This epic covers the planning, execution, and analysis of performance testing for QEMU's multiqueue block layer (io_uring) in OpenShift Virtualization. It aims to measure performance gains for VMs using high-performance storage (SSD/NVMe) and identify potential regressions or limitations.

Goals:

• Determine the performance improvement (throughput, IOPS, latency) provided by the multiqueue block layer for various I/O-intensive workloads.
• Validate that the feature does not introduce any performance regressions for existing workloads or storage configurations.
• Provide recommendations and guidelines for usage of the multiqueue feature.
• Identify any bottlenecks or areas for further optimization.

User Stories

• As a Cluster Admin, I want to measure and compare VM I/O performance with the multiqueue block layer both enabled and disabled, so that I can decide whether to enable it across the cluster for optimal performance.
• As a VM Owner, I want to compare multiqueue-enabled storage performance with traditional virtio-blk, so that I can determine which configuration best meets my VM’s workload requirements.

• As a Project Admin, I want to conduct long-duration stress tests on VMs with multiqueue enabled, so that I can verify the stability and performance of my project’s applications under sustained load.

• As a Cluster Admin, I want to monitor CPU, memory, and storage utilization when multiqueue is enabled, so that I can plan capacity and ensure resources are allocated efficiently across the cluster.

Non-Requirements

• Functional testing of the multiqueue block layer.
• Performance testing of other storage features not directly related to the multiqueue block layer.
• Performance tuning if there is not configuration to use at the virtualization layer.
• Identify optimal queue depths and I/O thread configurations for various workloads.