Tu sistema no es lento, lo estás midiendo mal
When performance drops, most teams scale up by adding more resources, GPUs, and cost, often without real gains.
In this session, we’ll show how to move from reactive scaling to a data-driven approach by understanding workload behavior and challenging misleading signals like CPU utilization.
You’ll learn a practical framework to optimize performance, cost, and efficiency for your workloads.
Most performance issues are solved the same way: add more CPU, more memory, or scale out. It works... but it’s often an expensive shortcut that hides the real problem.
In this session, we’ll explore a practical approach to performance engineering using open-source tools and reproducible methods. Instead of relying on intuition or single metrics like CPU usage, we’ll break down how to characterize workload behavior (CPU, memory, and I/O patterns), identify system limits through breaking-point testing, and measure efficiency using cost per unit of work.
Through different examples, we’ll analyze how multiple workloads behave under load, why average latency is misleading, and how to detect the “knee of the curve” where systems begin to fail. We’ll also dive into hands-on optimization techniques (including memory-efficient patterns, caching strategies, and SIMD vectorization) showing how code changes can outperform infrastructure scaling.
You’ll leave with...
A clear mental model to understand your workload behavior
A practical, repeatable framework you can apply immediately to decide when scaling is necessary...and when it’s just unnecessary spend.
Specialist Solutions Architect at AWS with roots in infrastructure and a passion for efficient compute, cloud, and performance engineering. From co-founding a trailblazing startup to architecting services across major cloud platforms, Guillermo bridges the gap between low-level optimization and real-world impact, helping teams build faster, leaner, and smarter.