NCSA Home
Contact Us | Intranet | Search

data link Story: Image Server for High-Resolution Movie Playback on Tiled Displays

News
datalink
0107
Globus
Current issue
Archives

Image Server for High-Resolution Movie Playback on Tiled Displays

The goal of this effort is to store high-resolution (e.g., 5100 x 2300 pixels) frames of the visualization of scientific data, such as those rendered from the data generated by scientists at the FLASH Center at the University of Chicago, and play them back as animations on tiled displays. The objective being to produce a system that only requires individual frames to be rendered once, at the greatest possible resolution, and then be capable of adapting the frames for playback a tiled display of any arbitrary resolution, potentially remotely located.

The intent is to use the striped/parallel gsiftp server to transfer the appropriately configured portion of the full-resolution image to each of the machines responsible for displaying the individual tiles. This means that for some tiled displays the original image will require some intelligent transformation to be applied to the data before transferring. The gsiftp server will act as the delivery mechanism for the distribution of the individual tiles to their needed location, capable of doing this in an intelligent manner based on the configuration of the tiled display and transparent to the user.

This tool will facilitate the more effective use of high-resolution renderings, requiring that the final movie only be rendered once regardless of display type. The tool will conserve the use of local disk, by allowing movies to be stored at a remote repository, instead of being stored locally on each tiled display. In the case of remote distribution of the tiled movies, by virtue of the parallel streaming of the gsiftp library, users will be able to effectively access data as if it was local.

The bandwidth requirements are: 1024x768x3=~2.36MB/tile x 15 tiles = ~35.39 MB/frame x 30 frames/sec = ~1.0b GB/sec aggregate bandwidth. In order to achieve this aggregate bandwidth, all of the backend nodes of the cluster will need to have high bandwidth external network.

The figure gives a general overview of the Tiled Movie Player architecture.

The player will make a third-party transfer request between a striped server running on a cluster used for storage and another running on a cluster used for visualization, which is connected to a tiled display. The front-end node on the storage cluster will forward the request on to its back-end nodes, which will each retrieve a portion of an image off of its shared (parallel) file system, perform any necessary transformations, and then send it directly to the appropriate back-end node on the visualization cluster, which will then display it on the tiled display once all of the tiles for a particular frame have been received. There could also be multiple data connections between each of the back-end nodes of the storage cluster and the back-end nodes of the visualization cluster.

Contact: Joseph Insley, insley@mcs.anl.gov, Futures Lab, Argonne National Laboratory

Contributors: Futures Lab, Argonne National Laboratory; Distributed Systems Lab, Argonne National Laboratory

URL: www.mcs.anl.gov/fl/ and www.mcs.anl.gov/dsl/