Results for system throughput measurements for ccdcor and decon are reported below.
The results are listed by machine type (SGI first, Compaq Alphas next, then PCs). The performance measure is the number of Megapixels (1024^2 pixels) divided by the total time taken. ccdcor and the libraries it uses do not do parallel processing so it will only take advantage of one processor on multiprocessor machines. The data was written to and from a local disk; the runs were done in conjunction with runs for the set B decon results.
| Model | Processor | Clock speed (Mhz) | Main Memory (Mb) | Secondary Cache (Mb) | OS | Priism | Data Dimensions | Throughput (Megapixel/s) |
| Indigo2 | R4400 | 250 | 128 | 2.0 | 6.5 + 3131, 3139, 3219 | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.4167 |
| O2 | R5k | 180 | 64 | 0.5 | 6.5.1m | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.3563 |
| O2 | R5k | 180 | 192 | 0.5 | 6.5.1m | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.3834 |
| O2 | R10k | 150 | 256 | 1.0 | 6.5.3m | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.4633 |
| O2 | RM 5200 | 300 | 128 | 1.0 | 6.5.5m | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.5197 |
| O2 | RM 5200 | 300 | 256 | 1.0 | 6.5.6m + 3048, 3065, 3077, 3131, 3139, 3159, 3172 | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.5489 |
| O2 | R12k | 300 | 256 | 1.0 | 6.5.6m | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.6974 |
| Octane | R10k | 225 | 128 | 1.0 | 6.5 + 3131, 3139, 3219 | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.9592 |
| Octane | 2 R10k | 195 | 128 | 1.0 | 6.5.1m | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.8734 |
| Octane | R12k | 270 | 128 | 2.0 | 6.5.4m + 3659 | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 1.1584 |
| Octane | R12k | 270 | 384 | 2.0 | 6.5.4m + 3659 | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 1.6272 |
| Power Challenge | 6 R10k | 194 | 512 | 1.0 | 6.5.1m + 3131, 3139, 3140 | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.7119 |
| Power Indigo2 | R8k | 75 | 256 | 2.0 | 6.5.1m | 3.3 6/99 | 700 x 650 x 21 x 3 x 1 (A) | 0.2485 |
| DS10 | EV67 | 617 | 512 | ? | RedHat 6.2 | 3.3 Dev 2/01 | 700 x 650 x 21 x 3 x 1 (A) | 2.4586 |
| DS10 | EV6 | 463 | 384 | 4.0 | 4.0F | 3.3 10/99 | 700 x 650 x 21 x 3 x 1 (A) | 1.3018 |
| DS10L | EV67 | 600 | 256 | ? | RedHat 6.2 | 6.3 2/01 | 700 x 650 x 21 x 3 x 1 (A) | 2.1669 |
| XP1000 | EV6 | 500 | 640 | 4.0 | 4.0E | 3.3 10/99 | 700 x 650 x 21 x 3 x 1 (A) | 1.3669 |
| 440BX, 100 MHz | P3 | 500 | 256 | 0.5 | RedHat 6.1 | 3.3 Dev. 7/00 | 700 x 650 x 21 x 3 x 1 (A) | 0.5110 |
| 440BX, 100 MHz | P3 (CuMine) | 800 | 256 | 0.5 | RedHat 6.1 | 3.3 Dev. 7/00 | 700 x 650 x 21 x 3 x 1 (A) | 0.6234 |
| AMD 760 | Athlon C | 1200 | 512 | 0.25 | Debian 2.1 | 3.3 3/01 | 700 x 650 x 21 x 3 x 1 (A) | 1.7182 |
| 860 Chipset | P4 Xeon | 2000 | 1024 | 0.5 | RedHat 7.2 | 3.3 11/01 | 700 x 650 x 21 x 3 x 1 (A) | 4.8461 |
decon results | top of the page
The results are listed by machine type (SGI first, Compaq Alphas next, then PCs). The performance measure is the number of Megapixels (1024^2 pixels) divided by the total time taken. decon and the libraries it uses do not do parallel processing so it will only take advantage of one processor on multiprocessor machines. I/O was done to and from a local disk. The runs for set B and set C, with the exception of the Pentium III decon result for the 8/99 development version, were done in conjunction with set A ccdcor runs. The runs for the other sets were done alone.
The amount of physical memory had a definite impact on these results. With test set B, an O2 with 64 Mb had 1.9 million page faults per run causing the throughput to drop by more than a factor of six; the Octanes with 128 Mb had 3000 page faults per run which probably translated to a ten percent drop in throughput from what would have been possible if they had had more memory. When memory is limited, the -nomem option can be a significant factor; running the same job with the -nomem option (the results for test set C in the table) boosted the performance on the O2 with 64 Mb by a factor of three.
Another important factor is the use of the FFTW library on the Alpha and Pentium-based machines. The results from Priism version 3.3 Dev. 7/00 are from executables using this library and show significantly improved throughputs compared to the older versions which used the MR TFFT routines. On the SGIs, the results are mixed but appear to be generally close to the results with the complib based version.
| Model | Processor | Clock speed (Mhz) | Main Memory (Mb) | Secondary Cache (Mb) | OS | Priism | Data Dimensions | Throughput (Megapixel/s) |
| Indigo2 | R4400 | 250 | 128 | 2.0 | 6.5 + 3131, 3139, 3219 | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.01203 |
| O2 | R5k | 180 | 64 | 0.5 | 6.5.1m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.00228 |
| O2 | R5k | 180 | 64 | 0.5 | 6.5.1m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (C) | 0.00706 |
| O2 | R5k | 180 | 192 | 0.5 | 6.5.1m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.01383 |
| O2 | R10k | 150 | 256 | 1.0 | 6.5.3m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.01704 |
| O2 | RM 5200 | 300 | 128 | 1.0 | 6.5.5m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.01626 |
| O2 | RM 5200 | 300 | 256 | 1.0 | 6.5.6m + 3048, 3065, 3077, 3131 3139, 3159, 3172 | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.01996 |
| O2 | RM 5200 | 300 | 256 | 1.0 | 6.5.6m + 3048, 3065, 3077, 3131 3139, 3159, 3172 | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.02199 |
| O2 | R12k | 300 | 256 | 1.0 | 6.5.6m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.0306 |
| O2 | R12k | 300 | 256 | 1.0 | 6.5.6m | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.02712 |
| O2 | R12k | 300 | 256 | 1.0 | 6.5.6m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (C) | 0.0246 |
| Octane | R10k | 225 | 128 | 1.0 | 6.5 + 3131, 3139, 3219 | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.03333 |
| Octane | 2 R10k | 195 | 128 | 1.0 | 6.5.1m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.03033 |
| Octane | R12k | 270 | 128 | 2.0 | 6.5.4m + 3659 | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.04298 |
| Octane | R12k | 270 | 384 | 2.0 | 6.5.4m + 3659 | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.04856 |
| Octane | R12k | 270 | 384 | 2.0 | 6.5.4m + 3659 | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.04777 |
| Octane | R12k | 270 | 384 | 2.0 | 6.5.4m + 3659 | 3.3 3/00 | 1020 x 1020 x 31 x 1 x 1 (D) | 0.01660 |
| Octane | R12k | 270 | 384 | 2.0 | 6.5.4m + 3659 | 3.3 Dev. 7/00 | 1020 x 1020 x 31 x 1 x 1 (D) | 0.01422 |
| Octane | R12k | 270 | 384 | 2.0 | 6.5.4m + 3659 | 3.3 3/00 | 512 x 512 x 110 x 1 x 1 (E) | 0.03229 |
| Octane | R12k | 270 | 384 | 2.0 | 6.5.4m + 3659 | 3.3 Dev. 7/00 | 512 x 512 x 110 x 1 x 1 (E) | 0.03861 |
| Power Challenge | 6 R10k | 194 | 512 | 1.0 | 6.5.1m + 3131, 3139, 3140 | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.02940 |
| Power Challenge | 6 R10k | 194 | 512 | 1.0 | 6.5.1m + 3131, 3139, 3140 | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.03027 |
| Power Indigo2 | R8k | 75 | 256 | 2.0 | 6.5.1m | 3.3 6/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.01034 |
| DS10 | EV6 | 463 | 384 | 4.0 | 4.0F | 3.3 10/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.0376 |
| DS10 | EV6 | 463 | 1024 | 4.0 | 5.0 | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.08053 |
| DS10 | EV67 | 617 | 512 | ? | RedHat 6.2 | 6.3 2/01 | 600 x 600 x 21 x 3 x 1 (B) | 0.09118 |
| DS10L | EV67 | 600 | 256 | ? | RedHat 6.2 | 6.3 2/01 | 600 x 600 x 21 x 3 x 1 (B) | 0.09036 |
| XP1000 | EV6 | 500 | 640 | 4.0 | 4.0E | 3.3 10/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.0370 |
| XP1000 | EV6 | 500 | 640 | 4.0 | 5.0 | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.08597 |
| XP1000 | EV6 | 500 | 640 | 4.0 | 5.0 | 3.3 3/00 | 1020 x 1020 x 31 x 1 x 1 (D) | 0.02528 |
| XP1000 | EV6 | 500 | 640 | 4.0 | 5.0 | 3.3 Dev 7/00 | 1020 x 1020 x 31 x 1 x 1 (D) | 0.02944 |
| XP1000 | EV6 | 500 | 640 | 4.0 | 5.0 | 3.3 3/00 | 512 x 512 x 110 x 1 x 1 (E) | 0.02560 |
| XP1000 | EV6 | 500 | 640 | 4.0 | 5.0 | 3.3 Dev. 7/00 | 512 x 512 x 110 x 1 x 1 (E) | 0.05985 |
| 440BX, 100 MHz | P3 | 500 | 256 | 0.5 | RedHat 6.0 | 3.3 Dev. 8/99 | 600 x 600 x 21 x 3 x 1 (B) | 0.01667 |
| 440BX, 100 MHz | P3 | 500 | 256 | 0.5 | RedHat 6.1 | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.03769 |
| 440BX, 100 MHz | P3 (CuMine) | 800 | 256 | 0.25 | RedHat 6.1 | 3.3 Dev. 7/00 | 600 x 600 x 21 x 3 x 1 (B) | 0.06116 |
| AMD 760 | Athlon C | 1200 | 512 | 0.25 | Debian 2.1 | 3.3 3/01 | 600 x 600 x 21 x 3 x 1 (B) | 0.09402 |
| 860 Chipset | P4 Xeon | 2000 | 1024 | 0.5 | RedHat 7.2 | 3.3 11/01 | 600 x 600 x 21 x 3 x 1 (B) | 0.12545 |
ccdcor results | top of the page
The input data set was 800x800x23 with 3 wavelengths and the output was 700x650x21 with 3 wavelengths. The arguments used (with the exception of the names of the input and output files) were:
| Argument | Value |
| -x | 50:749 |
| -y | 55:724 |
| -w | 540:605:460 |
| -area | 50:749:75:724 |
| -norder | 4:4:4:4:4 |
| -nout | 4:4:4:4:4 |
| -polyfit | 1:1:1:0:0 |
| -photon | 1:1:1:0:0 |
| -bleach | 1:1:1:0:0 |
| -zline | 1:1:1:0:0 |
| -itref | 0 |
decon results | ccdcor results | top of the page
The input data set was 700x650x21 with 3 wavelengths and the output was 600x600x21 with 3 wavelengths. The arguments used (with the exception of the names of the input and output file) were:
| Argument / Option | Value |
| OTF file | lens12.realctf |
| -x | 50:649 |
| -y | 25:624 |
| -w | 540:605:460 |
| -ncycl | 15 |
| -apodize | 0 |
| -nzpad | 32 |
| -method | DAA |
| -sub | 0:0:0:0:0 |
| -smooth | .15:.8 |
| -wiener | .9 |
| -prefilter | .3:.22 |
| -scale | 1 |
decon results | ccdcor results | top of the page
This is the same as test set B but uses the -nomem option.
decon results | ccdcor results | top of the page
The input data set was 1020x1020x31 with 1 wavelength (540 nm) and the output was 1020x1020x31 with 1 wavelength. The arguments used (with the exception of the names of the input and output file) were:
| Argument / Option | Value |
| OTF file | lens12.realctf |
| -ncycl | 15 |
| -apodize | 0 |
| -nzpad | 64 |
| -method | DAA |
| -sub | 0:0:0:0:0 |
| -smooth | .15:.8 |
| -wiener | .9 |
| -prefilter | .3:.22 |
| -scale | 1 |
decon results | ccdcor results | top of the page
The input data set was 512x512x110 with 1 wavelength (460 nm) and the output was 512x512x110 with 1 wavelength. The arguments used (with the exception of the names of the input and output file) were:
| Argument / Option | Value |
| OTF file | lens12.realctf |
| -ncycl | 15 |
| -apodize | 0 |
| -nzpad | 256 |
| -method | DAA |
| -sub | 0:0:0:0:0 |
| -smooth | .15:.8 |
| -wiener | .9 |
| -prefilter | .3:.22 |
| -scale | 1 |
decon results | ccdcor results | top of the page
modified May 20, 2003
Eric Branlund (eric@msg.ucsf.edu)