RAID Performance Analysis on Intel® Virtual RAID on CPU (Intel® VROC)

Documentation

Maintenance & Performance

000032376

09/12/2019

This white paper provides an analysis of RAID performance and describes:

  • the mathematics behind RAID performance.
  • the specific RAID implementations of Intel® Virtual RAID on CPU (Intel® VROC).
  • how operating systems, storage devices, and other components impact performance output. Below is a sample of the material from the white paper. Please see the entire paper for the full details.

RAID Performance Analysis on Intel® VROC (PDF) PDF icon
Size: 360 KB
Date: January 2019
Revision: 002

 

For additional information specific to Linux* and Windows*, please see:

Intel® Virtual Raid on CPU Windows* Performance (PDF) PDF icon
Size: 226 KB
Date: January 2019

Intel® Virtual Raid on CPU Linux Performance (PDF) PDF icon
Size: 532 KB
Date: January 2019

Note: PDF files require Adobe Acrobat Reader*.

 

Scope

RAID storage solutions provide users with a combination of data protection and performance acceleration. Each RAID level processes storage I/O in a different manner and stores data in a specific pattern across a set of RAID member disks. Our goal is to highlight those storage patterns for RAID levels 0/1/10/5 and explain how each pattern affects the performance of the storage solution. By understanding the underlying I/O process, we can understand the theoretical performance maximum for each RAID level. We'll explore how Intel VROC implements these RAID levels in an actual product. We'll see how Intel® VROC’s performance compares to the theoretical maximums laid out. In the future, we'll explain why any performance deltas exist between Intel VROC and the theoretical maximums caused by the Intel VROC RAID engine. We'll look at OS limitations, SSD architecture, and more.

Target audience

Data Center Administrators, Architects, and Managers; Storage Performance Testers; Server Platform Manufacturers.

RAID mathematical performance model

RAID mathematical models define the maximum theoretical performance of RAID systems, not including limitations imposed by hardware or software. In fact, some of these numbers may not be achievable in real-world settings due to hardware or software limitations. In such cases, an explanation will be provided later in this document.

Summary table

RAID mathematical models defining the maximum theoretical performance of RAID systems.

OS performance impacts on Intel VROC

In some cases, Intel VROC performance doesn't meet the theoretical limits explained in the section above. Sometimes this is due to the impact that the OS has on Intel VROC. In Windows, due to Intel VROC RAID Engine efficiency IOPS numbers are currently limited to:

  • ~1M IOPS for single RAID volume
  • ~1.4M IOPS for multiple RAID volumes

Conclusion

Understanding RAID fundamentals is critical to properly evaluating a storage solution. Each RAID level offers a trade-off between redundancy, performance, and cost. This document can be used to pick the right RAID level for specific storage needs. Furthermore, Intel VROC is a hybrid RAID solution that delivers RAID functionality to the NVMe. Using NVMe SSDs connected directly to the CPU, Intel VROC enables optimized RAID configurations that approach many of the theoretical maximums explored above.

Related topic
Intel® Virtual RAID on CPU RAID Scaling