FRIBParallelanalysis  1.0
FRIB Analysis Framework

This documentation describes a framework supplied at the FRIB to help you write code to analyze data from experments performed here. The framework supports parallel exectuion using the worker/farmer pattern. This allows you to write scalable, parallel analysis code whle only needing to focus on the segments of code that are relvevant to your problem.

Furthermore a SpecTcl jacket allows you to easily port SpecTcl event processors into this framework to trivially gain scalable speedup.

Analysis pipeline

Many experiments go through the same set of initial analysis steps:

Clearly any of these single steps or sets of steps may be iterated over until satisfactory/correct results are achieved.

The FRIB analysis framework provides frameworkds for each of the first three steps.

Processing framework.

The first two steps in the pipeline can be extraoridinarily processing intensive, depending on the actual computations required. Therefore a parallel processing framework is provided implemented on top of the Message Passing Interface (MPI). This standard for parallel computing is supported both on single systems and by most, if not all, job management systems in compute clusters. Thus scaling of your computation is limited only by event re-ordering, I/O bandwidth or communications overheads.

The parallel framework is organized as:

There is an implicit assumption in this structure that each event can be processed indpendnetly of all other events. For most experiments this is true. Some categories of experiments do have a processing stage for which this is not true; decay experiments which must match up implantation events with subsequent decay events.

These experiments can still take advantage of parallelism in two ways:

Where to go for more information: