Graphical applications on devcloud ?

[Plagne__Laurent]

Hi,

I wonder if there is a way to use graphical applications on devcloud ? It seems that there is no possibility to use X11 forwarding.

For the editor, I am able to use vscode remotely , but for vtune I have to manually scp the output of vtune command line and to visualize them on my own laptop (a bit tedious).

Is there an easier way for vtune-gui usage on devcloud ?

[Adweidh_Intel]

Hi Plagne,

Running VTune on the DevCloud

 For running VTune on the DevCloud, note the following:

1. VTune 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.

Also you could use VTune as a web server in addition to the classic desktop usage model. Kindly follow below steps :

1.   Launch Vtune web server from Devcloud by running the vtune-backend command:

cd <vtune-install-dir>/bin64/

vtune-backend --web-port=8880

2. Open the URL with the provided one-time token:

VTune Profiler Server outputs a URL to access the GUI. You can open the URL in browser on your local machine. In devcloud, you may need to do two-way tunneling to open the URL. For two way tunneling , open a new prompt and tunnel to login node of the Devcloud using the below command

ssh -L 8880:localhost:8880 devcloud_proxy

Then, tunnel to compute node as 

ssh -L 8880:localhost:8880 s001-n058

Where,

devcloud_proxy :- Host name

8880 :- Port in which vtune web sever is running

s001-n058:- compute node in which your vtune web server has started

3. Set a passphrase in the Set Passphrase dialog box.

Kindly find the below link for more information. Please feel free to reach us in case of any issues/queries.

https://software.intel.com/content/www/us/en/develop/documentation/vtune-help/top/launch/web-server-ui.html#web-server-ui

[Plagne__Laurent]

Thank you for this detailed answer.

Unfortunately I failed to apply the solution you proposed :

1 : I want to characterize a oneAPI executable running on a devcloud gpu. What would be the exact command to obtain the "APS" report ? (I tried vtune aps ./main)

2 : The webserver solution looks nice but I failed to follow your explanations.

*  I do not know if I need the "--allow-remote-ui" when I start the webserver on the computing node (is the computing node or the login node"

* I can't open the https adress on my local machine

* I don't really understand the double tunneling part...

* I do not see any dialog box for a passphrase.

Anyway, it looks too complicated for me to set up this each time I want to profile a kernel on devcloud.

[Adweidh_Intel]

Hi Plagne,

As you are trying to profile a GPU application I would suggest you to go with the Vtune web server option. Intel® VTune™ Application Performance Snapshot is mainly used for a quick view into different aspects of compute intensive applications' performance, such as MPI and OpenMP* usage, CPU utilization, memory access efficiency, vectorization, I/O, and memory footprint. You can refer more on https://software.intel.com/content/www/us/en/develop/documentation/application-snapshot-user-guide/top/analyzing-applications.html

Inorder to enable Vtune web server in your devcloud kindly follow the below steps:

Step 1 : Login to devcloud compute node.(Note the compute node name Example: s001-n011 )

Step 2 : Navigate to /opt/intel/inteloneapi/vtune/2021.1-beta08/bin64/

Step 3 : Start the web server by running ./vtune-backend --web-port=8880

Step 4 : Open another terminal and run ssh -L 8880:localhost:8880 devcloud then tunnel to the compute node using ssh -L 8880:localhost:8880 s001-n011 (Replace s001-n011 with your compute       node name). This is the two way tunneling part.

Step 5 : Now run the URL in browser which we received after starting web server(3rd step).Example : https://127.0.0.1:8880?one-time-token=******  You will be prompted to enter the passphrase once the url is accessed in your web browser, also you can set any passphrase of your choice. 

Note: You need not to pass --allow-remote-ui as you are accessing from localhost. 

Please feel free to reach us in case of any issues/queries.

Thanks,

Adweidh.  

[LaurentPlagne]

Thank you very much.

This time I was able to use vtune from the browser !

Unfortunately, the experience is not very good because it takes a very long time to filter the informations: after a long time to obtain the summary, I click on a given kernel and a spinning "loading" wheel starts and never ends...

On the double tunnel window, I see tons of error messages like:

error: [
'e.config.knobs.every is not a function',
'e.isInitialized@https://127.0.0.1:8880/perffilterbar/perffilterbar-610921.js:3:4079\n' +
'fn@https://127.0.0.1:8880/vendor-610921.js line 1 > Function:4:159\n' +
'$digest@https://127.0.0.1:8880/vendor-610921.js:1:186818\n' +
'$apply@https://127.0.0.1:8880/vendor-610921.js:1:188799\n' +
'$applyAsync/i<@https://127.0.0.1:8880/vendor-610921.js:1:188944\n' +
'nr/this.completeTask@https://127.0.0.1:8880/vendor-610921.js:1:209088\n' +
'ln/o.defer/i<@https://127.0.0.1:8880/vendor-610921.js:1:120644\n',
null
]

In addition, I see that I can't use the HW counters (on my own machine I don't have this issue because I launch vtune as root)...

[Adweidh_Intel · Accepted solution]

Hi Plagne,

We are sorry for the inconvenience caused, we are working on this internally meanwhile you can export the result to your laptop/desktop and view from vtune application.

Thanks,

Adweidh

[LaurentPlagne]

Yes, that is what I found the most suitable by now.

Thank you for your time !

[Adweidh_Intel]

Hi Plagne,

Thanks for your patience. We have tried to generate vtune report for your application using vtune webserver, where report was generated successfully without any error thorwn at kernel side, as you mentioned earlier its consuming lot of time for the graphics generation, we will provide feedback regarding this to the concerned team. As you were able to view the report using the workaround provided,shall we discontinue monitoring this thread ?

Thanks,

Adweidh

[Plagne__Laurent]

Yes !

Thank you again.

[Adweidh_Intel]

Hi Plagne,

Thanks for the confirmation. We would discontinue monitoring this issue, please raise a new thread if you have any further issues.

Thanks,

Adweidh