My goal is to obtain good writing performances on SSD NVME Disks (benchmarking tools seem to give performances in specific optimized context but far from what we can obtained in real life)

My server: 32x Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz RAM : 256GB DDR4 HD : 2x 1,5TB SSD NVME MICRON 9200 => RAID0 HD : 1x 1,5TB SSD NVME MICRON 9200 => DISK1 OS : debian 9.5 file system : XFS

I am using cp and dd to measure bandwidth of files copy from RAMDISK (30Go of /dev/urandom) to DISK1 and to RAID0 I am monitoring the writing operation with iostat I am using bs=1M and oflag with dd

a) with dd from RAMDisk to RAID0 without oflag=direct I get 15Gb/s and with oflag=direct I get nearly 20Gb/s. GOOD!

b) with dd from RAMDisk to DISK1 without oflag=direct I get 15Gb/s and with I get 13 Gb/s. I am not sure to understand the inversion

c) with cp from RAMDisk to RAID0, performances are poor and iostat samples every 2s shows 4 seconds without write between each write on RAID0

What is the best way to measure RAID0 SSD NVME in a realistic way without fluctuation)

Kind regards

  • I think 15Gbps for a single disk is not possible. Make sure you sync to clear write buffer after the dd copy. – Elliott B Feb 28 '20 at 6:17

You can try to use FIO for testing the storage and send fstrim command to the RAID-0 SSD pool prior to tests. Links for ref -

https://linux.die.net/man/1/fio or https://fio.readthedocs.io/en/latest/




Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.