Messages in this thread | | | Date | Mon, 11 Apr 2022 10:45:06 +0100 | From | Cristian Marussi <> | Subject | Re: [PATCH v3 00/21] x86/resctrl: Make resctrl_arch_rmid_read() return values in bytes |
| |
On Fri, Apr 08, 2022 at 06:30:40PM +0100, James Morse wrote: > Hi Cristian, > > On 15/03/2022 12:10, Cristian Marussi wrote: > > I tested this on an Intel(R) Xeon(R) Gold 5120T trying to compare gathered > > resctrl monitor data with and without your series and see if results > > were consistent. > > > > I started from this paper [0] from Intel itself for my basic setup with > > some minor variations: > > [..] > > > While doing that I collect in background (and out of node1 processors) all > > the mon_data from the lc_cos group every 100ms and dump those in a file one > > for each cache allocation mask. (mondata_LC_7f0_00f.txt etc) > > > > I tested first against a v5.17-rc1 mainline without your series (named as > > 5.17.0-rc1-mainline in the results) and then again with your series on top > > (named as 5.17.0-rc1-00021-g21c69a5706a5). Got your series from [1]. > > > > Then I used gnuplot to see what was the 'profile' of this data with and > > without your series by plotting the LC process llc_occupancy data against > > time for each one of the runs with the differerent cache allocated. > > (each colored graphs represent a different run with a different > > cache allocation as reported) > > > Hope this helps... >
Hi James,
> This is great, thanks! > > Can I take this as a Tested-by? >
Sure, sorry to have forgot that :D
Tested-by: Cristian Marussi <cristian.marussi@arm.com>
Thanks, Cristian
| |