January 18th, 2006

A look at IPTV


Triple Play is a wordbuzz in the industry for the last few years, most notable by equipment manufacturers who is trying to justify why telcos and ISPs should upgrade their infrastructure to handle the demand of video or IPTV. Numerous analysis reports from the likes of IDC also predicted a huge growth in demand; a simple Google search yields numbers like 36M subscribers in Europe by 2009, market boom from 262M in 2005 to 2.9B 2009, etc, thus fueling the interests in IPTV. Therefore, it is not surprising that IPTV has been the focus of CommunicAsia 2005.

However, the reality is not as rosy, as expected in any emerging technology. While there are some notable success in IPTV in various countries, the numbers generally hovers around the 200k to 500k subscribers. Countries which is deploying IPTV finds that it is taking off rather slowly, for a variety of reasons, like expensive CPEs (compared to cable) or monopoly of contents by incumbents (an interesting story for another day).

The main reason however is that many operators are dabbling with IPTV both with anticipations and with reservations. They never really gone all the way to promote the service. One of the reasons is the technical requirements to support IPTV. Assuming just 100k subscribers, to deliver HDTV quality video stream of 4mbps, we are looking at a total throughput of 400Gbps. At a primetime utilization of say just 50% thats still 200Gbps. While SCSI is much faster than SATA (320mbps vs 150mbps), SATA is a lot cheaper. This means we need 1400 SATA harddisk to pump 200Gbps. In practice, you need 50-100% more disks for redundancy, so lets lets round up to 2,000 SATA harddisk. If we pack 16 harddisks in a 2U rackspace, we need at least 5 racks (6-7 if we leave space for cooling) just for the harddisks alone. We will also need at least 20 (40 to be safe) of the media servers assuming we are using 10Gb Ethernet and each media server is capable of supporting 10,000 concurrent streams and thats another 5-10 racks. Most deployments also includes caching/replications of contents beside DSLAM for better performance and thus more disks, more spaces. Operation of such large scale storage is a huge challenge (harddisk failures are common) and normally not a core expertise of a telco.

Next, the infrastructure needed to deliver 200Gbps of traffic. The limitation here is not on the fibers but rather the streams are distributed across your network. Also as both operator and IPTV service provider, there is an expectation of QoS from your subscribers. This usually imply a major overhaul of your backbone infrastructure, a very significant investment. For example, British Telecom is spending £10B over the next five years for their BT 21CN project.

Thus if you are in a senior management of an operator, it does not take a lot to ask “Okay, so we going to do all these for 100k subscribers who will pay as additional US$5/mth”.US$6M/year isnt a lot of money to telcos. Oh, did I mention that the requirements are not scalable? If your subscribers increases to 200k, you need twice of what I describe above.

Does this means IPTV will not take off? No, I think it will. But IPTV couldn’t be the justification for the telcos to upgrade their infrastructure. IPTV will take off only after the infrastructure is ready and seeking for an application to fill the pipe. This is apparent in the places where IPTV do take off: Italy, Hong Kong, Japan, where Fiber to the Home, is readily available at consumer price.

Incidently, all the above calculation makes me even more excited about P2P IPTV. P2P IPTV could be trace back to an academic paper called Coolstreaming (see also Wikipedia) published about a year ago (the authors of the paper is rumored to be funded by Softbank after the publication). The release of their python code no doubt spurred the the creation of PPLive, Sopcast and Cybersky.

The most interesting conclusion of the paper is that P2P IPTV is extremely scalable. A (rough) calculation: Supposing you setup a P2P IPTV channel on your laptop and you only have enough CPU and bandwidth to support 4 clients, the 4 clients could then go on to serve 16 clients, and so on. By the 8th level, you would have served 262k viewers, anywhere on th Internet all from a single laptop!

Unfortunately, despite the scalablity, the technology is unfriendly to operators (they become just a pipe provider). It is difficult to see an telco deploying as it lacks the ability to control subscriptions and also quality which are important requirements for telcos. The business model for P2P IPTV is also yet to be seen.

Nevertheless, P2P IPTV has all characteristic of a disruptive technology which is what makes it even exciting. The fact that anyone on the Internet with a laptop could become a “TV” broadcaster is yet another story for another day.

ps: I wrote this at 4am in the morning, woke up by what I think a Gout attack T_T

Update: I think I screwed up the math quite badly here :( Never try to write this at 4am) 150MBps = 1.2Gbps so the calculation is off quite by a bit. Nevertheless, it is also important to note that that the internal disc drive speed is merely 72MBps or 576Mbps. So we looking at 300-500 harddisks instead of 2000. The racks requirements is down to 1-2 then.

Comments are closed.