Distribution & Delivery Global Viewpoint – May 2022

Removing the MCR core SDI/AES router and replacing it with a huge ethernet switch is not transitioning to IP. Our history is one of point-to-point communications where the connectivity is well defined and meets very high tolerances. To a certain extent we’ve been spoiled as we haven’t had to share SDI or AES connectivity with other services. But with IP, that all changes.

Flexible IP and ethernet networks are hugely complex. Not only do we have to think about other service users and how they’re using IP, but we must also think about traffic engineering, rate shaping, and forensic monitoring. Although ST2110 uses UDP fire-and-forget protocols to keep latency low, other services within the infrastructure may be using TCP connections to guarantee delivery, especially for messaging, control, and monitoring.

TCP transfers exist for a length of time determined by the type of data they are carrying. For example, an FTP transfer of a movie could last an hour, but the delivery of a mouse click could last a few hundred milliseconds. And then there are challenges around head-of-line blocking as the long transfers, often called elephant flows, can hold up short bursts, often called mice flows, thus creating latency for the mice flows and affecting the user experience. Unless efficient routing of elephant flows is provided through load balancing, congestion can easily occur leading again to increased latency.

Research has shown that in an internet datacenter, elephant flows make up only 20% of all TCP flows, but are responsible for 80% of all the data in the switch. So, what’s this got to do with broadcasting you may be asking as ST2110 uses UDP and not TCP. Well, we must think about how TCP flows will influence the network. For example, if we’re unaware of the importance of RTT (Round Trip Time) of the network then we could find that IP packets held in buffers due to head-of-line blocking are forcing a TCP timeout causing the IP packets to be resent, thus creating network congestion and further impacting latency.

One solution to this is to keep UDP and TCP messages apart on physically different networks. Yes, that might work, but it would be wholly impractical, especially if your using NDI and ST2110 together. Even so, you still need to consider the dynamic nature of TCP flows and how they are routed, especially if your network relies on ECMP (Equal-Cost Multi-Path) for routing.

Transitioning to IP is not easy, but the outcome often justifies the very steep learning curve. The better news is that a new type of System Integrator is emerging that not only understands how to glue SaaS services together, but also understands the intricacies of how to design, build, and monitor networks for broadcasters. Although we may well need rocket scientists to make the most flexible of networks operate efficiently, there are vendors out there who’s bread and butter is achieving this.