NVMe TP4176

Overview

NVMe TP4176 is a technical proposal that specifies Quality of Service for PCIe bandwidth and IOPS per SSD controller, providing the standards foundation for multi-tenant SSD performance isolation and consistent, advertisable performance. It lets operators configure per-tenant IOPS and bandwidth limits so that each tenant on a shared physical SSD receives an isolated slice rather than relying on vendor-specific SSD modifications.

The Phase 2 agreed rate-limit parameters are Total BW, Total IOPS, Write BW, Write IOPS, Write-to-Read BW Scalar, and Write-to-Read IOPS Scalar, all set in real-world units (IOPS in thousands, bandwidth in MB/s). The proposal also defines a soft-vs-hard rate-limit toggle. As of October 2025 TP4176 is still in development and discussion within NVMe, and its current scope is limited to controller-level rate limiting for read and write commands only — identified gaps include deallocates, bursty workloads, and credit sharing across virtual functions. Despite being in-development, the standard already has production application: Google uses TP4176 for live migration to throttle VM traffic and achieve convergence.