics across the US in cities such as New York, Miami, Dallas, Chicago, Los Angeles, Silicon Valley, and Seattle. In aggregate, I'm sure the sum total of traffic at these exchanges totals well over 1Tb/sec. I'll take a look at the statistics from AMS-IX and try to extrapolate them out, as you suggest. Looking at their historical data is a good way to validate the model I used all well - at or ahead of Moore's law.
Just a brief thought: 650 Tbps will give 650 million people each a 1 megabit/sec stream, which is enough to watch reasonable video quality, all at the SAME TIME. How much more bandwidth would the world need?