mengjiaoh / display_wall

DisplayWall 2

DisplayWall 2 is a light-weight framework for driving large tiled display wall.

Building from Source

Dependencies:

• libjpeg-turbo8-dev

Compiling and Installing Client Library

Client library is a stand alone library for communicating with the display wall. Users can compile, install the library and link it to any renderer.

• Create a build directory

    mkdir dw2/build
    cd dw2/build

• Open the CMake configuration dialog

    ccmake .. 

• Make sure to enable BUILD_CLIENT. And USE_TURBO_JEPG is recommended to enable in order to compress tile images. Then type 'c'onfigure and 'g'enerate. Then build it use

    make -jn install

Finding the client library install with CMake

Find the libray with CMake's find_package() function.

find_package(dw2_common REQUIRED)
find_package(dw2_client REQUIRED)
target_link_library(${target} 
                    dw2_common::dw2_common
                    dw2_client::dw2_client)

Compiling Display Wall Service

Display Service is used for driving X11 windows on display wall, letting renderer clients connect to and receiving tile images from clients.

Follow the instruction above and enable BUILD_SERVICE.

Compiling Test Renderers

Some examples of test renderer can be found in folder /test. Currently, there are 2 test renderers.

• Basic renderer: testFrameRender Compile with enable BUILD_TEST. dw2_testFrameRender is a simple renderer drawing continous colors to pixels.

• OSPRay renderer: testOSPRay Compile with enable BUILD_TEST_OSPRAY. dw2_testOSPRay is an example of using OSPRay. Currently, it renders spheres at random positions.

  • In order to run ospray example, compile and install dw2_common and dw2_client library first(following the instruction of compiling and installing the client library).
  • Clone ospray and clone ospray_wall module under ospray/module directory. Compile and install ospray library with enable ospray_module_wall.
  • Compile display wall code with enable BUILD_TEST_OSPRAY

Running:

Launching Displays

Usage

usage: ./dw2_service [args]*

Variable	Description
--num-displays/-nd <Nx> <Ny>	num displays (total) in x and y dir
--width/-w <numDisplays.x>	num displays in x direction
--height/-h <numDisplays.y>	num displays in y direction
--window-size/-ws <res_x> <res_y>	window size (in pixels)
--[no-]head-node / -[n]hn	use / do not use dedicated head node
--max-frames-in-flight/-fif <n>	allow up to n frames in flight
--bezel-width/-bw <Nx> <Ny>	assume a bezel width (between displays) of Nx and Ny pixels
--displays-per-node/-dpn <N> <display1> ...<displayN>]	This only makes sense if the mpi launch uses N ranks per display node. Specifies that each physical host in the mpi launch will have N physical displays attached to, with each display specified through three values: first, the X display is on (ie, :0, :1, etc), and the, pixel coords (x and y) of the lower-left pixel of that X display, with the i'th rank on any given node running the i'th such display.

Examples below are how to run on SCI powerwall, which has 9 * 4 monitors with resolution 2560 *1440.

service, no head node

    mpirun -ppn 1 -n 9 -hosts powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09 -genv DISPLAY=:0 ./dw2_service -w 9 -h 1 -nhn -fs

service, w/ head node

   mpirun -ppn 1 -n 10 -hosts powerwall00,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09 -genv DISPLAY=:0 ./dw2_service -w 9 -h 1 -hn -fs 

service, no head node, with individual ranks per display (meaning manual window placement and proper bezels applying):

• now have to use 4x9=36 ranks, each node has to appear 4 times:

• have to manulaly specify size of each display (glfw will report size of all four displays as one 'monitor', which is wrong if we want to drive them individually)

• have to specify position of the four displays on each node with --displays-per-node

• have to specify (global) --num-displays as 9x4, not 9x1

• CAN now specify bezel width, too...

Final command:

     mpirun -ppn 1 -n 36 -hosts powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09 -genv DISPLAY=:0 ./dw2_service -nhn --num-displays 9 4 --window-size 2560 1440 --displays-per-node 4 0 0 4320 0 0 2800 0 0 1440 0 0 0

service, w/ head node, with individual ranks per display (meaning manual window placement and proper bezels applying):

     mpirun -ppn 1 -n 37 -hosts powerwall00,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09 -genv DISPLAY=:0 ./dw2_service -nhn --num-displays 9 4 --window-size 2560 1440 --displays-per-node 4 0 0 4320 0 0 2800 0 0 1440 0 0 0

Note we do NOT use -ppn 4 (and instead explicitly specify each node four times) because ppn 4 would mean rank 0,1,2,3 would all get mapped to displaywall01, but those four ranks should drive the first four logical displays in the lower line of displays, NOT the four displays in the first column...

Running testFrameRender

test app, no head node

    mpirun -ppn 1 -n 10 -hosts powerwall00,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09  ./dw2_testFrameRenderer powerwall01 2903

test app, head node

    mpirun -ppn 1 -n 10 -hosts powerwall00,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09  ./dw2_testFrameRenderer powerwall00 2903

Running testOSPRay

test app, no head node

    mpirun -ppn 1 -n 10 -hosts powerwall00,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09  ./dw2_testOSPRay powerwall01 2903 --osp:mpi

test app, head node

    mpirun -ppn 1 -n 10 -hosts powerwall00,powerwall01,powerwall02,powerwall03,powerwall04,powerwall05,powerwall06,powerwall07,powerwall08,powerwall09  ./dw2_testOSPRay powerwall00 2903 --osp:mpi