Tridiagonal linear system: Intel Advisor and Intel vtune amplifier on the devcloud

[Oluwatosin]

Hello,

How do I launch and use Intel advisor and vtune amplifier on the devcloud ? Fot the vtune amplifier, I have tried to follow the vtune web server instructions, but I do not have very good network connection. Is there another way to launch it?

Kind regards,

Oluwatosin.

[AthiraM_Intel]

Hi,

Thanks for reaching out to us.

For running VTune/Advisor on the DevCloud, note the following:

1)VTune/Advisor can’t be used with a GUI. This is by design and not a bug.

2)The Intel DevCloud is a cluster with no windows subsystem – much like any other HPC cluster out there.

3)You can also use VTune APS “aps <command>” instead of the full vtune. This gives you a performance snapshot.

4) You can use vtune on DevCloud using vtune command line. Here is an example.

vtune-collect <analysis_type> [-target-system=<system>] [-knob <knobName=knobValue>] [--] <target>

please refer the following link to know more about command line syntax

https://software.intel.com/content/www/us/en/develop/documentation/vtune-help/top/command-line-interface/running-command-line-analysis.html

Once the results are collected you could view the results by copying the results directory to your local machine. And you can open and view the results on vtune-gui installed on your local machine.

Similarly you can use advisor on DevCloud,

advisor <--action> [--action-options] [--global-options] [[--] target [target options]]

please refer the below link for advisor command line syntax:

https://software.intel.com/content/www/us/en/develop/documentation/advisor-user-guide/top/command-line-interface/advisor-command-line-interface-reference.html

If you have any further issue, please let us know.

Thanks

[Oluwatosin]

Hello AthiraM_Intel,

I have tried to run a hotspot test. I got an error message and a folder rp000s-bad was also generated.

$ vtune -collect hotspots <path-to-executable>

>> vtune: Error: Cannot start data collection because the scope of ptrace system call application is limited. To enable profiling, please set /proc/sys/kernel/yama/ptrace_scope to 0. See the Release Notes for instructions on enabling it permanently.
vtune: Warning: Hardware collection of CPU events is not possible on this system. Microarchitecture performance insights will not be available.

[Stas-Neverov-Intel]

Hi Oluwatosin,

VTune Profiler server should work well for your setup. Please elaborate on network issues that prevent you from using the VTune server?

Thanks,
Stas

[Oluwatosin]

Hello Stas,

This kind of network problem-

client_loop: send disconnect: Connection reset by peer
client_loop: send disconnect: Broken pipe

It is my local network, it's mostly unstable. I get disconnected most times when working on the devcloud.

[AthiraM_Intel]

Hi,

Could you please try to profile the sample matrix application that is available along with Vtune.

You can find the sample application in the below path

<install-dir>/samples/en/C++/matrix

Steps :

•      Source <vtune-install-directory >/vtune-vars.sh
•      cd <vtune-install-dir>/samples/en/C++/matrix
•      vtune -collect hotspots -result-dir r000hs ./matrix

We are able to run the hotspot analysis in DevCloud without any issue

If you are still facing the same issue with the matrix application, please let us know.

Thanks.

[Oluwatosin]

Hello AthiraM,

Following the steps, I got this error:

>> vtune: Error: Cannot start data collection because the scope of ptrace system call application is limited. To enable profiling, please set /proc/sys/kernel/yama/ptrace_scope to 0. See the Release Notes for instructions on enabling it permanently.
vtune: Warning: Hardware collection of CPU events is not possible on this system. Microarchitecture performance insights will not be available.

However, I am able to use the vtune profiler server now, except that sometimes I can't access it or most of the time I get disconnected with this command line responses, that recursively displays on my screen, like using a while loop without a break

>> channel 2: open failed: connect failed: Connection refused

>> channel 3: open failed: connect failed: Connection refused

Using the server, I can view the sample matrix results, and perform analysis for other programs. I have some questions :

1. On the server, the CPU information for matrix results is the Intel (R) processor code named skylake (2.6 GHz, 8 logical cores); other analysis I have tried uses the Intel (R) Xeon(R) processor code named skylake (3.4GHz, 24 logical CPU cores). Is there a option to change processor type on the vtune profile server?

2. What are the names / model numbers for the Intel (R) processor code named skylake  (2.6 GHz, 8 logical cores) and Intel (R) Xeon(R) processor code named skylake (3.4GHz, 24 logical CPU cores) ? From wiki, based on the specifications given, the closest I can get for the second, is Xeon Platinum 8151, but I am not sure, I might be wrong

3. Can I access the Intel advisor using the web server method?

Thank you, I look forward to your response.

Kind regards.

[AthiraM_Intel]

Hi,

Regarding the error you are getting while analyzing the application,

Please make sure that you are in compute node to analyze the application in DevCloud.

You could reach an interactive compute node by using the below command, once you are connected to DevCloud.

qsub -I

After entering the login node try to profile the sample matrix application that is available along with Vtune.

You can find the sample application in the below path

<install-dir>/samples/en/C++/matrix

Steps :

• Source <vtune-install-directory >/vtune-vars.sh
• cd <vtune-install-dir>/samples/en/C++/matrix
• vtune -collect hotspots -result-dir r000hs ./matrix

Please find the below answers for your queries:

1. You can't change the processor type on vtune. However you can request different processor in DevCloud and profile the application using vtune webserver.

Please find the below DevCloud documentation to know more about the different processors available in DevCLoud.

https://devcloud.intel.com/oneapi/

2. Model name: Intel(R) Xeon(R) Gold 6128 CPU @ 3.40GHz, code named skylake (3.4GHz, 24 logical CPU cores)

Please find the below link for more details.

https://ark.intel.com/content/www/us/en/ark/products/120482/intel-xeon-gold-6128-processor-19-25m-cache-3-40-ghz.html

You could find the model name by using the below command after login to the compute node in DevCloud:

lscpu

3. You can't access intel advisor using webserver method. You could use the command line method as mentioned in the previous post. Once the results are collected you could view the results by copying the results directory to your local machine. And you can open and view the results on advisor-gui installed on your local machine.

If you have any further issue, please let us know.

Thanks

[Oluwatosin]

Thank you, my issue has been resolved.

[AthiraM_Intel]

Hi,

Thanks for the confirmation. Since your initial query is resolved, we are closing this thread and will no longer be monitored.

Regarding your last query we have moved that as a separate thread and your issue will be addressed in that thread. Please find the link to the new thread below:

https://community.intel.com/t5/Intel-DevCloud/Re-Re-Re-Tridiagonal-linear-system-Intel-Advisor-and-Intel-vtune/m-p/1291647#M2488

Thanks.