[lug] Debugging I/O performance

[Nick Golder]

On Mar 13, 2018, at 10:13, Rob Nagler <nagler at bivio.biz> wrote:
> 
> We have some Dell m620s in a m1000e chassis. I was switching around disk controllers to figure out if the Dell H310 (non-RAID) worked better than the S110 (non-RAID or AHCI) was faster. One of the m620s has twice as slow disk I/O, and I can't figure out why.
> 
> There are no errors in the logs. lshw, hdparm, etc. are identical. I walked through the BIOS config. Also, identical. Simple CPU benchmarks are identical. Fresh installs with the same CentOS 7 kickstart config (minimal install).
> 
> Many details have been documented here: https://github.com/radiasoft/devops/wiki/DellRAID#performance <https://github.com/radiasoft/devops/wiki/DellRAID#performance>
> 
> The two identical machines are f2 and f3 with H310s. f4 is almost the same but uses an onboard controller (AHCI).
> 
> I can afford to swap around m620s with some in our diskless cluster, which would avoid the problem. However, it seems to me there's something wrong, and I should understand what it is.
> 
> Any ideas how to determine what's causing the performance difference?
> 

Are the BIOS versions in lock-step? We’ve seen APIC order and ACPI MADT table changes in BIOS updates that have affected PCIe performance.

Nick
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lug.boulder.co.us/pipermail/lug/attachments/20180313/ba1afd47/attachment.html>