Re: [herbert-cryptodev-2.6:master] [rhashtable] e1d3422c95: stress-ng.syscall.ops_per_sec 98.9% regression

[Breno Leitao <leitao@debian.org>]

Hello Oliver,

On Tue, Jan 07, 2025 at 02:35:44PM +0800, Oliver Sang wrote:
> hi, Breno Leitao,
> 
> > I am trying to reproduce this report, and I would appreciate some help
> > to understand what is being measured, and try to reproduce the reported
> > problem.
> > 
> > On Fri, Dec 27, 2024 at 11:10:11AM +0800, kernel test robot wrote:
> > > kernel test robot noticed a 98.9% regression of stress-ng.syscall.ops_per_sec on:
> > 
> > Is this metric coming from `bogo ops/s` from stress-ng?
> 
> yes, it's from bogo ops/s (real time).
> 
> one thing we want to mention is the test runs unstably upon e1d3422c95.
> as below, %stddev for it reaches 67%.

Thanks. From what I understand, this is clock time, which can vary a
lot.

I see a small variation, but, inside the standard deviation:

Kernels I've tested:
	Kernel A: 6.13-rc6 (9d89551994a43)
	Kernel B: 9d89551994a43 + cherry pick of e1d3422c95f003e ("rhashtable: Fix potential deadlock by moving schedule_work outside lock")

Average result: (Bogo op/s)
	* kernel A:
		* Average      : 2776.70
		* Min value    : 1946.22
		* Max value    : 3278.67 
		* Standard dev :  387.01

	* Kernel B:
		* Average      : 3158.60
		* Min value    : 1850.91
		* Max value    : 4120.10
		* Standard dev :  507.19

Host: Intel(R) Xeon(R) D-2191A CPU with 64GB of RAM.

I booted the machines 2 times, with A, B, kernel sequentially, and
for each case I run the following command 10 times:

	stress-ng --timeout 10 --times --verify --metrics --no-rand-seed --syscall 224