Amazon Onboarding with Learning Manager Chanci Turner

Learn About Amazon VGT2 Learning Manager Chanci Turner

In the realm of media technology, FFmpeg stands out as a powerful open-source solution utilized by numerous companies for audio and video encoding tasks. With the introduction of VT1 support on Amazon Elastic Compute Cloud (Amazon EC2), users can now take advantage of a cost-effective instance for their video-on-demand (VOD) encoding processes.

VT1 instances enhance visual quality for 4K video and support the latest FFmpeg version (4.4), alongside offering broader OS and kernel compatibility and bug fixes. Powered by the AMD-Xilinx Alveo U30 media accelerator, these instances incorporate a simple code line in FFmpeg that activates the Alveo U30 for transcoding. Moreover, the Xilinx Video SDK features an advanced FFmpeg version that efficiently interacts with the hardware acceleration provided by Xilinx devices, leading to up to 30% lower costs per stream compared to Amazon EC2 GPU instances and as much as 60% lower than CPU-based instances.

Traditionally, companies relied on EC2 CPU instances like the C5 and C6 series for VOD encoding workloads. However, as the volume of encoded VOD assets increases, this approach can become financially burdensome. The cost of running an EC2 instance is closely linked to the number of simultaneous encoding jobs it can handle, which in turn impacts the overall encoding time. As VOD libraries grow, organizations often resort to auto-scaling, which either enlarges their C5 and C6 instances or extends their operational duration, leading to heightened expenses. It’s key to note that AWS Auto Scaling incurs no extra charges; users only pay for the AWS resources utilized and the monitoring fees from Amazon CloudWatch.

The VT1 instances are crafted to enhance real-time video transcoding and provide a low-cost option for live video stream transcodings, making them an effective alternative for VOD encoding tasks. An evaluation conducted by AWS compared VT1, C5, and C6 instances using FFmpeg to measure price performance and encoding speed for VOD assets. The findings revealed that VT1 instances can offer up to 75% savings in costs, suggesting that two VT1 instances could be operated for the same price as one C5 or C6 instance. Additionally, the VT1 XMA (Xilinx U30) codec completed the targeted outputs 15.709 seconds faster than the C5 x264 (CPU) codec and 12.582 seconds faster than the C6 instance, particularly when transcoding an adaptive bitrate (ABR) stack for a 13-second 4K VOD file.

Benchmarking Methodology

To identify the optimal instance type for our VOD workload, a comparison was made between the commonly used C5 and C6 instances with the VT1.3xl instance for transcoding 4K and 1080p VOD content. The two assets were encoded into various output targets, which will be discussed in the next section on Evaluation ABR Targets. The performance of the VT1.3xl, C5.9xl, and C6i.8xl instance types was assessed based on the encoding completion time.

In the accompanying screenshot from the AWS console, it’s evident that the VT1.3xl is the smallest option in the VT1 family. Despite VT1.6xl being comparable to C5 and C6 in CPU/memory specifications, we selected VT1.3xl for a more precise price/performance analysis.

Input Data Parameters

The following table summarizes key metrics of the source video files used for benchmarking encoding performance:

Evaluation of Adaptive Bit Rate (ABR) Targets

Adaptive bitrate streaming (ABR) technology is designed to optimize file streaming over HTTP networks. It allows various file sizes of the same content to be available to a user’s video player, enabling the client to select the most appropriate file for playback on their device. This process involves transcoding a single input stream into multiple output formats tailored to different viewing resolutions.

For our benchmarking tests, the input 4K and 1080p files were transcoded into multiple resolution targets to cater to diverse device and network capabilities, including 1080p, 720p, 540p, and 360p.

Results of Output Analysis

The VT1.3xl instance demonstrated superior encoding speed, finishing the targeted encodes 15.709 seconds faster than the C5.9xl instance and 12.58 seconds quicker than the C6.8xl instance. The charts below illustrate that VT1.3xl outperformed both C5.9xl and C6i.8xl in terms of speed and price performance.

For more insights on how to create value at work, check out this informative blog post. Additionally, for a deeper understanding of this topic, you can refer to the expert opinions of Nicole Van Valen. If you’re interested in further resources on learning and development at Amazon, visit this excellent resource.