Add Bookmark(s)



Bookmark(s) shared successfully!

Please provide at least one email address.

Microchip background

High-speed storage in a hyper-converged solution – enabling new possibilities.

High-speed storage in a hyper-converged solution – enabling new possibilities.

1.2 Tib/s Storage Spaces Direct: Mellanox, HPE and Micron Show NextGen All-Flash Data Center

Mellanox, Hewlett Packard Enterprise (HPe) and Micron recently collaborated to redefine “fast” – as in fast networking, fast throughput, fast data.  How fast was fast? 1.2Tib/second on a 12-node cluster fast!

More than the sum of the parts

To reach this performance level, the teams selected each component carefully:

  • Windows Server 2016 Datacenter Storage Spaces Direct
  • HPE Proliant DL380 Gen9 Servers
  • Mellanox ConnectX-4 100 Gb/s RoCE adapters
  • Mellanox Spectrum Ethernet Switches and LinkX cables
  • Micron 9100 MAX 2.4TB NVMe SSDs

And the result is more than the sum of the parts: extreme performance and extreme scale.  Take a look at the numbers:

CSV FS IOPS Reads Writes BW (MB/s) Read
Total 304,423 304,411 17 159,328 159,327
ignite1 26,680 26,678 1 13,734 13,734
ignite10 22,345 22,342 2 11,714 11,714
ignite11 28,452 28,450 2 14,897 14,897
ignite12 26,028 26,028   13,646 13,646
ignite2 23,857 23,856 1 12,507 12,507
ignite3 23,257 23,256 1 12,193 12,193
ignite4 29,391 29,390 1 15,409 15,409
ignite5 23,180 23,177 2 12,152 12,152
ignite6 24,177 24,175 2 12,675 12,675
ignite7 24,058 24,056 2 12,612 12,612
ignite8 26,823 26,822 1 14,062 14,062
ignite9 26,182 26,181 1 13,726 13,726

The Servers Still Had Plenty of Headroom!

The servers were barely breathing hard during this testing, leaving plenty of CPU resources for other tasks.  With 1.2Tb/sec of sustained storage traffic running across the cluster, each node in the cluster consumed only 25% of available CPU capacity leaving plenty of power to perform additional computer tasks. 

Results, results, results!

This joint Windows Server 2016 demonstration simulates application and user workloads using DiskSpd and VM Fleet* across a 12 node cluster using Storage Spaces Direct, Hyper-V, Windows Clustering, local high performance NVMe SSDs and 100GbE RoCE - eliminating complex and costly external shared storage requirements from Windows Server 2016 SDS solutions. The demonstration used DiskSpd to generate 512KiB read requests from each of the 336 Windows Server 2016 Server Core VMs in the cluster, controlled by VM Fleet.  

Want to learn more?

Take a look at the Solution Brief or if you have questions about the testing we ran, tweet me @GreyHairStorage or engage with us @MicronStorage and on LinkedIn.

About Our Blogger

Doug Rollins Doug is a Senior Technical Marketing Engineer for Micron's Storage Business Unit, with a focus on enterprise solid state drives.
Login or Sign Up Now for an account to leave a comment.