ServerlessVideo: How AWS Is Running Point in the Video Market in 2024

In 2023, the streaming world was shaken when Amazon Prime decided to move away from serverless. Serverless architecture, also known as serverless backend or serverless computing, was considered to bring flexibility and scalability, improve efficiency, and free the platform operators' hands so they could devote their resources to the platform itself. However, due to several issues, Amazon had to move away from this architecture model. In this article, we'll briefly overview issues related to the serverless approach and explore where AWS stands with it today.

Moving Away From Serverless

As a platform that offers thousands of channel live streams to their customers, Amazon Prime needed to monitor their channel streams at a large scale (both audio and video) in order to take real-time actions. While Amazon has already developed tools for monitoring individual streams, the challenge was rooted in scaling up this process to handle the vast volume of streams concurrently. Hence, managing the monitoring operation at an industrial level was the key challenge here.

Part of the problem was the way video frames were passed around different components. To reduce computationally expensive video conversion jobs, the team has built a microservice that split videos into frames and temporarily uploaded those frames to an Amazon S3 bucket. Defect detectors, each of which also ran as a separate microservice, then downloaded frames and processed them concurrently using AWS Lambda. However, the high number of Tier-1 calls to the S3 bucket was expensive.

Orchestration was another problem. Orchestrating services, when operating at scale, inherently introduced friction where the system must scale up to match the throughput of its fastest component. Additionally, all the functions resided within the same business domain. This essential aspect influenced the architecture breakdown, as the team opted for a distributed system approach with numerous components to facilitate rapid iterations during the solution's development, likely due to unclear requirements and multiple reworks. Hence, adjustments became necessary as the application reached feature stability and required high scalability with real-time responsiveness.

To re-architect the audio and video quality inspection solution, reduce operational costs and address scalability problems, Amazon Prime moved the workload to EC2 and ECS compute services and created the tool for inspecting the quality of audio and video streams that could detect various user experience quality issues and trigger proper repair actions. As a result of this move, Amazon Prime achieved a 90% reduction in operational costs.

Where does it leave the streaming space today? And does the Amazon's move from serverless mark the fall of serverless itself? Let's see what's in here back in 2024.

AWS and Serverless Advocacy in 2024

At the end of 2023, AWS launched ServerlessVideo, a full-stack video streaming application that can be used to broadcast videos and that is apparently built on a serverless architecture.

ServerlessVideo can be used for both live streams and videos played on-demand (VOD). Along with the broadcasted video, the app uses Amazon Interactive Video Service (AIVS), which handles the streaming and ensures its interactivity. For instance, the same technology is used on Twitch, where viewers can send over their emojis and reactions during the live stream.

For VOD, the application goes through a very interesting architecture that is all about plugins. And basically, since the architecture is pluggable, nearly all kinds of other plugins can be contributed to this ecosystem thanks to AWS Step Functions and Amazon EventBridge. Generative AI plays an important role in this solution as well: the use of Amazon Bedrock and Amazon Transcribe allows the creation of descriptive video titles and the generation of video transcriptions automatically.

Besides actual video broadcasting enablement, ServerlessVideo automates image processing tasks, including background removal and dynamic resizing, by integrating AWS Step Functions, AWS Lambda, Amazon EventBridge, and AI and machine learning capabilities via Amazon Rekognition.

With renewed focus on cost, sustainability, and speed, ServerlessVideo promises to reduce image processing time from weeks to minutes and save substantial costs, as it did for the NFL. A solution for the NFL was built with AWS serverless technologies and Adobe Photoshop APIs. And I'm sure we'll see more examples that power up AWS serverless advocacy quite soon.

Embracing the Power of AWS with DataArt

As an AWS Advanced Consulting Partner, DataArt has solid experience with a broad variety of AWS expertise and technologies, backed by hundreds of completed projects.

One of them, specifically related to video streaming, is a startup that provides a multi-cam live streaming platform for bloggers, influencers, and content creators, helping them manage and monetize their video content. To stay ahead of the curve and engage video content creators and streamers, the client had to ensure that their web platform and mobile apps could effectively cope with high-resolution videos, provide smooth and lag-free camera switching, and generally provide a top-notch user experience.

To help the client reach their goals, DataArt implemented the following advancements:

• Organized collecting of all the streams from all the cameras on a single server to avoid hops
• Used UDP-based WebRTC and SRT to reduce latency and eliminate freezes and stuttering
• Applied AWS VGPU plugin + Nvidia MPS to ensure multi-processing (up to 10 concurrent processes)
• Reduced switching latency from 3-4 sec to 250 ms by forcing keyframes every 0.5 sec
• Used GStreamer to manage media streaming and switching
• Implemented WebRTC to deliver high-resolution video streams

Embarking Your Serverless Journey

At DataArt, we specialize in AWS solutions and help companies start their AWS cloud migration journey with DataBase Migration Assesment - a consultative engagement conducted by our experienced migration consulting engineers.

Check out DataArt's AWS Migration Checklist to help you make more informed decisions that align with your organization's strategic goals. Embark on your digital transformation journey with our guide and a team of experts ready to support you every step of the way.

Originally published here.