How To Optimize Video Streaming At Scale with Wowza & Graviton

I recently hosted a webinar with Mario Popish, Sr Partner Solutions Architect at AWS, and Viesturs Proškins, Head of Video at Playtech. During this session, we highlighted the shift in streaming infrastructure from traditional x86 to ARM-based AWS Graviton processors. The core message: Graviton offers superior efficiency, performance, and cost savings, particularly for compute-intensive video streaming workloads.

This blog recaps what we discussed on this webinar, for the detailed discussion we encourage you to watch the full VOD.

Comparing x86 with ARM-Based Graviton Instances

During the webinar, we identified several key insights that provide a rationale for Graviton and ARM-based devices when compared to x86 instances.

1. Graviton processors provide better performance at lower power requirements and lower costs.
2. In Graviton, each vCPU is a physical core with no symmetric multi-threading (or hyper-threading like Intel).
3. Workloads can safely push Graviton utilization to the 90% mark (compared to typical 75% for scale-out policies) without impacting P99 performance metrics. 
4. Graviton instances generally offer more network bandwidth available compared to x86 instances of the same size (up to 20% more in Playtech’s experience).
5. Wowza Streaming Engine benchmarks showed the Graviton 4 to be “super high performance,” handling 22 inputs and 66 encodes on a 60 FPS 1080p stream.

Graviton’s ARM-based architecture is designed for compute-intensive tasks like streaming and AI. The physical cores are crucial for consistent workloads like video transcoding, allowing each thread to run on its own. Wowza Streaming Engine is optimized for Graviton and ARM in general, using ARM-optimized libraries for codecs. With Wowza, Graviton can “sweat” and maximize efficiency, using less energy and making it more sustainable environmentally and financially.

ARM also unlocks new possibilities for edge and micro-edge workloads, such as running on devices like drones, offering incredible processing power on a very small footprint with very low battery consumption. For on-prem environments, ARM helps with rack capacity and real estate limitations. Density (channels per rack unit) is significantly improved, which, combined with lower power consumption and heat generation, is critical for optimizing data centers.

How Playtech Migrated Video Workflows to AWS Graviton & Wowza for More Efficient, Performant Streaming

Customers like Playtech report consuming less watts and spending less dollars per stream. Despite double-digit year-on-year growth, Playtech kept its AWS spending “relatively flat” by systematically optimizing costs.  Viesturs Proškins, Head of Video at Playtech, shared a practical, real-world roadmap for migrating video workflows to AWS Graviton, focusing on risk mitigation, performance optimization, and Total Cost of Ownership (TCO).

Playtech’s Migration Approach & Strategy

For those starting a new cloud journey, Viesturs challenged the audience to attempt going Graviton first. The migration required “some homework” on their CI/CD pipelines to produce both x86 and ARM64 compatible images (AMIs), which was not considered a major change. Playtech began their migration with side services that are not production-critical, such as the observability stacks. This approach helped them take initial risks with more confidence.

Instead of multiple Availability Zones, Playtech’s fundamental design uses multiple active-active regions. This architecture allows them to introduce new features (like Graviton) in one of the active regions. If anything goes wrong, they can literally turn that region off, and the other active region takes over the load. They rolled out Graviton to 50% of their operating regions following this architectural decision.

Playtech’s Real-World Outcomes

Playtech’s scaling trigger is usually network-out capacity, not CPU. They found that Graviton instances generally have more network bandwidth available, up to 20% more for the same instance size compared to x86 instances. This allows them to have less instances running because each one is “maxed out significantly more.”

Graviton also safely supports 90% CPU utilization, unlike x86 systems where 75% often triggers scaling out. This is because the Graviton’s P99 performance metrics (on whatever dimension the organization tracks) do not shift even under this significant load. Viesturs’ number one takeaway was to “Let that graviton sweat.”

Playtech is taking a tactical pause to observe the 50/50 split of regions. They are aiming for 98-99% coverage when it comes to compute migration to Graviton for their media payloads running Wowza. Viesturs stressed that TCO should be the most important aspect for decision-making, covering more than just raw instance price.

Building Your Media Technology Stack with Wowza & Graviton

To summarize, if you are looking to migrate or adopt Graviton into your media workflows, you can minimize risk while realizing the full benefits by:

• If you’re new to the cloud, Go Graviton First
• Adopt A Gradual Migration Strategy, starting with non-critical services
• Use Multiple Active-Active Regions to isolate risk
• Update CI/CD Pipelines for both x86 and ARM/ARM64
• Start Small And Test Often before gradually transitioning load over

In today’s day and age, TCO is key. By weighing factors like raw instance price, rack space (for on-prem), power usage, and availability/SLAs, organizations can significantly reduce costs while achieving the same or better performance. AWS and Wowza both have dedicated programs and specialists to help customers with their migration journey. For more detailed insights, Playtech’s full story, or in-depth benchmark comparisons, watch the full webinar VOD here. And if you’re ready to talk to a Wowza expert about how you can build more efficient, intelligent, reliable video streaming workflows, contact us.