Depot Uses Simulation to Optimize CI Infrastructure Latency and Costs

According to a recent LinkedIn post from Depot, the company is using simulation and hyperparameter optimization to refine how it manages a standby pool of EC2 machines for continuous integration workloads. The post describes a tradeoff between keeping enough warm capacity to avoid latency spikes and minimizing idle-instance costs.

The LinkedIn post explains that Depot built a discrete-event simulator in Python using SimPy, seeded it with real customer CI job data, and calibrated it against production latency and cost metrics. This framework was then used with Optuna to run multi-objective hyperparameter optimization, seeking parameter sets that jointly minimize latency and infrastructure spend.

As described in the post, the optimization process reportedly produced a configuration that reduced p99 latency by about 2 seconds while also lowering standby pool costs by roughly 2%. The resulting policy was to maintain more machines in standby during daytime usage peaks and fewer at night, with the simulator providing a way to quantify risk before deployment.

The post also notes that the same simulation framework was reused for CI job scheduling, where testing suggested that assigning jobs to the first host with sufficient capacity outperformed a “run hot” strategy of loading the most-used host. This type of data-driven operational tuning may help Depot improve service performance and margin profile as scale grows.

For investors, the post suggests a focus on building internal tooling that can repeatedly evaluate latency–cost tradeoffs across infrastructure decisions. If sustained, such optimizations could modestly improve unit economics and support higher service reliability, potentially strengthening Depot’s competitive position in the CI infrastructure market over time.