Optimizing Cost and Performance with arm64

Honeycomb.io, a Series B startup in the observability space, evaluated the arm64 processor architecture in order to improve cost and performance of its telemetry ingest and indexing workload. Over a year, 92% of all its compute workloads migrated successfully to arm64, cost of compute dropped by 40%, and end-user visible latency improved modestly. However, the journey was not without roadblocks and challenges such as lack of full software compatibility, hidden performance quirks, and additional complexity. This talk describes the process of setting up the evaluation, full migration, and improvements made to the ecosystem to make the workload run smoothly at scale in the end.