The Synthetic Aperture Radar Atlas (SARA) is a web-based distributed data application which allows users with access to the World-Wide Web and the Internet to view images of the Earth's surface taken by a synthetic aperture radar [13]. The data upon which these images are generated are collected by satellites as they pass over the Earth, resulting in long, rectangular regions, or tracks of data. Because these data are stored in raw uncompressed SAR format, it is partitioned and replicated across several high-capacity storage sites. Via a web page and a Java applet, users of the SARA system can request an image of an arbitrary sub-region of the track with certain features of the data highlighted.
The SARA application is essentially comprised of three logical phases, the first of which is the data retrieval phase. During data retrieval, raw SAR data corresponding to the requested region of a track is located and retrieved, usually from a high-capacity storage subsystem like the High-Performance Storage System (HPSS) filesystem. This is done with the help of a metadata server which maps data tracks to sets of servers which have the data. Data stored in these filesystems achieves access times ranging from seconds (if the files are in a disk cache) to perhaps even hours (if the files are not in disk cache and require operator intervention to mount a set of tapes). The final step of this phase requires moving the raw data to a processing node in preparation for processing the image.
In the data processing phase, raw data is converted into an image file. The data is filtered, reduced, and encoded, based on two user inputs: the features the user wishes to highlight, and the image format the user requests. The final logical phase of the application begins when the image file is prepared.
The image transfer phase involves moving the image file from the processing node to the user's machine. The current SARA application has only a Web-based interface, meaning that the image transfer is done via an HTTP connection, and the image is displayed in a web browser.
The job of scheduling an application such as SARA can be logically performed in a hierarchical and distributed fashion: scheduling decisions made at one phase can be used to make subsequent choices. No scheduling is required of the data server node, since a single data server needs only fetch data from the storage subsystem which it serves and transfer it back to the requester. However, during the data retrieval phase, processing nodes act as resource selectors and schedule planners in deciding which data servers should be employed to retrieve which data sets. These results can then be used by the Java client to make a similar decision about which processing node should be responsible for generating which parts of the requested image. In the next sections, we describe our initial implementation of and results from the resource selection and planning components of the AppLeS agent which operate at the processing node level. A description of further plans for the development of an ``end-to-end'' SARA AppLeS are described in Section 6.