Skip to content

Latest commit

 

History

History
211 lines (151 loc) · 7.3 KB

server.md

File metadata and controls

211 lines (151 loc) · 7.3 KB
Raw
Setup

Setup TDW on a Linux server

You can run the TDW build inside and outside a Docker container. There are advantages and disadvantages to either approach:

If you run TDW inside a Docker container, it can't use a GPU. This means that rendering will be slow and not as photorealistic. The Docker container is mainly suitable only for projects that don't require rendering or don't have sudo access. Even then, TDW in a Docker container is relatively slow because the physics engine normally utilizes the GPU.

If you run TDW outside a Docker container, it can use a GPU and run much faster. However, you need sudo access to configure the server.

This document describes how to run TDW outside a Docker container. To learn how to run TDW inside a Docker container, read this.

Requirements

  • Python 3.8+ (optionally, conda)
  • A Linux server. Typically, this is Ubuntu 20 or newer
  • An NVIDIA GPU
  • Up-to-date NVIDIA drivers Check online for which package to install.
  • Xorg and an active X server sudo apt install -y gcc make pkg-config xorg
  • The server must have 8 or less GPUs. If there are more, you can temporarily disable them by editing xorg.conf. This is a Unity bug that they have told us they won't fix.
  • Ability to use sudo

Why you need to use sudo to install TDW

Installing TDW on a server requires some admin privileges.

TDW runs on Unity3D, a video game engine. Unity3D requires a valid display buffer even if it is not rendering. Unity3D does offer a "headless" mode but because TDW often renders, it can't use headless mode. Therefore, TDW needs an active X server, which requires sudo to start.

Unity's rendering looks more photorealistic if the computer has a GPU. Without a GPU, Unity cannot render as realistically and rendering will be slower. Unity also offloads many physics calculations to the GPU; without a GPU, Unity's physics will be slower. The server must therefore physically have a GPU, and the GPU must have up-to-date drivers, which requires sudo to install.

The xorg.conf file will tell X displays to use a GPU. Without a valid xorg.conf file, TDW can't run with a GPU. Generating an xorg.conf file requires sudo.

You can get around this by running TDW in a Docker container, but there are caveats to doing so. See the beginning of this document, or read this.

Install Xorg

First, you need to create an xorg.conf file to tell Xorg that the computer has displays and that they can be run on GPUs.

  1. Run nvidia-xconfig --query-gpu-info. This will list all the GPUs and their bus ids
  2. Run nvidia-xconfig --no-xinerama --probe-all-gpus --use-display-device=none. This will generate xorg.conf (/etc/X11/xorg.conf) file
  3. Make n copies of this file for n GPUs. You can name each file as xorg-1.conf, xorg-2.conf ... xorg-n.conf
  4. Edit each file:
    1. Remove ServerLayout and Screen section
    2. Edit Device section to include the BusID and BoardName field of the corresponding GPU. You can get GPU list by running nvidia-xconfig --query-gpu-info

For example if nvidia-xconfig --query-gpu-info outputs two GPUs:

Number of GPUs: 2

GPU #0:
  Name      : Tesla V100-PCIE-16GB
  UUID      : GPU-bbf6c915-de29-6e08-90e6-0da7981a590b
  PCI BusID : PCI:0:7:0

  Number of Display Devices: 0


GPU #1:
  Name      : Tesla V100-PCIE-16GB
  UUID      : GPU-2a69c672-c895-5671-00ba-14ac43a9ec39
  PCI BusID : PCI:0:8:0

  Number of Display Devices: 0

Then create two xorg.conf files and edit the device section for first file:

Section "Device"
    Identifier     "Device0"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
EndSection

...to...

Section "Device"
    Identifier     "Device0"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
    BoardName      "Tesla V100-PCIE-16GB"
    BusID          "PCI:0:7:0"
EndSection

And

Section "Device"
    Identifier     "Device1"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
EndSection

...to...

Section "Device"
    Identifier     "Device1"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
    BoardName      "Tesla V100-PCIE-16GB"
    BusID          "PCI:0:8:0"
EndSection
  1. Run x-server. For each xorg configuration file run sudo nohup Xorg :<Display server name> -config <configuration file name> & :
sudo nohup Xorg :1 -config /etc/X11/xorg-1.conf & 

sudo nohup Xorg :2 -config /etc/X11/xorg-2.conf &
  1. When successfully done, running nvidia-smi should show the x-server proccess with corresponding GPU version

Install TDW

  1. ssh into the server
  2. pip install tdw
  3. Download the latest build of TDW and extract the zip file.
  4. Create this Python script and run it:
from tdw.controller import Controller

port = 1071
c = Controller(launch_build=False, port=port)
print("Hello world!")
c.communicate({"$type": "terminate"})

In another shell:

  1. ssh into the server
  2. cd to wherever you downloaded the build.
  3. DISPLAY=:0 ./TDW.x86_64 -port 1071 Replace :0 with a valid display value.

Result: The terminal window will print messages about downloading a build. Then, it will launch a windowed application, print Hello world!, kill the windowed application process, and exit.

Test image capture by repeating the same steps but with this controller:

from tdw.controller import Controller
from tdw.tdw_utils import TDWUtils
from tdw.add_ons.third_person_camera import ThirdPersonCamera
from tdw.add_ons.image_capture import ImageCapture
from tdw.backend.paths import EXAMPLE_CONTROLLER_OUTPUT_PATH

camera = ThirdPersonCamera(position={"x": 2, "y": 1.6, "z": -0.6},
                           avatar_id="a",
                           look_at={"x": 0, "y": 0, "z": 0})
path = EXAMPLE_CONTROLLER_OUTPUT_PATH.joinpath("image_capture")
print(f"Images will be saved to: {path}")
capture = ImageCapture(avatar_ids=["a"], path=path)

c = Controller(launch_build=False)
c.add_ons.extend([camera, capture])
object_id = c.get_unique_id()
commands = [TDWUtils.create_empty_room(12, 12)]
commands.extend(c.get_add_physics_object(model_name="iron_box",
                                         position={"x": 0, "y": 0, "z": 0},
                                         object_id=object_id))
c.communicate(commands)
c.communicate({"$type": "terminate"})

Result:

Always set launch_build=False

On Linux servers, always set launch_build=False in your controllers. For more information, read this.

Docker

The requirements for Docker are the same as those for non-Docker setups. The host computer have a GPU + drivers, xorg.conf file, and must be running Xorg. In the future, we hope to make it possible to run Xorg from within a Docker file with its own internal xorg.conf file.

To build our Docker container, first clone this repo and then:

cd tdw/Docker
./install.sh
./build.sh

To pull from Dockerhub instead:

cd tdw/Docker
./pull.sh

To run the TDW Docker container:

cd tdw/Docker 
./run.sh DISPLAY PORT ADDRESS

For example: ./run.sh :0 1071 localhost\

Troubleshooting

See this list of common problems. If you don't see your problem on the list, let us know by opening a GitHub Issue.