UFS and NVMe 存储 performance in embedded systems

Embedded 存储 systems have experienced a rapid pace of innovation in the last couple of years. While eMMC has been a mainstream 存储 of choice for many embedded solutions, recently introduced 存储 technologies — such as UFS and NVMe — are offering great interface speeds that allow systems to make use of the underlying speed of NAND technologies.

In embedded systems, the 存储 software (SW) stack has a profound impact on user-level speeds. UFS and NVMe differ greatly in their 存储 SW stacks. The UFS 存储 stack builds on the SCSI stack, whereas NVMe has a 存储 stack designed specifically for managed NAND devices, one that has been significantly simplified compared to the SCSI stack to realize high 存储 speeds at the user level.

在微米, we compare and quantify the differences between UFS and NVMe user-level 存储 performances. Our conclusion, based on measurements and estimation, show that due to the simpler 存储 SW stack, NVMe achieves 28% faster system-level speed in sequential writes and 15% faster speeds in sequential reads, and 30% higher IOPS in random writes and 16% higher IOPS in random reads. This inherent advantage of NVMe is expected to yield even greater benefits with faster 存储 devices.

Click here to read the complete white paper comparing UFS and NVMe 存储 stack and system-level performance in embedded systems.