User Tools

Site Tools


biac:cluster:performancetests

Preliminary Node Tests

Comparison of process duration

For tasks using concurrent processes, average times are reported. Individual process times are included in the Notes section for the task.

Task biacnode1 [860] biacnode2 [1950]
FSL Batch Job 1 2h 36m 2h 36m
FSL Batch Job 2 46m 48m
FSL Batch Job 3 1h 20m 1h 23m
FSL Batch Job 4 36m 38m
Freesurfer Job 1 14h 52m (892m) 15h 40m (940m)
Freesurfer Job 2 19h 45m (1185m) 22h 4m (1324m)
Matlab Job 1 39m 42m
Matlab Job 2 48m 54m
Dummy Task 1 x-d x-h x-m x-d x-h x-m

Notes

Dummy Task 1 - Tester

A brief description of this task including information about software and datasets (size, runs …) used.

FSL Batch Job 1 - Chris

a dataset from MIRECC with 10 runs (180 timepoints each), 23 copes … job includes slice-time correction, motion correction, coregistration and stats within FSL, using a bash script that i (chris) wrote (calls FEAT and uses SED for some replacements).

  • run1 finished in 17minutes(node1) | 16minutes(node2), the entire job took 2hr36m (~15m6s per run)
  • needed to replace the fsl distribution of gnuplot with the standard version supplied with linux
  • data read/wrote from/to the linux cluster, it was never transfered across the network to a storage server
FSL Batch Job 2 - Chris

the completed dataset from “batch job 1”, was used for a second level FSL analysis … 10 runs were combined for a within subject average (for all 23 copes).

  • also using a batch script written by me (chris)
FSL Batch Job 3 - Chris

a dataset from MIRECC with 5 runs (180 timepoints each), 23 copes … job includes slice-time correction, motion correction, coregistration and stats within FSL, using a bash script that i (chris) wrote (calls FEAT and uses SED for some replacements).

  • node1 had 2 instances of the same job running concurrently, each job finished in 1h20min (~16m each)
  • node2 had 4 instances of the same job running concurrently, each job finished in 1h23min (~16.6m each)
  • data read/wrote from/to the linux cluster, it was never transfered across the network to a storage server
FSL Batch Job 4 - Chris

the completed dataset from “batch job 3”, was used for a second level FSL analysis … 5 runs were combined for a within subject average (for all 23 copes).

  • also using a batch script written by me (chris)
  • node1 had 2 instances of the same job running, all finished in ~36m
  • node2 had 4 instances of the same job running, all finished in ~38m
Freesurfer Job 1 - Josh

Running a modified version of Jim's “runfreesurfer” on a 256x256x166 (1mm^3) T1 dataset. BIAC Scan ID 31816, series600. This script will run “recon-all” with all options “autorecon1”, “autorecon2” and “autorecon3”.

  • Had to edit “runfreesurfer” to add the subject directory to the input arguments of “recon-all” (using the -sd flag).
  • All data was processed locally on the node in /home/bizzell/freesurfer
  • Time for biacnode1: 892m, 43.227s (14h,52m,43s)
  • Time for biacnode2: 940m, 25.618s (15h,40m,26s)
Freesurfer Job 2 - Josh

Running the modified version of “runfreesurfer” described above on two datasets:

  • Dataset #1: 256x256x166 (1mm^3) T1, BIAC scan ID 31812, series600
  • Dataset #2: 256x256x166 (1mm^3) T1, BIAC scan ID 31885, series600

For biacnode1, I am running Freesurfer on both of these datasets in parallel. For biacnode2, I made one copy of each dataset, and am running Freesurfer on all 4 datasets (31812, 31885, 31812 copy, and 31885 copy) in parallel.

Time for biacnode1:

  • One dataset took 1096m, 17.684s (18h,16m,18s)
  • Other dataset took 1274m, 11.860s (21h,14m,12s)
  • Average time 1185m (19h,45m)

Time for biacnode2 (I incorrectly used the command “time” for this one, so I had to get the processing time from the Freesurfer log files):

  • Dataset1 31812: Start 4/6/07 12:08:28, End 4/7/07 12:05:46; 23h, 57m; 1437m
  • Dataset2 31812 copy: Start 4/6/07 12:08:28, End 4/7/07 11:48:57; 23h, 40m; 1420m
  • Dataset3 31885: Start 4/6/07 12:08:28, End 4/7/07 8:29:03; 20h, 21m; 1221m
  • Dataset4 31885 copy: Start 4/6/07 12:08:28, End 4/7/07 8:25:28; 20h, 17m; 1217m
  • Average 31812: ~1429m; 23h, 49m
  • Average 31885: 1219m; 20h, 19m
  • Average all: ~1324m; 22h, 4m
Matlab Job 1 - Trong-Kha

SENSE spiral recon on exam 32904, run #7 (spiral-in, 64×64, 30 slices, 168 time points)

  • 1 process on biacnode1: 39m
  • 1 process on biacnode2: 42m
Matlab Job 2 - Trong-Kha

same as Matlab Job 1

  • 2 concurrent processes on biacnode1: 48m | 48m | average: 48m
  • 4 concurrent processes on biacnode2: 53m | 55m | 54m | 54m | average: 54m

Node Specs

biacnode1
  • processor : Intel(R) Xeon(R) CPU 3070 @ 2.66GHz
  • cache size : 4096 KB
  • cpu cores : 2
  • memory : 3988948 kB
biacnode2
* processor 1    : Intel(R) Xeon(R) CPU 5150  @ 2.66GHz 
* cache size     : 4096 KB 
* cpu cores      : 2 
  • processor 2 : Intel(R) Xeon(R) CPU 5150 @ 2.66GHz
  • cache size : 4096 KB
  • cpu cores : 2
  • memory : 8131428 kB
biac/cluster/performancetests.txt · Last modified: 2014/08/04 16:03 (external edit)