Optimizing software infrastructure to compute photo-zs in the LSST scale: preparing for DR1

Heloisa Mengisztki

In this project, we use the DP0.2 dataset and the DESC's RAIL pipelines to model the total duration of computing photometric redshifts (photo-z) at LIneA's computer cluster Apollo as a function of a list of variables such as (i) the kind of storage hardware used, (ii) the core algorithm adopted, (iii) the file formats for input and output data, and (iv) the output data compression (photo-z PDF representation). We perform stress tests in these different scenarios to verify if the total runtime scales linearly with the size of the dataset and to identify bottlenecks, sources of slowness, and opportunities for workflow optimization in preparation for LSST DR1

 

This poster will be displayed on Monday and Tuesday.

 

Career Stage: 
Undergrad Student