Changes

Page history
Update home authored by van Vliet's avatar van Vliet
Show whitespace changes
Inline Side-by-side
home.md
View page @ 89c506dd
......@@ -1283,126 +1283,17 @@ will limit the number of simultaneously running tasks from this job array to 25
We recommend to use this option.
## Not possible anymore !! Remote GPU-accelerated Visualization on res-hpc-gpu01
## Remote GPU-accelerated Visualization on res-hpc-lo02
At this moment the login node res-hpc-lo01 does not have a powerful GPU card.
If you want to run a graphical program that will show 3D animation, movies or any other kind of simulation/visualization, we have reserved 1 GPU on res-hpc-gpu01 for this.
If you want to run a graphical program that will show 3D animation, movies or any other kind of simulation/visualization, we have the second login node **res-hpc-gpu01** for this.
This server has a powerful Tesla T4 GPU card (16Gb memory).
Steps for setting a up a remote GPU-accelerated visualization:
* login to res-hpc-gpu01
* start a vncserver
* connect from your remote desktop to the vcnserver
* start your visualisation program
* connect your remote desktop (X2Go) to the second login node res-hpc-lo02
* start your visualization program (with "vglrun" in front of it if needed)
### login to res-hpc-gpu01
As a normal user, you are not allowed to login to a node, in this case res-hpc-gpu01.
You have to create a small job which will allocate a reservation on res-hpc-gpu01, so that you are allowed to login to res-hpc-gpu01
Run:
```
srun -n 1 -c 1 -p gpu -w res-hpc-gpu01 -t 2:0:0 --mem=200M --pty bash
```
You are now logged in on res-hpc-gpu01
### create a vncserver setup and start a vncserver session
If you are setting up a vcnserver session for the first time, it will ask you a few questions and after that you have to adapt the configuration and startup file:
```
vncserver
You will require a password to access your desktops.
Password:
Verify:
Would you like to enter a view-only password (y/n)? n
Desktop 'TurboVNC: res-hpc-gpu01.researchlumc.nl:1 (username)' started on display res-hpc-gpu01.researchlumc.nl:1
Creating default startup script /home/username/.vnc/xstartup.turbovnc
Starting applications specified in /home/username/.vnc/xstartup.turbovnc
Log file is /home/username/.vnc/res-hpc-gpu01.researchlumc.nl:1.log
```
You should choose a strong password, the password should not be the same as the user login password.
Kill the vncserver connection:
* vncserver -kill :1
Now adapt the **xstartup.turbovnc** file:
* vi $HOME/.vnc/xstartup.turbovnc
```
#!/bin/sh
unset SESSION_MANAGER
unset DBUS_SESSION_BUS_ADDRESS
XDG_SESSION_TYPE=x11; export XDG_SESSION_TYPE
exec icewm-session
```
Adapt/create a **turbovncserver.conf** file for the vncserver with some useful settings:
* vi $HOME/.vnc/turbovncserver.conf
```
$geometry="1280x1024"
$depth=24
```
Now start the vncserver:
```
vncserver
Desktop 'TurboVNC: res-hpc-gpu01.researchlumc.nl:1 (username)' started on display res-hpc-gpu01.researchlumc.nl:1
Starting applications specified in /home/username/.vnc/xstartup.turbovnc
Log file is /home/username/.vnc/res-hpc-gpu01.researchlumc.nl:1.log
```
You can list you vncserver session with the following command:
* vncserver -list
```
vncserver -list
TurboVNC sessions:
X DISPLAY # PROCESS ID
:1 33915
```
You can/should kill your vncserver session when you are done with running your own application:
* vncserver -kill :1
```
vncserver -kill :1
Killing Xvnc process ID 47947
```
#### vncserver and port numbers
Everytime someone start a vcnsession and there is already a vncsession running, the port number will increase.
The first connection will be on **:1** which is port 5900 + 1, where the port 5900 is the standard port number for VNC.
For example:
```
Desktop 'TurboVNC: res-hpc-gpu01.researchlumc.nl:3 (username)' started on display res-hpc-gpu01.researchlumc.nl:3
```
In this case, the port to connect to is number **3**.
So you connect to "res-hpc-gpu01.researchlumc.nl:3".
You can always list your own open vnc session with the **vncserver -list** command.
Remember to kill you vnc session when you are done running your own application.
### connect from your remote desktop to the vcnserver
If you are running the **X2Go** client, open a terminal.
Then run the command:
```
/opt/TurboVNC/bin/vncviewer res-hpc-gpu01.researchlumc.nl:1
```
It will ask you for your vnc password:
![alt text](images/vncviewer-02.gif "vncviewer-02")
Or if you run: "/opt/TurboVNC/bin/vncviewer" without the hostname, you have to enter the hostname: "res-hpc-gpu01.researchlumc.nl:1"
![alt text](images/vncviewer-01.gif "vncviewer-01")
This will then open a (second) icewm windows manager. From there you can launch your accelerated visualization.
Once you are in your remote desktop, open a terminal
For the GPU acceleration, you have to run the VirtualGL command: **vglrun** in front of the real program you want to run.
Examples:
......@@ -1412,7 +1303,21 @@ Examples:
With the "glxinfo" program, you should check for the strings:
* direct rendering: Yes
* OpenGL renderer string: TITAN Xp/PCIe/SSE2
* OpenGL renderer string: Tesla T4/PCIe/SSE2
```
vglrun glxinfo | egrep "rendering|OpenGL"
direct rendering: Yes
OpenGL vendor string: NVIDIA Corporation
OpenGL renderer string: Tesla T4/PCIe/SSE2
```
```
glxinfo | egrep "rendering|OpenGL"
direct rendering: Yes
OpenGL vendor string: VMware, Inc.
OpenGL renderer string: llvmpipe (LLVM 9.0.0, 256 bits)
```
#### Programs that can run with "vglrun"
......@@ -1424,7 +1329,7 @@ How to run:
![alt text](images/fsleyes-01.gif "fsleyes-01")
Check for: **OpenGL renderer: TITAN Xp/PCIe/SSE2**
Check for: **OpenGL renderer: Tesla T4/PCIe/SSE2**
With the **nvidia-smi** command, you can also check if your program is running on the GPU.
Below you can see 2 programs running on the GPU: the Xorg server and the fsleyes program:
......@@ -1432,39 +1337,17 @@ Below you can see 2 programs running on the GPU: the Xorg server and the fsleyes
```
nvidia-smi
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 2 19234 G ...0.3/fslpython/envs/fslpython/bin/python 17MiB |
| 2 24920 G /usr/libexec/Xorg 17MiB |
| 0 N/A N/A 25565 G /usr/libexec/Xorg 25MiB |
| 0 N/A N/A 28059 G ...ng/fsl/FSLeyes/bin/python 2MiB |
+-----------------------------------------------------------------------------+
```
##### Bugs
* Problem: sometimes, if you start the program, the program seems to be very slow and is running at a frame rate of 1 frame per second
* Solution: stop the program and start it again. The second time it should run at normal/smooth speed
##### Remember:
* kill your own vnc session with "vncserver -kill:[port number] when you are done with your application
* do not close (exit or ^D) the **bash shell** which you got with the "srun" command before you are done with your application
* if you close your bash shell (srun), the vnc session will be terminated.
You will see something like this if you run the "vncserver -list" command:
```
vncserver -list
TurboVNC sessions:
X DISPLAY # PROCESS ID
Warning: res-hpc-gpu01.researchlumc.nl:1 is taken because of /tmp/.X1-lock
Remove this file if there is no X server res-hpc-gpu01.researchlumc.nl:1
:1 31636
```
In this case you have to manually remove the lock file.
### Remote visualization with a "reverse SSH tunnel"
You can skip a step in the above explanation about how to setup a remote visualization.
With a reverse SSH tunnel you can make a quick connection to a remote desktop.
We assume that you already have setup correctly your **vncserver**.
......@@ -1475,14 +1358,13 @@ We are using the SSH proxy server for this:
#### Setting up the reverse SSH tunnel
You have to follow the following steps:
* srun -n 1 -c 1 -p gpu -w res-hpc-gpu01 -t 2:0:0 --mem=200M --pty bash
* [on gpu01:] vncserver
* [on res-hpc-lo02:] vncserver
```
Desktop 'TurboVNC: res-hpc-gpu01.researchlumc.nl:1 (username)' started on display res-hpc-gpu01.researchlumc.nl:1
Desktop 'TurboVNC: res-hpc-lo02.researchlumc.nl:1 (username)' started on display res-hpc-lo02.researchlumc.nl:1
Starting applications specified in /home/username/.vnc/xstartup.turbovnc
Log file is /home/username/.vnc/res-hpc-gpu01.researchlumc.nl:1.log
Log file is /home/username/.vnc/res-hpc-lo02.researchlumc.nl:1.log
```
* [on gpu01:] ssh -R 8899:localhost:5901 -l username 145.88.35.10
......
......