For a long time, professional-grade, real-time media transport, processing and monitoring functionality was provided by dedicated hardware. If additional functionality was required in the network, a “box” would be purchased that specifically performed that task.
This dedicated hardware approach made technical sense because of the amount of processing required, as well as the need for low latency, high reliability and minimum power consumption. However, this was inflexible and not always cost effective.
The trend for high-tech equipment in many industries has been a progressive move from dedicated hardware to specialized platforms running software, and then onto COTS (Commercial Off-The-Shelf) IT hardware, such as servers based on X86 CPU architecture, running software applications. For real-time transport, processing and monitoring of media signals, the concept of software-defined platforms is now becoming established. The question is though: how suitable is COTS for live broadcast production?
Intuitively, a COTS approach makes sense, as IT equipment is ubiquitous, open and proven in many environments. Despite some initial hesitancy in the industry, we are beginning to see media processing products appearing that are based on COTS.
OVERCOMING THE CHALLENGES
Currently, COTS products for live media transport, processing and monitoring still only provide limited functionality compared to leading software defined platforms. Before it is possible for COTS hardware to be used for all aspects of live production there are several technical challenges that the industry must firstly work to overcome.
But what exactly are these challenges and what could overcoming them mean for the role of COTS hardware in live production in the future?
At this stage, the main obstacle to using COTS hardware is its real-time processing performance. While it can handle applications—such as audio processing—that involve flows under 10Mbs reasonably well, it can be less efficient with higher rate flows, such as those required for real-time video encoding.
However, developments are taking place that will help running video processing applications on COTS platforms. Firstly, we are beginning to see newer generations of standards, such as JPEG XS encoding, that have been designed from the ground-up for software processing. Secondly, and more fundamentally, chip vendors, such as Intel and Xilinx, are launching or already offer new “generic” FPGA acceleration boards.
The NICs (Network Interface Cards) used for stream acquisition can also be a processing bottleneck. To get the performance needed to handle very high volumes of data in real-time, it helps if the NICs can communicate directly with the FPGA, rather than via the CPU and memory. Vendors such as Mellanox have developed NICs designed to optimize the throughput to the processing functions.
It is worth noting though that a contributing factor to some of the perceived inferior performance on COTS is poor programming; for example, it is too easy to add buffering to solve problems rather than address them in a way that keeps latency down. Lean and efficient code is key to powerful and scalable media functions implementation.
Traditionally, professional media transport has been linear in nature, meaning that the output of equipment utilizes a constant, defined bit rate. In IP terms, this means that packets are transmitted at a constant and steady pace.
COTS equipment is inherently nonlinear, which is a challenge to broadcast orthodoxy. Typically, the software and hardware—due to the resource scheduler—will natively try and push out as much data onto the network as quickly as possible, without any consideration for pacing.
The problem with nonlinear transmission is that bandwidth usage becomes very unpredictable and timing consistency cannot be maintained. And in real-time applications, where packet delays are unacceptable, there should always be enough bandwidth to accommodate all possible flows and keep an accurate packet pace.
The nonlinearity of COTS devices creates the need for substantial additional bandwidth, which is largely unused most of the time and is not desirable economically.
Consequently, the prevailing view in the industry is that COTS-based media applications should be developed in a way that paces packets more carefully.
CPU-based COTS devices have noticeably higher power consumption than bespoke hardware or software-defined platforms based on FPGAs. In some production environments, such as OB vans or remote sites, this can prove to be a challenge.
Over time, COTS power consumption is likely to be reduced, as generic systems introduce FPGA acceleration.
It is also likely (though not necessarily desirable) that broadcast environments will become more accommodating of the extra power needed for COTS.
DOES THE FUTURE LIE IN COTS?
There is no doubt that, for most real-time broadcast media transport, processing and monitoring, the best approach currently is to use software-defined platforms, built on hardware optimized for performance.
While truly generic COTS IT hardware running software is not yet a viable option for many applications—especially video processing that requires very high bandwidth—it is already being used successfully for some applications. Therefore, as technology continues to evolve and is capable of overcoming the challenges and limitations outlined above, it is likely that COTS will be suitable for use in all aspects of real-time broadcast production.
Olivier Suard is vice president of marketing for Nevion.