Observability startup Dash0 has announced the acquisition of Berlin-based continuous profiling specialist Polar Signals. The primary technical outcome is the integration of advanced CPU time allocation tracking into SignalStore, Dash0's OpenTelemetry-native data platform.
Bridging Traces and GPU Kernels
Continuous profiling provides developers with an ongoing view of resource consumption within running applications. While many observability platforms now include basic forms of this capability, Polar Signals offered a distinct advantage: the ability to profile Nvidia CUDA workloads in production environments.
The acquired technology allows for visibility down to individual GPU kernels during both AI training and inference phases. This granularity helps teams pinpoint exactly where performance bottlenecks occur within complex deep learning models. By feeding this profiling data into Agent0, Dash0 enables its operations agent to correlate telemetry with source code repositories. Consequently, the system can prepare pull requests that address specific runtime issues identified in GPU execution paths.Storage Architecture and Data Unification
The acquisition extends beyond software capabilities; it involves a strategic shift in storage architecture for high-cardinality data generated by profiling activities. Dash0 is acquiring Polar Signals' Great Lakes storage engine, which was built specifically to handle the volume of profile data.
The stated goal is eventually replacing ClickHouse beneath SignalStore with this new engine. This move aims to unify metrics, logs, traces, and profiles within a single database instance rather than maintaining separate silos for different telemetry types. Additionally, Parca, Polar Signals' open source continuous profiling project, will remain under the Apache 2 license and continue maintenance by Dash0.Automated Optimization via AutoTune
Dash0 is developing a capability called AutoTune within Agent0. This feature analyzes scheduling profiles to identify opportunities for improving CPU and memory usage in production environments. When an optimization opportunity is identified, the system generates a pull request containing proposed changes backed by profiling evidence.
It is important to note that this automation does not merge code directly into repositories without human oversight; developers must review these proposals before accepting them.Related CloudNinjas coverage: DevOps.