Your browser is not supported. Please update it.

23 October 2014

EBU completes work on broadcast SLAs for Telco IP networks

Broadcasters and content distributors now have tools to specify SLAs (Service Level Agreements) when dealing with Telcos or CDN providers for media transport, following the completion of a project by the EBU (European Broadcasting Union).

This has become an increasingly urgent requirement as more video transport is moved to IP networks, with more stringent requirements for QoS than other forms of data, such as email, which do not require real time delivery. Video content is highly susceptible to absolute delay and jitter or delay variation, which needs to be accounted for in SLAs. The business terms are also different, given that some content such as live premium sports requires higher QoS than say on demand archived content.

As a result the typical measure of availability for a Telco network specified in a number of 9s is not appropriate. A level of 99.999% equates to five minutes downtime a year, but for a broadcaster the impact would depend on which five minutes. By this same token the traditional Telco business model where an SLA is framed adds a partial insurance policy with a rebate scaling in proportion to downtime is not right either for a broadcaster, given that the impact on reputation of a major outage may be beyond immediate financial compensation.

As a result the EBU’s project has come up with a mechanism for relating media transport requirements with measures of Telco network QoS and availability. The model only applies to Telco-owned IP networks and not the public Internet, although the latter can sometimes be addressed by combining SLAs through one or more CDNs coupled with an ISP network for the final delivery. This is why we are seeing major OTT providers such as Netflix building out their own CDNs and connecting them directly to ISPs so that they can guarantee end to end SLAs, effectively bypassing the Internet.

The EBU project began by examining the two fundamental SLA aspects of performance and availability, splitting these into sub-categories. There are two fundamental elements of performance, capacity, typically measured as a bit rate for data networks, and latency or delay. For bit rate the SLA will depend on which stage in the lifecycle the video is at, so that an uncompressed SDI signal for example requires 270 Mbps, but this has to be translated correctly into a specification for the target IP service, taking account of the real throughout achieved. The point here is that an IP service may specify an interface bit rate of say 100 Mbps but only actually achieve 80 Mbps over the link after various overheads have taken their toll. Therefore the video SLA has to incorporate some headroom to account for that.

Much the same applies to latency, given that variable delays are incurred as a result of the jittery nature of bursty packet networks. This requires a de-jitter buffer at the end of the link to smooth out the flow of packets by pushing them out at regular intervals. Again requirements vary, so that a good two way video interview would require end to-end latency of no more than 50ms, while a one way satellite broadcast could tolerate up to 1s.

On the availability front the EBU recognized three mechanisms that need to be incorporated in SLAs operating at different levels of the communications hierarchy. Firstly at the signal transmission level there is Forward Error Correction (FEC), used to protect against short-term transient disturbance to the signal, by using algorithms that add inbuilt redundancy by adding a few bits of error-correction data. This data is entered by the encoder at the start of the link and then used by the receiver to verify the integrity of the signal and in the event of errors, reconstruct the signal if possible. From the SLA perspective it adds both to latency and bit rate consumed.

Secondly there is retransmission at the IP packet level to protect against short-term disturbances in packet flow. This includes mechanisms in TCP at the transport level enabling the receiver to request retransmission of missing packets. The point here for the SLA is that an end to end system designed to carry live broadcast signals needs to incorporate a ‘de-jitter buffer’ again to ensure a smooth signal flow downstream. Also this is too slow for interactive services because any packets needing to be retransmitted would arrive too late.

Finally at the link level there is path protection switching, including various techniques for rerouting or taking duplicate paths. For example there is packet merging, operating in the IP layer to merge two IP streams together to overcome upstream network disturbances. This can be used as an alternative to FEC, trading off bit rate against latency. It doubles the network bandwidth required, but eliminates the need for an FEC buffer, reducing the end-to-end latency.

There are a number of other factors that have had to be taken into account, such as the difference between fiber and copper. Fiber is essentially immune from transient physical problems and is either available or not, with the latter usually being caused by the cable being severed. Copper performance on the other hand can be affected by electromagnetic interference. This factor may also need to be taken into account when framing SLAs.