Every growing engineering org faces the same challenges when self-hosting CI. This talk shares lessons from building a high-performance CI cloud optimized for speed and scale. We’ll cover why bursty, single-core, I/O-heavy workloads need tailored infrastructure, how we used Firecracker microVMs for sub-3-second provisioning, gaming CPUs for compile performance, NVMe SSDs with copy-on-write file systems, and trade-offs like sacrificing durability to remove I/O bottlenecks. We’ll also look at efficient workload bin-packing for cost-effective scaling.
