Cinegy Cloud Broadcasting

What might a cloud deployment look like using Cinegy? The scope can range significantly. It might be as simple as a single Cinegy Playout engine for specific events, or as comprehensive as a full product suite. This suite includes Cinegy Convert for ingest, followed by media management and editing through Cinegy Desktop and Cinegy Archive, and culminates in broadcasting with Cinegy Air (divided into Air automation and Playout Engine) and content recording using Cinegy Capture for Video on Demand workflows.

This diagram illustrates various components of Cinegy’s software. But how exactly do these integrate with cloud technology?

When we align the different elements of Cinegy software with cloud services, a seemingly intricate deployment simplifies into just four core cloud capabilities. These include cloud compute, addressing CPU/GPU needs; database services for data management; SMB storage service functioning as network shared storage; and object storage for original media.

Alternatively, Cinegy Archive can also operate on a cloud compute instance. This streamlined approach demystifies the complexity of cloud deployments.

When it comes to selecting the appropriate compute instances for various playout configurations in the cloud, we often encounter questions like:

Later in this document, we will provide practical examples of playout configurations alongside the corresponding cloud instance sizes and their load capacities.

For these tests, we utilize a typical cloud playout setup as our standard configuration for the Cinegy Playout engine. We’ll adapt this base configuration as needed, highlighting any significant modifications.

Examining the configuration details:

This configuration represents a balanced approach, providing a reliable foundation for testing various playout scenarios in the cloud environment.

Following the engine configuration, now it’s time to test on older, smaller cloud instances. Starting with one HD channel without graphics, we’ll monitor the instance’s CPU and GPU load. Gradually, we’ll add more channels, one by one, to observe the impact on resource usage. Before initiating these tests, we’ll record the idle state load of the instance for comparison.

Summarizing the results, we started with around 10% CPU, 6% GPU utilization and as we moved up adding more channels, it was the CPU that was maxing out at an average of 72%.

The GPU was only 42% and this demonstrates how we can deliver three simple HD channels from the smallest of the older generation cloud instances.

Could we add another HD channel to the instance? Yes, with a few tweaks:

These adjustments could enable one more HD channel to run smoothly on the existing cloud instance.

Moving from older to newer cloud instances faced a technical shift: NVIDIA’s newer chipsets don’t support interlaced encoding in H.264, necessitating a switch to progressive output. Question - how would this impact performance on the newer hardware?

To find out, we mirrored our earlier approach, adding playout engines one at a time on a newer instance. Despite the shift to progressive output, the results were promising:

We began at a lower baseline than the older instance (3% CPU and 1% GPU utilization). As we scaled up to four HD channels, the CPU peaked at about 83% - slightly higher than the older instance, but still manageable. The GPU maintained an average of 54%, indicating good performance.

These results underscore the advantages of newer cloud instances. While they’re not the latest hardware, they still offer significant improvements over older generations.

Integrating graphics into high-definition output raises an important question: how does this affect the load on the cloud instance?

We noticed that when the animated graphic starts, there’s an initial spike in CPU usage, reaching up to 42%, before settling back to 30% during the animation. The GPU follows a similar pattern, peaking at 18% and then reducing after the animation ends. These observations highlight how complex graphics temporarily elevate resource demands on the cloud instance.

Transitioning to Ultra High Definition (UHD) broadcasting raises a crucial question: can it be efficiently delivered? To set this up, we simply adjusted our engine configuration for UHD: switching to UHD player mode and increasing the output bitrate to 35 Mbps, still utilizing the H.264 codec.

In our trials with a newer generation cloud instance for a single UHD channel, the CPU load fluctuated around 30%, while the GPU maintained a steady utilization of about 14%.

Interestingly, the CPU load occasionally dipped to 16 -18%, and the GPU load slightly increased during playback. This variation was due to the use of two different video clips encoded with ProRes and HEVC respectively. The different encoding methods of these clips influenced the instance’s load.

The outcome is clear: delivering UHD content is not as challenging as it might seem. Even with media files requiring different codecs, the CPU load peaked at a manageable 60%, and the GPU load at 18%.

This demonstrates the efficiency and adaptability of Cinegy software, even on a cloud instance that isn’t the most advanced available. It reassures that running multiple UHD channels on a single instance is not only feasible, but also efficient.

Migrating workloads to the cloud involves extensive testing and exploration. It’s crucial to run repeated tests across various configurations. Maintaining both a development and a staging environment is a best practice, as it allows for the swift creation and deletion of test environments.

For automation, while cloud providers offer their own services, Cinegy prefers a more neutral solution. TerraformTerraform, an open-source infrastructure-as-code tool, is our go-to choice. It provides a consistent command-line interface (CLI) workflow for managing numerous cloud services. Terraform utilizes various file types for defining and deploying your environment’s components.

A notable feature of Terraform is its ability to document your infrastructure as code. This documentation can be stored in a version-controlled repository, like Git. This approach streamlines the creation and removal of environments, as Terraform tracks all deployments. It simplifies the deployment process, eliminating the need to remember the sequence of service deployments or prerequisites like Active Directory setup. Cinegy frequently uses Terraform for cloud deployments. We have even demonstrated its success on our YouTube channel, showing how to establish 150 playout channels in just 10 minutes. This showcases Terraform’s efficiency and our reliance on it for cloud-based operations.

Moving on to the aspect of monitoring, it’s crucial to understand its impact on cloud-based operations. As we’ve observed, even connecting to monitor resources can add to the load on a system.

Cinegy has long emphasized the significance of big data and analytics, which are incredibly valuable in providing both real-time and historical insights into system performance. This data helps in identifying trends and making informed decisions.

A powerful and cost-effective solution for this is the combination of Metricbeat and Elasticsearch for data collection and management, with Grafana for data visualization. This trio offers a robust and free-to-use system for monitoring.

What does this combination achieve? It enables the creation of detailed dashboards that offer a comprehensive view of system operations. For instance, you can clearly visualize GPU load management. Additionally, for more granular data, like CPU specifics, you can interact with the dashboard to get point-in-time statistics. This integration provides a thorough understanding of how your system is performing, enhancing the ability to make data-driven decisions in managing cloud resources.

You might wonder if it’s possible to delve deeper into monitoring. The answer is a resounding yes, especially with Cinegy Air. We have designed Cinegy Air to provide useful metrics that can be integrated with systems like Elasticsearch or Prometheus.

By harnessing these metrics, we can construct detailed graphs. These visuals offer insights into critical aspects of broadcasting, such as output frame drops, system timing variations, and media access delays. But we don’t stop there. By merging data from Cinegy Air and tools like Metricbeat, we achieve a comprehensive overview of both the hardware and software.

This holistic approach allows us to correlate any issues with the state of the machine at the exact moment of occurrence. This level of detail is crucial in cloud environments, where you don’t have control over the underlying infrastructure. Such insights are invaluable for pinpointing and addressing potential issues effectively.