[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ba3be1f7-53ee-e794-700a-a9d5314b9c11@amd.com>
Date: Wed, 4 Jan 2023 12:06:44 -0600
From: "Moger, Babu" <babu.moger@....com>
To: Reinette Chatre <reinette.chatre@...el.com>, fenghua.yu@...el.com
Cc: tglx@...utronix.de, mingo@...hat.com, bp@...en8.de,
dave.hansen@...ux.intel.com, x86@...nel.org, hpa@...or.com,
corbet@....net, linux-kernel@...r.kernel.org,
linux-doc@...r.kernel.org, eranian@...gle.com,
peternewman@...gle.com
Subject: Re: [RFC PATCH 2/3] x86/resctrl: Move the task's threads to the group
automatically
Hi Reinette,
On 1/4/23 10:43, Reinette Chatre wrote:
> Hi Babu,
>
> On 1/3/2023 2:06 PM, Babu Moger wrote:
>> Some micro benchmarks run multiple threads when started. Monitoring
>> (or controlling) the benchmark using the task id is bit tricky. Users
>> need to track all the threads and assign them individually to monitor
>> or control. For example:
>> $stream_lowOverhead -codeAlg 13 -nRep 100000 -cores 0 1 2 3 -memMB 32
>> -alignKB 8192 -aPadB 0 -bPadB 0 -cPadB 0 -testMask 1
>>
>> $pidof stream_lowOverhead
>> 6793
>>
>> This benchmark actually runs multiple threads underneath on the cores
>> listed above. It can be seen with the command:
>> $ps -T -p 6793
>> PID SPID TTY TIME CMD
>> 6793 6793 pts/2 00:00:00 stream_lowOverh
>> 6793 6802 pts/2 00:01:25 stream_lowOverh
>> 6793 6803 pts/2 00:01:25 stream_lowOverh
>> 6793 6804 pts/2 00:01:25 stream_lowOverh
>> 6793 6805 pts/2 00:01:25 stream_lowOverh
>>
>> Users need to assign these threads individually to the resctrl group for
>> monitoring or controlling.
>>
>> $echo 6793 > /sys/fs/restrl/clos1/tasks
>> $echo 6802 > /sys/fs/restrl/clos1/tasks
>> $echo 6803 > /sys/fs/restrl/clos1/tasks
>> $echo 6804 > /sys/fs/restrl/clos1/tasks
>> $echo 6805 > /sys/fs/restrl/clos1/tasks
>>
>> That is not easy when dealing with numerous threads.
> How about:
>
> # echo $$ > /sys/fs/resctrl/clos1/tasks
> # stream_lowOverhead -codeAlg 13 -nRep 100000 -cores 0 1 2 3 -memMB 32\
> -alignKB 8192 -aPadB 0 -bPadB 0 -cPadB 0 -testMask 1
Yes, That works. We can use that as a work-around for now.
I will drop this patch for now. I will have to figure out how to handle
failures in the middle of task/thread movement.
Will have to revisit this in the future. Will re-submit patch 1 and 3.
Thanks
Babu
>
> Reinette
--
Thanks
Babu Moger
Powered by blists - more mailing lists