| |
 |
| Paper: |
Fast Approximate Method for Computing Simulated Galaxy Fluxes on Top of Apache Spark |
| Volume: |
535, Astronomical Data Analysis Software and Systems XXXI |
| Page: |
379 |
| Authors: |
Tallada-Crespi, P.; Carretero, J.; Serrano, S.; Castander, F. J.; Cesar, E.; Eriksen, M.; Fosalba, P.; Gaztanaga, E.; Merino, G.; Neissner, C.; Tonello, N.; Torradeflot, F. |
| Abstract: |
The standard approach to estimate galaxy bandpass-averaged spectral flux density (from now on, galaxy fluxes) from cosmological simulations is a complex algorithm that requires solving several integrals. For very large simulations, the computational cost of producing those fluxes for every wavelength band (filter) of a photometric survey can become prohibitive. This work describes the approximate method that has been designed and developed to estimate the observed light fluxes, reducing considerably the computational time and with the precision imposed by scientific constraints. This method consists of refactoring and splitting the standard algorithm based on integrals, into three independent steps: the first step consists on the calculation of the flux from the galaxy spectral energy distribution (SED) (including the intrinsic dust extinction), the second adds the spectral emission lines contribution (also extincted), and the third corrects for the local light extinction caused by the Milky Way. This method has been calibrated to not introduce a difference in galaxy magnitude larger than ∆(m) < 0.01. The new algorithm has also been deployed on top of Apache Spark to be able to take advantage of distributed computing resources. Running on a 300-core Spark cluster at the Port d’Informació Científica (PIC) it can compute around 380 million galaxy fluxes per minute, with just 47 microseconds per flux on average, about 750 times faster compared to the integral method. |
|
|
|
|
|
