The public cloud represents the ultimate destiny for broadcasting, the end of a journey from dedicated infrastructures optimized for live video through more traditional data centers to end up with virtualized networks allowing use of distributed commodity hardware that can be turned on and off almost like a tap.
The challenge is to do this without throwing the baby out with the bathwater. In this case the baby embraces QoS, SLAs and all those things associated with absolute audio-video quality for all services including live, while the bathwater are all those attendant costs and inflexibilities associated with traditional broadcast systems. Put another way the aim is to achieve all the benefits of public cloud, such as fast provisioning, resource elasticity, reduced hardware dependency and support for all forms of delivery including OTT over unmanaged networks, from a common head end. In the process this should convert rather unpredictable Capex into sustainable Opex, while ensuring QoS for all services including crucially live OTT, which is a fast emerging requirement for many broadcasters, operators and content owners.
The major broadcast system vendors are all coming along with new platforms, or more likely enhancements to existing ones, to enable virtualized public cloud, with Thomson Video Networks the latest to set out its stall. It has just launched a virtualized edition of its ViBE VS7000 multi-screen and HEVC encoder, which it says is specifically engineered for cloud-based video processing of content delivered to OTT and multiscreen platforms in hybrid deployments alongside traditional broadcast services. It would also work for OTT-only deployments, but might be overkill for smaller operators.
Thomson emphasizes correctly that such a cloud platform must support not just file and live transcoding for all leading codecs including H.264 or HEVC up to UHD resolution, but also all the associated workflow requirements, including audio and video processing functions such as branding, resizing, packaging, scrambling, multiplexing, loudness control, ad insertion and splicing. All these functions are in fact already available with the existing VS7000, but have now been extended to virtualized environments.
Thomson can be congratulated for at least being very clear what it means by a virtualized environment and over the progressive steps broadcasters have to take to get there. The firm started off by observing that many broadcasters are renewing their architecture to reduce costs and prepare for full multiscreen hybrid services, but face a number of options, whose relative merits depend on factors such as geography, scale and service profile. Options include retention of a dedicated optimized architecture, migrating to a hybrid one by introducing some level of private virtualization, or going towards the whole hog by starting to move some workflows to a public cloud solution offering greater flexibility and scale economies. The decision has to be made against various unchanging requirements, such as video quality, maximum audience reach, reliability of service delivery, ability to conduct on-air maintenance, and ease of operation. In some cases, perhaps for a broadcaster continuing to transmit solely via DTT while relying on third parties for other forms of distribution, there may be no hurry to move away from a dedicated infrastructure. In other cases, perhaps for a broadcaster with plenty of existing IT capacity serving a variety of services, an in-house private cloud would improve operational efficiency without risk of relying on external providers.
But for a growing number of cases the public cloud will at least be part of the solution, perhaps in tandem with in-house capacity in some hybrid arrangement. Again as Thomson observed, a public cloud and possibly even a private one can also deliver optimum efficiency, cost reduction and elasticity, if it is fully virtualized. This is where the definitions come in useful, for there is confusion here with software video platforms, which are really just an interim step towards virtualization.
Software transcoding packages achieve independence from the underlying hardware, but not from its OS. They enable expensive dedicated hardware to be replaced with a generic commercially available systems, but still tethered to hardware running a particular OS, which might mean servers running a version of Windows or Linux. Software solutions insulate the transcoding from hardware specifics and made it easier to modify and evolve services, which is increasingly important now that broadcasting and video distribution have become faster moving. The advent of OTT and multiscreen has only added to the software impetus by increasing pace of innovation, introducing multiple target device formats and screen sizes, while adding adaptive bitrate (ABR) encoding profiles to improve video quality over unmanaged networks.
These considerations drove the development of Thomson’s MediaFlex OS multimedia framework, which initially was intended largely for internal use to enable faster shoehorning of the VS7000 software to different target OSs. But then Thomson realized that without that much modification, MediaFlex OS was just the answer to full virtualization, insulating encoding and video processing from the OS as well as the hardware, by functioning as a single higher level OS above all the underlying OSs tethered to underlying hardware.
The basic idea is over 40 years old now, having been pioneered by IBM with its VM (Virtual Machine) architecture launched in 1972. This was a time when IBM dominated the world of business computing and almost uniquely had a requirement for an operating system that would tie its own machines together and enable applications to run across all of them. This comprised a control program called the hypervisor running on the physical hardware, creating the virtual machine environment on top. The hypervisor then took over the traditional OS tasks, such as input/output, resource-sharing and management of the now enlarged virtual storage arsenal. This also meant all users could be presented with a unique virtual machine dedicated to their own needs, with their own address space and virtual devices, capable of running any application supported by at least one of the constituent underlying operating systems.
This is really what Thomson has now enabled for ViBE, but with the difference that the virtual hypervisor incorporates features needed for modern day video distribution, without which it could not deliver adequate performance or flexibility. There are a number of obvious requirements that the virtual OS must fulfil, one being support for live ingest. The point here is that virtualized infrastructures rely on intermediate software with IP as the sole communications protocol for exchanging video. This means ingesting uncompressed legacy HD-SDI content to a virtualized infrastructure requires the signal to be compressed to a level low enough to save on bandwidth, but high enough to preserve sufficient video quality for further processing in the cloud. Here Thomson, like others, has developed a mezzanine encoder capable of converting multiple HD-SDI signals to IP while protecting the signal against errors that inevitably occur at the IP network level.
The more interesting aspect of cloud virtualization concerns minimizing the loss of hardware optimization. Thomson is good enough to admit that there will always inevitably be a trade-off, for even software encoding can be optimized significantly by linking intimately with the underlying OS, so that the whole package can indirectly exploit features of the hardware. Virtualized implementations in principle enable the service to run on any server, or at least a large subset of server brands and models that merely must meet quite basic hardware requirements. This removes OS and therefore hardware dependency, with the benefit of enabling transcoding deployment on an existing server infrastructure very quickly, but at the expense of CPU optimization. The upshot is that for a given level of performance more hardware has to be provisioned, which offsets the savings through economies of scale. This is why a public cloud solution will not be right for every single situation, because it will increase costs for operators with highly predictable workloads that experience relatively blunt peaks and troughs.
But there will be a diminishing number of cases where a public cloud will not save money and improve competitiveness through faster provisioning, at least as part of the overall infrastructure. So as Thomson points out, the service will come at a cost, but it is likely to be the best solution in many cases when transcoding workload varies over time for such services as file transcoding or temporary live services, or even when in-house platform ownership is not financially viable. In this case, a pay-per-usage or pay-as-you-grow model may be more feasible. As Thomson also pointed out, transcoding a VoD catalog to a new format is a typical use case that is perfectly suited to public cloud services. Here there is no real-time requirement because hundreds or thousands of Virtual Machines (VM) can be provisioned in parallel to process the full catalog. If some jobs fail they can simply be restarted at no cost or inconvenience to the VoD service provider.
There is one health warning worth making, which is not automatically to trust promises made by public cloud providers, like “unlimited capacity and elasticity”. There is no such thing and it is worth checking exactly how much spare capacity a given public infrastructure has at peak times and what back up provisioning if any that provider has. A public cloud is not by itself a universal panacea for all the challenges of multiscreen broadcasting.