IRAF Benchmarks
Robert Emery wrote on Mar 25, 2007
For standardized IRAF tasks, are there any benchmark statistics published for various current hardware platforms and configurations?
1) I plan to put IRAF on a new PC running Linux. I would like to get a feel beforehand of how much processing power I would need to run standardized tasks in an acceptable time interval. I currently have no experience with IRAF on any platform.
2) Can IRAF benefit from a multiple core PC? I could build the PC with a single single-core CPU, a single dual-core CPU or a dual dual-core CPU. Given a dedicated machine, I don't see how IRAF could take advantage of the extra cores. All the CPUs would be AMD 64-bit and cutting edge in terms of operating frequency.
3) Is there a Linux distro that is particularly recommended for IRAF? If so, which one and why?
Thanks.
1) I plan to put IRAF on a new PC running Linux. I would like to get a feel beforehand of how much processing power I would need to run standardized tasks in an acceptable time interval. I currently have no experience with IRAF on any platform.
2) Can IRAF benefit from a multiple core PC? I could build the PC with a single single-core CPU, a single dual-core CPU or a dual dual-core CPU. Given a dedicated machine, I don't see how IRAF could take advantage of the extra cores. All the CPUs would be AMD 64-bit and cutting edge in terms of operating frequency.
3) Is there a Linux distro that is particularly recommended for IRAF? If so, which one and why?
Thanks.
Mike Fitzpatrick wrote on Mar 25, 2007
Robert,
You need to quantify terms like 'standardized' and 'acceptable' to get a meaningful answer, but maybe the following will help.
We don't generally keep benchmarks due to the infinite number of configurations, but a "standard" benchmark script we've used before creates some dummy images, does basic CCD processing, combines and image and does a median filter of 2k x 2k images; all the standardized things I think you mean. On a dual 3-Ghz linux (RH9) server with 2GB RAM the execution times are roughly:
This test pretty well covers the CPU- and disk-intensive applications but of course times will vary based on things like your disk speed, size of images, number to process, etc.
Only in the sense the system in inherently multi-process, i.e. the CL, the task you're running, maybe a graphics kernel etc are all running for one "task". You can run parallel pipelines to use the CPU/core but an individual image won't be split across mutiple cores for processing. You can certainly write a pipeline that keeps a multi-cpu/core machine occupied, but not just by running CCDPROC from the prompt.
Note also that IRAF is not (yet) 64-bit compatible so you'll be limited to 2Gb memory and file sizes, and compilation under 64-bit systems can also be tricky. Existing 32-bit binaries will run but keep this in mind if you'll be developing something other than scripts.
The most popular is probably still RedHat Enterprise/Fedora but there is no specific recommendation. Hope this helps.
Cheers,
-Mike
You need to quantify terms like 'standardized' and 'acceptable' to get a meaningful answer, but maybe the following will help.
We don't generally keep benchmarks due to the infinite number of configurations, but a "standard" benchmark script we've used before creates some dummy images, does basic CCD processing, combines and image and does a median filter of 2k x 2k images; all the standardized things I think you mean. On a dual 3-Ghz linux (RH9) server with 2GB RAM the execution times are roughly:
Total execution time = 16.0 seconds
Total time Make 5 imgs Proc 3 imgs Combine 3 imgs Median 1 img
16.0 5.0 3.0 1.0 7.0
This test pretty well covers the CPU- and disk-intensive applications but of course times will vary based on things like your disk speed, size of images, number to process, etc.
2) Can IRAF benefit from a multiple core PC? I could build the PC with a single single-core CPU, a single dual-core CPU or a dual dual-core CPU. Given a dedicated machine, I don't see how IRAF could take advantage of the extra cores. All the CPUs would be AMD 64-bit and cutting edge in terms of operating frequency.
Only in the sense the system in inherently multi-process, i.e. the CL, the task you're running, maybe a graphics kernel etc are all running for one "task". You can run parallel pipelines to use the CPU/core but an individual image won't be split across mutiple cores for processing. You can certainly write a pipeline that keeps a multi-cpu/core machine occupied, but not just by running CCDPROC from the prompt.
Note also that IRAF is not (yet) 64-bit compatible so you'll be limited to 2Gb memory and file sizes, and compilation under 64-bit systems can also be tricky. Existing 32-bit binaries will run but keep this in mind if you'll be developing something other than scripts.
3) Is there a Linux distro that is particularly recommended for IRAF? If so, which one and why?
The most popular is probably still RedHat Enterprise/Fedora but there is no specific recommendation. Hope this helps.
Cheers,
-Mike
Chris Hunter wrote on Mar 25, 2007
fitz
Robert,
Note also that IRAF is not (yet) 64-bit compatible so you'll be limited to 2Gb memory and file sizes, and compilation under 64-bit systems can also be tricky. Existing 32-bit binaries will run but keep this in mind if you'll be developing something other than scripts.
Fairly certain 32bit OS is limited to 4GB memory. I think the maximum file size is 2GB. There was a limitation of 256MB memory for iraf from using "shared libraries". I think this has been removed ?
There are some variables such as MAXWORKSET that affect the memory available to iraf. I really don't know if they actually do anything now.
Regards,
--Chris
Robert Emery wrote on Mar 25, 2007
Mike,
I appreciate the response. The figures you gave indicate anything I build will be fast enough for typical IRAF tasks. Thanks.
I appreciate the response. The figures you gave indicate anything I build will be fast enough for typical IRAF tasks. Thanks.
Last post on Mar 25, 2007