Overview
Optimyze built Prodfiler, software that continuously sampled running systems to show where computing time was spent. It helped engineers find expensive functions across applications, libraries, and infrastructure without adding custom measurement code to each application.[1]
Elastic acquired Optimyze on November 1, 2021. Prodfiler became Elastic Universal Profiling, which remains part of Elastic's observability products.[2][3][4]
Founding Story
Thomas Dullien started Optimyze in February 2019 with Sean Heelan. They first offered consulting in exchange for a share of customers’ infrastructure savings. Dullien says small customers offered limited savings and larger companies struggled with contracting, so the pair shifted toward a fleet-wide profiler. They built a prototype in 2019 and raised funding in early 2020.[5][6]
The problem was practical: a cloud bill can reveal that a service consumes resources, but not which part of the code is responsible. A profiler samples a running program and aggregates those observations, helping an engineer decide where an optimization may matter. A sample is an estimate, so conclusions still require representative workloads and validation.
Timeline
- 2019–2020: The team builds an early prototype, raises funding, and begins production deployments.[6]
- 2020: a16z leads a seed round, according to contemporary reporting.[7]
- August 2021: Optimyze launches Prodfiler.[6]
- November 2021: Elastic completes the acquisition.[2]
- Later: The team ports the product to Elastic's stack and relaunches it as Universal Profiling.[6][3]
What They Built
Prodfiler used eBPF, a Linux technology that lets controlled programs observe kernel events, to collect profiling data across running systems. It sampled call stacks—the sequence of functions active at a moment—and associated them with readable function names. Engineers could aggregate those samples across services and machines to identify code that consumed substantial CPU time.[1][6]
Continuous collection matters because expensive behavior may appear only under a particular load. The team still had to keep overhead low enough for production and process a large volume of samples. Dullien describes rebuilding a slow query component before launch; easy installation did not remove the engineering cost of storing and querying the results.[6]
Market Position
Target Customers
Optimyze targeted engineering teams operating cloud infrastructure, especially teams seeking to understand CPU use across many services. Its own site presented customer examples from BioCatch and DataStax; these are company-selected testimonials, not an independent measure of typical savings.[5]
Market Size
The available sources do not establish a reliable dollar market size or number of paying Prodfiler customers.
Competition
Profilers were available from open-source projects and observability vendors. Elastic now offers Universal Profiling as a whole-system profiler and documents CPU stack sampling. A new product must compare itself with that active successor and other existing tools.[4]
Business Model
Prodfiler was offered as a software service. The reviewed sources do not establish its contract prices, gross margin, retention, or revenue. The founder described early production deployments before the 2021 launch.[3][6]
Traction
Dullien described customer enthusiasm for the product and its deployment experience. These are founder observations; the available sources do not establish customer count, operating results, or investor returns.[6]
Outcome and evolution
Profiling joined observability
Elastic said it wanted to combine profiling with logs, metrics, and traces, allowing engineers to investigate performance in the context of other operational data. Its annual report confirms that it acquired all of Optimyze's shares on November 1, 2021. Elastic reported $20 million in total consideration, including cash and future cash payments for certain unvested options.[1][2]
Elastic's current documentation describes Universal Profiling as an always-on Linux profiler and notes that it currently supports CPU profiling through stack sampling. Its product lineage continues, although the independent company no longer does.[4]
Key Lessons
- Measure the work behind the bill. Profiling connects resource consumption to code that engineers can investigate.
- Make deployment practical. Optimyze focused on collecting fleet-wide samples without custom instrumentation in each service.
- Build the analysis path too. Its team had to rebuild a slow query component before launch.
- Place data beside related evidence. Elastic saw value in combining profiles with logs, metrics, and traces.