COSBench

COSBench tests object-storage services by generating operations such as uploading and retrieving objects. You configure the workload, including object sizes and the number of concurrent requests, then observe how the storage behaves. It is useful when you need to compare systems using the operations your application actually performs. A meaningful result includes errors and response times as well as throughput, and the test should use disposable data rather than production objects.

Deployment and operating notes

The recovered chart packages Intel’s COSBench object-storage benchmark. Its own README says it is not a product with a full-time support team and expects users to investigate bugs and contribute. This review verified the project identity and benchmark purpose, but did not establish a supported current Java, chart or cloud-authentication combination.

Use a dedicated bucket and credentials with access only to disposable test objects. Define the object-size distribution, concurrency, read/write mix, duration and warm-up period before comparing systems. Check that the chosen adapter implements the authentication and API behavior of the target service; retries and failures can distort a throughput number. Record client CPU, network limits, error rates and latency percentiles alongside server metrics. Start with a small run and verify cleanup so an old workload file cannot delete production objects or create uncontrolled storage cost. If replacing COSBench, reproduce the workload semantics rather than comparing different tools’ headline requests-per-second figures.