Matthias Loibl

Matthias Loibl is director of cloud at Dash0 and an open-source observability maintainer focused on making production systems’ resource consumption measurable and actionable. He previously led Polar Signals Cloud and joined Dash0 when it acquired Polar Signals in August 2026, incorporating continuous CPU, memory, and GPU profiling into its observability platform.

Based in Berlin, Loibl previously worked at Red Hat and Kubermatic and was a senior software engineer at Polar Signals before becoming director of its cloud product. His open-source contributions include Prometheus, Prometheus Operator, and Parca; he maintains the Prometheus project’s prometheus-mixin and is an emeritus maintainer of Thanos. He also maintains Pyrra, a Prometheus-based platform for service-level objectives and error budgets, alongside UX designer Nadine Vehling.

Technical contributions

• Always-on production profiling. Loibl uses Linux eBPF and low-overhead sampling to identify expensive CPU activity across production applications without instrumenting each service. Continuous observation captures recurring performance problems that local testing and isolated measurements can miss.

• Continuous GPU profiling. His approach to GPU efficiency correlates NVIDIA utilization, power, temperature, memory, and PCIe measurements with CPU call stacks, exposing accelerator underuse caused by slow data preparation, transfer bottlenecks, or thermal throttling. CUDA kernel timing connects work on the GPU to the application code responsible.

• Time-aware flame charts. Loibl extended chronological flame charts to CPU profiles, making garbage-collection pauses, uneven processor usage, lock contention, and short-lived latency spikes visible when aggregated flame graphs obscure their timing.

• Data-local reliability calculations. For Pyrra deployments using Thanos, he redesigned expensive SLO queries around local Prometheus preaggregation, reducing cross-zone data movement while giving users control over the accuracy trade-off.

• Evidence-grounded AI performance engineering. Loibl connected AI coding assistants to production profiling data through the Model Context Protocol. His MCP sandwich tool narrows analysis to a function’s callers and callees, limiting context consumption while grounding automated diagnoses in runtime evidence.

Loibl also helped organize PromCon in Berlin and runs the Berlin Prometheus Meetup.