#xpcs-hadoop
The instruction for how to compile the code with "Maven" are in the INSTALL file. This document describes major changes in each version of the code. The feature list here is not comprehensive, please consult the confluence page at https://confluence.aps.anl.gov/display/XPCS/XPCS+Analysis for more details.
Version 0.4.2
- Added the pipeline launch scripts along with the required files (jars, scripts etc). This provides a complete launch package for XPCS analysis.The pipeline version was taken from the Tag 0.1-20140311 (See https://subversion.xray.aps.anl.gov/pipeline/java/tags/0.1-20140311/).
Version 0.4.1
The version 0.4.1 is the most complete version that is in production since late 2013 at Sector 8. Here is a brief list of features in this version:
-
The complete multitau analysis of data in all three formats :Sparse, Non- Sparse, and Kinetics.
-
Process the IMM file stored directly in the Hadoop file system (HDFS).
-
Read the job configuration parameters from the HDF5 file (See the confluence page for more information on how the hdf5 parameters are interpreted along with their type: https://confluence.aps.anl.gov/display/XPCS/Job+Parameters)
-
The final results produced by the MapReduce are converted to a format that can be written as a HDF5 dataset. (See the confluene page https://confluence.aps.anl.gov/display/XPCS/HDF5+Output+Format for more details on the output format)
-
This version provides a partial fix for the concurrency issues involving processing of input parameters from HDF5 file. Our code reads the input parameters from the fixed location within the HDF5 file. As of this version it is the dataset named '/xpcs'. Before this version the '/xpcs' would be read twice: first, for configuring the job parameters (input, qmap etc), second, for determining the location of the output dataset in the same HDF5 file. The problem arises when an external program (e.g. Suresh's Matlab) modifies the value under '/xpcs' in between these two reads. We fixed it by reading all the information needed including the location of the output folder. However, this is a problem that require a bit re- engineering in particular when it will come to submitting multiple jobs using the pipeline.