ParaView Tiled Display Mode
Using ParaView to render to a Tiled Display is becoming increasingly common as LCD display resolutions have slowed their growth and density. A Tiled Display in ParaView operates much like a Distributed Server, and therefore you should read about that first.
This document covers the differences between Distributed Server and Tiled Display mode.
For typical use, follow the same instructions that you use for a regular Distributed Server run, but add the following commandline arguments:
- -tdx=<width in tiles> - This sets the width of the tiled display, in number of tiles (monitors). Required
- -tdy=<height in tiles> - This sets the height of the tiled display, in number of tiles (monitors). Required
- -tmx=<horizontal mullion in pixels> - This specifies the gap to leave between tiles in the Horizontal Direction, in pixels. This is optional and typically not used on projector-based systems.
- -tmy=<vertical mullion in pixels> - This specifies the gap to leave between tiles in the Vertical Direction, in pixels. This is optional, and typically not used on Projector based systems.
Note that Process 0 will manage the Upper-left tile, and Process 1 will manage the tile to its right in a layout similar to the diagram shown below (for a 4x3 arrangement).
If your nodes are not numbered in such a fashion, you will need to find a way to specify the way that the processes are mapped to nodes. One such way is with the MPI host file. The processes are started across the nodes in the order they are specified in the host file (first host gets the first process, etc). Different versions of MPI may support other methods.
To compute the mullion width, first measure the size of the monitor and find out the resolution you'll be operating at. For a 16" wide x 12" height monitor running at 1600x1200, that's 100dpi. If the bezels between two monitors create a 1" gap, then your mullion is 100 pixels.
The options presented for Tiled Display mode mimic the options shown for Distributed Server mode, with one small addition.
- Client Collect - When this option is enabled, when the visible geometry is within the amount of memory specified by the slider, the geometry will be sent to the client to be rendered locally. This slows down the rendering, due to the extra bandwidth required to collect the geometry and transmit it to the client, but can make manipulating some interface elements (probes, cutting planes, etc.) easier as they must be manipulated on the client screen.
- In Tiled Display mode, each node needs to have X configured and running full-screen on the visible monitor. Access controls need to be set to allow any user to access the X session. No window manager is necessary.
- Tiled Display mode uses IceT for compositing. In IceT, every node renders the portion of its data that is visible on every display, and each node is responsible for collecting the necessary data from the other nodes and compositing it. The result is a very efficient and fast compositing and rendering algorithm that consumes a massive amount of bandwidth between the nodes.