Sunday, November 16, 2008

Telco IPTV Network Architecture

Both Alcatel-Lucent and Cisco have introduced edge routers that facilitate Telco IPTV networks by supporting fast channel change, packet retransmission, and ad insertion using a large flash memory in the router. Both companies are talking about having enough memory in these routers to support video on demand caching. That is, the most popular video on demand assets would be stored on the routers which would eliminate the network traffic required to provide them from a centralized server.

I think that these developments are a real step forward for IPTV networks. I think that their ability to support fast channel change and packet retransmission are fundamental improvements. The ad insertion and video on demand capabilities will provide significant improvements, up to a point.

The fundamental issue will be the amount of storage provided by the router. In a personalized ad insertion system, this will work fine as long as the number of personalized ads does not become too large. The issue with video on demand will be the hit rate, that is the number of video on demand requests that are served by the content on the router vs. the centralized router.

My own thinking on this issue has evolved. I did a report for MRG in the scaling of IPTV networks three years ago that identified that the amount of traffic required to support high levels of on demand traffic will be a major issue. At that time, I felt that the caching approach that Alcatel-Lucent and Cisco are talking about would be the answer.

I looked at this issue again last year in the report Networking Strategies for TelcoTV Services. With further analysis, I realized that the traffic to fill the distributed video on demand caches would be enough by itself to swam metro networks. My conclusion was that the IPTV traffic had to be removed from the IP/Ethernet networks and move down to the optical level.

My recent report Telecom 2020: Transformation Strategies looked at the issue again and took into account the consolidation of fiber access systems into large centers that would serve very large numbers of subscribers. This architecture would mean that there would be enough video on demand traffic to one of these major fiber access centers to fill a 10 Gbps or even a 100 Gbps optical link; therefor, there is no benefit for bringing this traffic up to the packet layer. It will be much more economical just to pass directly to the fiber access center over a direct optical link.

There are too philosophical points that drive my thinking. The first is that as IPTV moves from multicast to on demand and unicast, the TV traffic will dominate the network. It just makes sense to optimize the network for the dominant traffic type.

The second point is that it is very difficult to understand how a packet network can provide an acceptable SLA or QoS when 99 percent of the traffic on the network is deterministic TV traffic. On the other hand, it is easy to see how this will work over direct optical links.

There is a real question of timing here. For many carriers, the cached approach introduced by Alcatel-Lucent and Cisco will be a big help over the next three to five years. However, I do think that these carriers will be thinking seriously about a direct optical approach after that.

No comments: