Skip to content
This repository has been archived by the owner on Sep 12, 2024. It is now read-only.

Latest commit

 

History

History
57 lines (42 loc) · 5.7 KB

gpu-visualization.markdown

File metadata and controls

57 lines (42 loc) · 5.7 KB
Raw
layout
default

Remote Visualization Guide for HPC Cloud

In this guide we will describe how to setup and use remote interactive visualization on the HPC Cloud. It allows the user of the HPC cloud to fully use virtual machines equipped with GPUs for interactive rendering, with the result being transferred over the internet to their own local laptop or desktop.

Remote interactive visualization

Getting access

By default, you will not have access to the GPU nodes of the HPC Cloud. In order to get this, please send an e-mail to helpdesk@surfsara.nl.

If GPU access has been enabled for your account, you will have to use a datastore that is enabled on the GPU nodes. This datastore is called gpu_images_ssd and behaves the same way as the local_images_ssd datastore, except that it is accessible on the GPU nodes only. In other words, using this datastore makes sure that your VM is run on the nodes which have GPU's installed on them.

Note that the ceph datastore is also enabled on the GPU nodes. However, as described here, it is best to put your OS image on either the gpu_images_ssd or the gpu_images_ssd datastore. ceph can then be used for larger data images and for persistent data.

OpenGL rendering

Normally, an OpenGL-based visualization application running on your local machine will send OpenGL rendering commands to the GPU and let it handle the rendering and displaying of the output. On the Cloud, this is slightly different in this respect, as the rendered output needs to be transferred back to the user. To accomplish this in a fairly user-transparent manner the packages VirtualGL and TurboVNC are used. VirtualGL provides a way to intercept the rendering output of an OpenGL-based application with high performance, while TurboVNC is capable of compressing and serving that output on-the-fly. OpenGL applications do not have to be modified in any way in this scheme, but the user does need to use an extra command when starting applications, which will be described below.

Setting up Remote Visualization on the HPC cloud

You can either set up your own cloud image, or use an image provided by SURFsara. In this guide we will use one of the standard images provided by SURFsara.

  1. In the Dashboard on https://ui.hpccloud.surfsara.nl/, under Storage -> Apps, select the Ubuntu 16.04 image with GPU (Ubuntu 16.04 GPU VIS) and import it with the Import button in the upper right. import AppMarket Appliance
  2. In the next screen, make sure to select “gpu_images_ssd” as your datastore, and give both your image and template a distinctive name. import AppMarket Appliance
  3. Under Storage-> Images, select the image you just created and, with the “dots” button, set your image to persistent state (Make persistent). After this, refresh the list with the button of the two arrows chasing each other and make sure the status of your image is “ready” Make image persistent
  4. Under Templates-> VMs, select the template you created, and click the “Update” button. This will open the properties of your VM template. Here you can set memory and cores of your virtual machine. A good default is 16GB memory and 4 CPU&VCPU. Update VM
  5. Next, under the “Other” tab, click “+ Add PCI Device”, and select the GK104GL [GRID K2] GPU. Make sure to click the green “Update” button to save your changes to the VM Add GPU to template
  6. Under Templates -> VMs, select your template, and click the “Instantiate” button, and in the pop up window click the green “Instantiate”, to start your VM. Start VM
  7. Under Instances -> VMs wait for your machine to boot. If the status is “running”, you can proceed with the next step. Note the IP address your machine received.
  8. Using SSH, login to your new machine, using ssh ubuntu@<ip address>
  9. You need to configure the VirtualGL context: sudo /opt/VirtualGL/bin/vglserver_config Answer in order: 1, n, n, y, X.
  10. The VM by default runs a firewall. Use the command sudo ufw allow from <your home IP address> to any to allow access from your own client IP address. If you have a dynamic IP address, you can specify a range using the / syntax, e.g. 145.100.1.0/24, which allows all ip addresses starting with 145.100.1.*
  11. Reboot using the command sudo reboot for the changes to come into effect
  12. On your own local machine, you need a VNC client. We advice TurboVNC (https://sourceforge.net/projects/turbovnc/) or TigerVNC (http://tigervnc.org/) but any other VNC client will suffice.

Workflow

You only need to do the previous steps once. For every subsequent work session, you need only need to reinstate your VM and start your virtual desktop as shown in the steps below:

  1. Log in onto your VM using SSH. Note that you most likely will have to use a different IP address each time you reinstate your VM
  2. Start your virtual desktop with /opt/TurboVNC/bin/vncserver -geometry 1920x1200. You can vary the resolution as desired using the -geometry option.
  3. Start the VNC client on your local machine, and login to your VNCserver: vncviewer <your VM's ip address>:1
  4. You should now have access with your VNC viewer to the remote desktop on the cloud node
  5. (Optional) To test if everything works correctly, you can use the application glxgears located in the mesa-utils package (on Ubuntu). You run the command vglrun glxgears, and if you see three red, green a blue cogs, everything works correctly. Test with GLXgears