Challenges and solutions for adaptive bitrate (ABR) scaling and high-density transcoding

Under the influence of the COVID-19 pandemic, Internet video traffic is growing at an accelerated rate. Video transcoding workloads in e-sports, telemedicine, e-commerce, entertainment, and distance learning are becoming more common, and computing density is increasing. Content delivery network (CDN) providers that deliver content services are under tremendous pressure to efficiently distribute live video content with minimal capital expenditure (CAPEX), for which they have to re-evaluate their hardware infrastructure and software capabilities.

In short, content delivery networks typically face one or more of the following major challenges:

Cost: Supporting a large number of scaling functions requires high-performance transcoding servers, which leads to higher costs;

Flexibility: Customers’ changing bandwidth needs require flexible configuration and scaling of hardware infrastructure support;

TCO: How to reduce total cost of ownership (TCO) and improve performance of video transcoding workloads within existing dedicated frameworks;

Ease of use: It is hoped that hardware acceleration performance can be achieved by using a general software framework such as FFmpeg without the need for underlying hardware development;

Low latency: Ability to stream live content with the lowest possible latency for applications such as live streaming video games, where a great user experience relies on real-time user engagement and collaboration.

For many years, Xilinx has provided FPGA-based hardware-accelerated video transcoding solutions for content delivery networks. With a deep understanding of the challenges faced by content delivery networks, Xilinx has launched a solution that can meet these challenges well - the Xilinx U30 Software Developer Kit (SDK). This is a complete software stack that simplifies development by provisioning resources and managing capacity for large-scale video streaming infrastructure.

Making ABR more efficient and cost-effective

The Xilinx U30 SDK combined with the Alveo U30 accelerator card enables acceleration of compute-intensive real-time adaptive bitrate (ABR) video transcoding workloads with the highest video channel density and lowest cost.

There are many video transcoding solutions on the market that provide ABR capabilities, but they all suffer from some inefficiencies. These solutions often have one of the following two major flaws, both of which increase costs and cause other problems:

The first type: CPU resources are required for ABR scaling. There are transcoding cards dedicated to encoding and decoding, but they require the CPU on the server to perform ABR scaling, which adds a heavy load to the CPU computing resources. In many cases, the result is that more expensive and more powerful servers are required to support ABR, which uses up capital expenditures for no benefit.

Second: ABR scaling on the accelerator card reduces the channel density that the accelerator card can support.

Some commonly used GPU transcoding cards do not delegate ABR scaling to the CPU, but use additional hardware resources on the card to perform the scaling, which results in fewer channels on a single card. More cards mean more servers, higher costs, and more management required.

The Xilinx solution provides scalar H.264, H.265/HEVC video codec capabilities on a single Xilinx U30 accelerator card, helping free up CPU resources and improve performance while also saving 80% in cost, 90% in power, and 75% in space.

Make transcoding easier to develop and deploy

One of the main features of the U30 SDK is to speed up development and support the use of the FFmpeg industry standard to fully utilize the hardware acceleration capabilities of the high-channel density video transcoding solution based on the Xilinx Alveo U30. In addition, if you are working under a proprietary framework, you can easily integrate with the proprietary framework using the C-based API provided in the U30 SDK to leverage the performance of the U30 solution.

The U30 SDK is designed to grow with video services, providing resource allocation and capacity management for large-scale video streaming infrastructure. The Xilinx Resource Manager (XRM) provided with the U30 SDK can manage and allocate all hardware acceleration functions, supporting multiple video processing jobs to run on multiple Alveo U30 accelerator cards, achieving seamless workload scaling.

The Xilinx U30 SDK GitHub page provides extensive documentation, video transcoding examples, and video quality assessment tools to quickly stream video applications.

Direct transcoding and FTRT transcoding

Content delivery networks and video service providers often use codecs such as H.264 and HEVC to compress streaming video content for distribution to consumers. Providers need to convert HEVC encoded streams to H.264/AVC video encoding formats to fully utilize HEVC's excellent bit rate savings, or re-compress content libraries from one encoder format to another to reduce storage requirements. Xilinx U30 SDK provides direct real-time transcoding capabilities that can seamlessly convert from one format to another.

In addition, there are some use cases that require faster-than-real-time video transcoding. For example, car-buying websites often want to get videos back to customers as quickly as possible; security services want to get video footage back to customers as quickly as possible. The U30 SDK supports features that exceed real-time video transcoding speeds, allowing content delivery networks to transcode 60 minutes of 1080p 60fps high-quality video in 20 minutes on a single U30 accelerator.

Highest density, lowest total cost of ownership

As an adaptable accelerator card designed for high performance and efficiency, Xilinx Alveo accelerates dynamic workloads in local data centers or in the cloud.

The Alveo U30 accelerator card is based on the Xilinx Zynq® UltraScale+™ MPSoC, a power-optimized, fully programmable system-on-chip (SoC) that integrates a video codec and graphics engine for ultra-high-definition video. The U30 card supports both H.264 and HEVC (H.265) codec formats, and can support up to 16 1080p 30 frames per card.

With a high-channel density real-time video transcoding solution based on the Alveo U30 accelerator card, Xilinx provides optimal compression efficiency for content delivery networks, with a certain number of channels and lower cost per stream and power consumption than software encoding and GPU encoding solutions.