IPTV primer: an overview of the fusion of TV and the Internet

By Iljitsch van Beijnum
Ars Technica

March 10, 2011

http://arstechnica.com/telecom/news/2011/03/iptv-tv-and-the-internet-a-match-made-in-heaven.ars


Profound as it may be, the Internet revolution still pales in comparison 
to that earlier revolution that first brought screens in millions of 
homes: the TV revolution. Americans still spend more of their non-sleep, 
non-work time on watching TV than on any other activity. And now the 
immovable object (the couch potato) and the irresistible force (the 
business-model destroying Internet) are colliding.

For decades, the limitations of technology only allowed viewers to watch 
TV programs as they were broadcast. Although limiting, this way of 
watching TV has the benefit of simplicity: the viewer only has to turn 
on the set and select a channel. They then get to see what was deemed 
broadcast-worthy at that particular time. This is the exact opposite of 
the Web, where users type a search query or click a link and get their 
content whenever they want. Unsurprisingly, TV over the Internet, a 
combination that adds Web-like instant gratification to the TV 
experience, has seen an enormous growth in popularity since broadband 
became fast enough to deliver decent quality video. So is the Internet 
going to wreck TV, or is TV going to wreck the Internet? Arguments can 
certainly be made either way.

The process of distributing TV over a data network such as the Internet, 
a process often called IPTV, is a little more complex than just sending 
files back and forth. Unless, that is, a TV broadcast is recorded and 
turned into a file. The latter, file-based model is one that Apple has 
embraced with its iTunes Store, where shows are simply downloaded like 
any other file. This has the advantage that shows can be watched later, 
even when there is no longer a network connection available, but the 
download model doesn't exactly lend itself to live broadcasts—or instant 
gratification, for that matter.


Streaming

Most of the new IPTV services, like Netflix and Hulu, and all types of 
live broadcasts use a streaming model. Here, the program is set out in 
real time. The computer—or, usually by way of a set-top-box, the 
TV—decodes the incoming stream of audio and video and then displays it 
pretty much immediately. This has the advantage that the video starts 
within seconds. However, it also means that the network must be fast 
enough to carry the audio/video at the bitrate that it was encoded with. 
The bitrate can vary a lot depending on the type of program—talking 
heads compress a lot better than car crashes—but for standard definition 
(SD) video, think two megabits per second (Mbps).

To get a sense just how significant this 2Mbps number is, it's worth 
placing it in the context of the history of the Internet, as it has 
moved from transmitting text to images to audio and video. A page of 
text that takes a minute to read is a few kilobytes in size. Images are 
tens to a few hundred kilobytes. High quality audio starts at about 128 
kilobits per second (kbps), or about a megabyte per minute. SD TV can be 
shoehorned in some two megabits per second (Mbps), or about 15 megabytes 
per minute. HDTV starts around 5Mbps, 40 megabytes per minute. So 
someone watching HDTV over the Internet uses about the same bandwidth as 
half a million early-1990s text-only Web surfers. Even today, watching 
video uses at least ten times as much bandwidth as non-video use of the 
network.

In addition to raw capacity, streaming video also places other demands 
on the network. Most applications communicate through TCP, a layer in 
the network stack that takes care of retransmitting lost data and 
delivering data to the receiving application in the right order. This is 
despite the fact that the IP packets that do TCP's bidding may arrive 
out of order. And when the network gets congested, TCP's congestion 
control algorithms slow down the transmission rate at the sender, so the 
network remains usable.

However, for real-time audio and video, TCP isn't such a good match. If 
a fraction of a second of audio or part of a video frame gets lost, it's 
much better to just skip over the lost data and continue with what 
follows, rather than wait for a retransmission to arrive. So streaming 
audio and video tended to run on top of UDP rather than TCP. UDP is the 
thinnest possible layer on top of IP and doesn't care about lost packets 
and such. But UDP also means that TCP's congestion control is out the 
door, so a video stream may continue at full speed even though the 
network is overloaded and many packets—also from other users—get lost. 
However, more advanced streaming solutions are able to switch to lower 
quality video when network conditions worsen. And Apple has developed a 
way to stream video using standard HTTP on top of TCP, by splitting the 
stream into small files that are downloaded individually. Should a file 
fail to download because of network problems, it can be skipped, 
continuing playback with the next file.


Where are the servers? Follow the money

Like any Internet application, streaming of TV content can happen from 
across town or across the world. However, as the number of users 
increases, the costs of sending such large amounts of data over large 
distances become significant. For this reason, content delivery networks 
(CDNs), of which Akamai is probably the most well-known, try to place 
servers as close to the end-users as possible, either close to important 
interconnect locations where lots of Internet traffic comes together, or 
actually inside the networks of large ISPs.

Interestingly, it appears that CDNs are actually paying large ISPs for 
this privilege. This makes the IPTV business a lot like the cable TV 
business. On the Internet, the assumption is that both ends (the 
consumer and the provider of over-the-Internet services) pay their own 
ISPs for the traffic costs, and the ISPs just transport the bits and 
aren't involved otherwise. In the cable TV world, this is very 
different. An ISP provides access to the entire Internet; a cable TV 
provider doesn't provide access to all possible TV channels. Often, the 
cable companies pay for access to content.

A recent dispute between Level3 and Comcast can be interpreted as 
evidence of a power struggle between the CDNs and the ISPs in the IPTV 
arena.


Walled gardens

For services like Netflix or Hulu, where everyone is watching their own 
movie or their own show, streaming makes a lot of sense. Not so much 
with live broadcasts.

So far, we've only been looking at IPTV over the public Internet. 
However, many ISPs around the world already provide cable-like service 
on top of ADSL or Fiber-To-The-Home (FTTH). With such complete 
solutions, the ISPs can control the whole service, from streaming 
servers to the set-top box that decodes the IPTV data and delivers it to 
a TV. This "walled garden" type of IPTV typically provides a better and 
more TV-like experience—changing channels is faster, image quality is 
better, and the service is more reliable.

Such an IPTV + Internet access service is a lot like what cable networks 
provide, but there is a crucial difference: with cable, the bandwidth of 
the analog cable signal is split into channels, which can be used for 
analog or digital TV broadcasts or for data. TV and data don't get in 
each other's way. With IPTV on the other hand, TV and Internet data are 
communication vessels: what is used by one is unavailable to the other. 
And to ensure a good experience, IPTV packets are given higher priority 
than other packets. When bandwidth is plentiful, this isn't an issue, 
but when a network fills up to the point that Internet packets regularly 
have to take a backseat to IPTV packets, this could easily become a 
network neutrality headache.


Multicast to the rescue

Speaking of networks that fill up: for services like Netflix or Hulu, 
where everyone is watching their own movie or their own show, streaming 
makes a lot of sense. Not so much with live broadcasts. If 30 million 
people were to tune into Dancing with the Stars using streaming, that 
means 30 million copies of each IPTV packet must flow down the tubes. 
That's not very efficient, especially given that routers and switches 
have the capability to take one packet and deliver a copy to anyone 
who's interested. This ability to make multiple copies of a packet is 
called multicast, and it occupies territory between broadcasts, which go 
to everyone, and regular communications (called unicast), which go to 
only one recipient. Multicast packets are addressed to a special group 
address. Only systems listening for the right group address get a copy 
of the packet.

Multicast is already used in some private IPTV networks, but it has 
never gained traction on the public Internet. Partially, this is a 
chicken/egg situation, where there is no demand because there is no 
supply and vice versa. But multicast is also hard to make work as the 
network gets larger and the number of multicast groups increases. 
However, multicast is very well suited to broadcast type network 
infrastructures, such as cable networks and satellite transmission. 
Launching multiple satellites that just send thousands of copies of the 
same packets to thousands of individual users would be a waste of 
perfectly good rockets.


Peer-to-peer and downloading

Converging to a single IP network that can carry the Web, other data 
services, telephony, and TV seems like a no-brainer.

Multicast works well for a relatively limited number of streams that are 
each watched by a reasonably sized group of people—but having very many 
multicast groups takes up too much memory in routers and switches. For 
less popular content, there's another delivery method that requires no 
or few streaming servers: peer-to-peer streaming. This was the 
technology used by the Joost service in 2007 and 2008. With peer-to-peer 
streaming, all the systems interested in a given stream get blocks of 
audio/video data from upstream peers, and then send those on to 
downstream peers. This approach has two downsides: the bandwidth of the 
stream has to be limited to fit within the upload capacity of most 
peers, and changing channels is a very slow process because a whole new 
set of peers must be contacted.

For less time-critical content, downloading can work very well. 
Especially in a form like podcasts, where an RSS feed allows a computer 
to download new episodes of shows without user intervention. It's 
possible to imagine a system where regular network TV shows are made 
available for download one or two days before they air—but in encrypted 
form. Then, "airing" the show would just entail distributing the 
decryption keys to viewers. This could leverage unused network capacity 
at night. Downloads might also happen using IP packets with a lower 
priority, so they don't get in the way of interactive network use.


IP addresses and home networks

A possible issue with IPTV could be the extra IP addresses required. 
There are basically two approaches to handling this issue: the one where 
the user is in full control, and the one where an IPTV service provider 
(usually the ISP) has some control. In the former case, streaming and 
downloading happens through the user's home network and no extra 
addresses are required. However, wireless home networks may not be able 
to provide bandwidth with enough consistency to make streaming work 
well, so pulling Ethernet cabling may be required.

When the IPTV provider provides a set-top box, it's often necessary to 
address packets toward that set-top box, so the box must be addressable 
in some way. This can eat up a lot of addresses, which is a problem in 
these IPv4-starved times. For really large ISPs, the private address 
ranges in IPv4 may not even be sufficient to provide a unique address to 
every customer. Issues in this area are why Comcast has been working on 
adopting IPv6 in the non-public part of its network for many years. When 
an IPTV provider provides a home gateway, this gateway is often 
outfitted with special quality-of-service mechanisms that make 
(wireless) streaming work better than run-of-the-mill home gateways that 
treat all packets the same.


Predicting the future

Converging to a single IP network that can carry the Web, other data 
services, telephony, and TV seems like a no-brainer. The phone companies 
have been working on this for years because that will allow them to buy 
cheap off-the-shelf routers and switches, rather than the specialty 
equipment they use now. So it seems highly likely that in the future, 
we'll be watching our TV shows over the Internet—or at least over an IP 
network of some sort. The extra bandwidth required is going to be 
significant, but so far, the Internet has been able to meet all 
challenges thrown at it in this area. Looking at the technologies, it 
would make sense to combine nightly pushed downloads for popular 
non-live content, multicast for popular live content, and regular 
streaming or peer-to-peer streaming for back catalog shows and obscure 
live content.

However, the channel flipping model of TV consumption has proven to be 
quite popular over the past half century, and many consumers may want to 
stick with it—for at least part of their TV viewing time. If nothing 
else, this provides an easy way to discover new shows. The networks are 
also unlikely to move away from this model voluntarily, because there is 
no way they'll be able to sell 16 minutes of commercials per hour using 
most of the other delivery methods. However, we may see some 
innovations. For instance, if you stumble upon a show in progress, 
wouldn't it be nice to be able to go back to the beginning? In the end, 
TV isn't going anywhere, and neither is the Internet, so they'll have to 
find a way to live together.

_______________________________________________
Medianews mailing list
[email protected]
http://lists.etskywarn.net/mailman/listinfo/medianews

Reply via email to