Ubuntu
linux-aws package

  1. Jammy (22.04)
  2. Bug #2009325
  3. Comment #79

Comment 79 for bug 2009325

Revision history for this message
Allan G Soeby (soeby) wrote :

Hi Du ChengEn,

Thanks for you feedback, and understanding of our issue.

I will be watching the nfs mailing list as well, but kindly post references to the bug here, once the bug is opened.

I support your idea, of a separate mount-option, if it is not possible to address both issues.

Looking at the overall impact of this in our environment, we seem to be trading some GETATTRs for ACCESS calls. While the isolated increase in ACCESS calls is very high, the combined GETATTR+ACCESS is not as high. However, we are still talking about 4-5x here - 3-400% increase.

As our workloads are very meta-data intensive it will push our NFS servers into unacceptable load levels (~50k -> ~200k NFS ops).

Good news with regards to suggested increases in LOOKUPs, though. I cannot confirm any issues here. I have been running 5.15.69, 5.15.60 and your 1st test kernel (5.15.0-67-generic #74+test20230307b2h2cbb6062f8eb) for days. Once caches were settled, they are all hovering at the same levels of LOOKUP calls. So I see no regression on this matter.

Thanks in advance.