I have joined KubeCon + CloudNativeCon North America from November 11–15, 2024, and had the chance to dive into some interesting sessions. Here’s a quick recap of two talks on Operations + Performance and Observability.
ARM-Wrestling: Overcoming CPU Migration Challenges to Reduce Costs
Speakers: Laurent Bernaille (Principal Engineer) & Eric Mountain (Staff Engineer), Datadog
Datadog shared their experience migrating workloads to ARM-based CPUs, aiming to lower costs and boost performance. With ARM adoption growing (think AWS Graviton), the talk emphasized the practicalities of future-proofing infrastructure.
Key Terms for beginner:
• ARM CPUs: These are processors designed with energy efficiency in mind, commonly used in mobile devices but increasingly popular in cloud computing (e.g., AWS Graviton) due to their cost and performance advantages over traditional x86 processors.
• Multi-Architecture Images: These are Docker images that include versions of the software for different CPU types (e.g., x86 and ARM) so they can run seamlessly on various systems without manual intervention.
Key Highlights:
• Preparation: They modified their Kubernetes clusters to work with ARM nodes, updating tools like kubelet and containerd.
• Challenges: Debugging was a recurring theme—issues like Go runtime bugs and libc incompatibilities needed creative fixes.
• Solutions: The team used multi-architecture images and cross-compilation to simplify deployments across ARM and x86.
Why It Matters:
ARM instances with AWS Graviton2 offer 40% better value than x86—more cost-effective and faster. However, achieving this requires careful planning and adjustments to your systems.
Here’s a simple workflow of how Datadog transitioned to ARM nodes:
Personal Takeaway: Datadog was upfront about the bumps in the road, which made their journey relatable. The process showed that even large, well-resourced companies need to iterate and adapt.
Measuring All the Costs with OpenCost Plugins
Speaker: Alex Meijer (Staff Engineer, Stackwatch)
This session introduced OpenCost Plugins, designed to measure and visualize Kubernetes costs. What stood out was Datadog being the first reference implementation, connecting cost data to their observability platform. It’s an interesting collaboration, but it also raised questions about balancing simplicity with flexibility.
Key Terms for beginner:
• FOCUS Specification: A standard set of billing fields created by the FinOps Foundation to make cost data consistent across tools. It ensures plugins provide the right information to OpenCost without needing deep technical knowledge of the platform.
• Plugins: Add-ons that let users customize how they track costs in OpenCost by connecting it to specific data sources (like Datadog). Plugins allow flexibility without altering OpenCost’s core functionality.
Key Highlights:
• Core vs. Extended Interfaces: The plugin design simplifies common use cases while allowing advanced customizations for those who need it.
• Community Contributions: OpenCost actively encourages user-built plugins, even offering a bounty program to drive adoption.
• FOCUS Specification: This schema standardizes cost data, making it easier for tools like Datadog to integrate and provide consistent results.
Why It Matters:
With cloud costs growing, having tools like OpenCost Plugins can be a game changer. They make cost tracking straightforward while leaving room for deeper analysis. The integration with Datadog adds another layer of value, though it feels more like a first step than a fully matured solution.
Personal Takeaway: The talk felt empowering, but also a little ambitious. While the collaboration with Datadog shows potential, it might take time to see widespread adoption or seamless integration.
Conclusion:
Both talks touched on how tech teams can optimize their workflows—whether by cutting costs with ARM or simplifying cost tracking with OpenCost Plugins.
Top comments (0)