With all of the hype surrounding the migration of video services to IP networks, an issue that often gets overlooked is whether or not there is enough capacity on the Internet backbones to handle the increased load. Not surprisingly, the limits may be closer than what many users expect. In fact, unless significant capacity increases continue to be made, we could run out of bandwidth before the end of the current decade.
THE NUMBERS ARE IN
(click thumbnail)The comScore Video Metrix ranking for March 2007 show Google as the top U.S. streaming video property with 1.2 billion streams initiated.Consider the trends. According to comScore, U.S.-based companies delivered more than 7 billion video streams over the Internet to 126 million unique viewers in March 2007. The Reston, Va.-based Internet marketing research firm said the average viewer consumed 55 streams, or roughly two per day. YouTube alone was responsible for 1.1 billion streams to 55 million unique viewers. This represents a tenfold increase in the past year in the number of video streams delivered per month by YouTube, which was widely reported to routinely deliver 100 million streams per month before being acquired by Google last summer.
A lot of bandwidth is required for making this many video deliveries. Using 3 minutes as an average length of a video, and 4 MB per minute for the low-resolution video (just over 500 kbps), gives an average file size of 12 MB. Delivering 1.1 billion streams would require about 12.5 PB (that’s Petabytes, or millions of gigabytes) of bandwidth for the month, or about 40 Gbps of traffic if the downloads were evenly spread out across the month, which, of course, they are not. Now this seems like a fairly heavy load on the Internet. Indeed, speaking at a cable TV conference in Amsterdam last February, Vincent Dureau, head of TV technology at Google, said, “the Web infrastructure, even Google’s [infrastructure], does not scale. It is not going to offer the quality of service that consumers expect.”
Things could get really ugly if even a small subset of today’s television viewers abandon their current supplier (such as broadcast, CATV or satellite) and try to get their programming over the Internet. If 1 million viewers (which is a pretty small audience by U.S. television broadcast standards) are all watching a modest-quality 1 Mb broadcast (not good enough to be called SD), total bandwidth demand would be 1 Tbps (1 million Mbps). To put this in perspective, the total bandwidth from one major content delivery network is only 1 Tbps, and that has to be shared among all their clients.
CAN MULTICASTING HELP?
Unfortunately no, because the Internet is not multicast enabled. What this means in practical terms is that a separate stream of video data must be created at the source for each viewer, and transported all the way to whatever device is being used to view the stream. Even if two people in the same building are watching the same stream at the same time, a separate stream needs to be delivered from the source to each one. Unfortunately, because of the varying age and state of the equipment that makes up the Internet, it is simply not feasible to enable multicasting, even though that would go a long way toward solving the bandwidth problem. For private networks, where the age and configuration of routing equipment can be controlled, multicasting is a powerful way to distribute real-time video streams to multiple destinations simultaneously.
P2P TO THE RESCUE?
Video also puts an enormous load on the Internet though the phenomenon of file sharing using peer-to-peer networks. These networks operate by first storing content files within user devices (such as home PCs) and then using these devices to serve up content to other users by way of the upstream data connections from these devices back to the Internet. P2P utilities such as BitTorrent, eDonkey, and FastTrack are in widespread use throughout the world. Many of the files exchanged on P2P services are movies that have been ripped from DVDs, and can easily occupy 5 GB or more. You can create a lot of data traffic by moving even a relatively few of these files around.
According to data collected and analyzed by CacheLogic, a content delivery network provider in the United Kingdom, 50 to 65 percent of all downstream traffic (from the Internet to users) is occupied by P2P applications, and 75 to 90 percent of the upstream traffic is driven by P2P. Video makes up a significant portion of this traffic—more than 60 percent according to CacheLogic. A number of new video distribution services that have been announced by major studios are using P2P for file distribution (for example, In2Movies from Time Warner).
P2P solves one Internet video problem while creating another. It gets around the problem of scaling up a central server to deliver files to millions of viewers by distributing that function to every member of the network. At the same time, it creates a problem for ISPs, because it creates a great deal of traffic from subscribers back to the core, which is not how many of the delivery technologies have been designed to work. Because this upstream bandwidth is typically less than the downstream bandwidth, it becomes the limiting factor in a P2P network.
Andrew Parker, chief technical officer of CacheLogic and I spoke about some of the techniques his firm has developed to use P2P to handle these increasing loads.
The secret, he says, is to tap both the power of centralized servers for storing some forms of content and use classic P2P technology to deliver content between users that are in close proximity. The result, he says, is a system that is both robust and flexible enough to handle large amounts of data in a way that is well suited to today’s Internet.
DOES IPTV HAVE THIS PROBLEM?
Not in any practical sense. In this column, we have been talking about transmitting video over the public Internet, and not about purpose-built IPTV networks. The latter networks, particularly those that are constructed and managed by a single service provider from end-to-end are really not at risk, because the amount of bandwidth occupied by each viewer is carefully controlled. In addition, on a private IPTV network, different priorities can be assigned to different types of traffic, and multicasting can be used. The latter is particularly important, because multicasting allows a single IP video source to feed a number of destinations. This results in very efficient bandwidth utilization because copies of each stream are created only where they are needed as they reach branching points.
IS THERE A FUTURE?
Certainly, the loads placed on the Internet by video and other services will continue to increase. But, as in the past, capacity will be added whenever there is an economic justification to do so. Therefore, as it has for most of its history, the capacity of the Internet will continue to exceed the loads placed on it, if only just barely. Innovation will also continue, in the form of better ways to compress video, in the form of ever more powerful routers to switch the data, and in the form of higher capacity optical backbones.
The bottom line is that there is no crisis today, and likely not to be one in the future, unless millions of people suddenly decide to cancel their TV subscriptions and only watch programming delivered over the Internet. Let’s hope that the shift is gradual.
