3

We are testing the speed of two Intel DC P3600 NVMe drives on a CentOS 7.1 system. They are configured in a software RAID1 using mdadm.

After upgrading the kernel from the (very old) 3.10 to the latest 4.3 (or 4.2.5) we see can observe a very dramatic read speed degradation - from 4GB/s to max 500MB/s. We can replicate the test by selecting the kernel version at boot time.

The tests are conducted using fio with using the following command:

fio --name=read --ioengine=libaio --iodepth=16 --rw=read \
    --bs=128k --direct=1 --size=32G --numjobs=8 --runtime=300 \
    --time_based=1 --group_reporting

What change in the Kernel could cause this? Everywhere I read, the upgrade is even recommended as the NVMe driver reached version 1.0 after kernel 3.19. Any suggestions would be more than welcome.

Improve this question
1
  • Is Centos 7.1 with a 4.2.5/4.3 kernel supported by Centos? Serverfault is for professional sysadmins keen to run supportable platforms, perhaps superuser.com might be more appropriate for more 'experimental' scenarios
    – Chopper3
    Nov 4, 2015 at 18:10

1 Answer 1

Reset to default
2

Red Hat kernels are quite different from vanilla ones. Even tunables (eg: sysctl) have significant different default values.

I strongly suggest you to stay with RH kernels unless absolutely necessary. Please at least consider using ELRepo kernels, if you really need a more update kernels.

Anyway, this question should be issued in the linux kernel mailing list, I suppose.

Improve this answer
1
  • We did use the ELRepo kernel, for production we intend to stay on the official ones. NVMe developer documents suggests upgrading the kernel to a version greater than 3.19 as it should have better driver support. Thanks for the answer
    – markz
    Nov 4, 2015 at 18:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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