dragnet:psrfits2fil

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Next revision		 Previous revision
dragnet:psrfits2fil [2016-08-03 20:08] – created Sotiris Sanidasdragnet:psrfits2fil [2017-03-08 15:27] (current) – external edit 127.0.0.1
Line 1: Line 1:
 ==== Benchmarks for filterbank creation with psrfits2fil ==== ==== Benchmarks for filterbank creation with psrfits2fil ====
  
-A series of tests were ran on dragnet (drg01), directly (now through slurm), on fits files from a random LOTAAS observation.+A series of tests were ran on dragnet (drg01), directly (now through slurm), on fits files from a random LOTAAS observation. Number of cores,means number of different fits files processed simultaneously. The total time is an extrapolation of this benchmark.\\
  
-=== Data read/written on the same disk ===+Only one test was ran for each occasion. Repeating them showed differences in execution time, but I doubt that the results change qualitatively. 
 + 
 +=== Input/Output on the same disk ===
 1-core:354sec\\ 1-core:354sec\\
 Total time for 16 files:5664sec\\ Total time for 16 files:5664sec\\
 +
 +3-cores:630sec\\
 +Total time for 16 files:3360sec\\
  
 4-cores:720sec\\ 4-cores:720sec\\
 Total time for 16 files:2880sec\\ Total time for 16 files:2880sec\\
 +
 +5-cores:1106sec\\
 +Total time for 16 files:3539sec\\
  
 8-cores:1735sec\\ 8-cores:1735sec\\
Line 16: Line 24:
 Total time for 16 files:3778sec\\ Total time for 16 files:3778sec\\
  
-When writing the filterbank in the same disk with the fits file, the best performance is achieved by having ~4 psrfits2fil instances running. +When writing the filterbank in the same disk with the fits file, the best performance is achieved by having 4 psrfits2fil instances running. 
  
 === Input/output on different disks === === Input/output on different disks ===
Line 28: Line 36:
 Total time for 16 files:1802sec\\ Total time for 16 files:1802sec\\
  
-16-cores:1686sec\\ +16-cores:1513sec\\ 
-Total time for 16 files:1686sec\\+Total time for 16 files:1513sec\\ 
 + 
 +24-cores (hyperthreaded):2252sec scales perfectly!\\ 
 +32-cores (hyperthreaded):3346sec performance loss already evident;~340 sec slower than expected 
 + 
 +Using all the cores in the dragnet nodes gives the best performance. Moreover, multithreading behaves exactly as having the extra cores.
  
-Using all the cores in the dragnet nodes gives the best performance.+Moving a fil file from /data2 to /data1 takes 1 minute. 
  • Last modified: 2016-08-03 20:08
  • by Sotiris Sanidas