Future directions

Measuring FITS I/O

Goal

Measure the FITS I/O speed for CTA format event lists and compare to ROOT and HDF5.

For FITS look at the speed of:

• CFITSIO
• CFITSIO Python wrapper
• astropy.io.fits
• astropy.io.table
• GammaLib (e.g. write with ctobssim and read / write with ctselect)

Methods

Define test datasets of different sizes (1 MB to 10 GB with ~ 10 values and log spacing).

Do operations so that I/O speed dominates execution time:

• Event selection using box cuts
• Histogram or simply sum data

These days memory is pretty large (10 GB to 100 GB are typical), maybe we should also measure memory to CPU I/O bound operations?

Some references

• http://www.hdfgroup.org/HDF5/RD100-2002/HDF5_Performance.pdf
• http://heasarc.gsfc.nasa.gov/fitsio/c/c_user/node119.html
• http://adsabs.harvard.edu/abs/2009CoPhC.180.2499A
• http://adsabs.harvard.edu/abs/2011JPhCS.331c2010A