SwiftMontage

 

 

Overview

There are many scientific applications out there today. One such application is the Montage application. Montage is used to create science grade astronomical mosaics. Depending on the size of mosaic that is being generated the size of they data can be in the hundreds of giga-bytes. Using computing resources efficiently to create these mosaics is a necessity.

This is where Swift comes into the picture. Swift is a parallel scripting language with the ability to use multiple computing resources during the execution. These computing resources are known as sites. One mechanism that Swift implements for use in this multi-site execution model is Coasters which allows the execution to only incur the queue time penalty once. The queue time penalty is the amount of time a job must sit and wait in the queue before execution can begin on these computing resources since most resources use some sort of queue system to assure fairness for all users.

Here is a link to the SwiftMontage API. There is also a pdf version for download.

You can download the latest version of Montage v3.3 and Swift v0.92.1.

 

Google project hosting for the project is at http://code.google.com/p/swiftmontage. This is used for issue tracking and downloading the software package once the code is stabilized.

 

 

Multisite

multisite.gif

Above simple visual representation of how Swift works when the Coasters mechanism is used in the site execution model. For a more detailed description about how Coasters is executed, see wiki.cogkit.org/wiki/Coasters.



Since Swift uses a scripting language like syntax, workflows are very easily written. When Swift is applied to the Montage application the inherent parallelism of the problem is exposed. Below is a visualization of the general workflow for the Montage application.

 

 

Workflow

Montage.gif

This is visualization of the workflow for a 10 image data set using the Montage application represented as a Directed Acyclic Graph(DAG).

 

 

Mosaic

m42 cluster

This is a colored mosaic of the m42 cluster. It was generated from the h, j, and k bands of a 1 square degree of the sky centered at the m42 nebula. Each band consists of 56 images.