ContactDownloadsThesisAbout

Main Menu
Home
Ph.D. Pursuit
Materials
EM Topics
F A Q
All Else

 Subscribe in a reader

Enter your email address:

Delivered by FeedBurner


Recommendations

Life as a Physicist

Ars Technica

Macenstein

strawberry hedgehog vegan bath products


Advertisements
Adding a Mac Pro to the Data Processing Arsenal
Wednesday, 17 October 2007
mac pro
The Mac Pro sits on the desk near a motivational gift given to me by a grad student colleague.

Things have recently changed at my desk with the addition of a new Mac Pro computer. After a few years in graduate school I made the switch to an Apple laptop and have been happy with it. The Pro is intended to provide useful computing power for chugging through the large volumes of data that have been collected by the end of an experiment. It's been processing data for about a month and during that time I have made some observations and come up with some suggestions for other people who are relatively new to Macs.

The Devil is in the Details

I spent a few weeks trying to determine whether this machine would actually prove useful. Of course it is going to work, but will it contribute in such a way as to shorten my completion time or improve the quality of the analysis? I concluded that it will make life much easier compared to the other options (all which can still be used in conjunction). In the end, however, I missed one important detail.

The primary (only?) analysis program I use, IDL, can not presently run in 64-bit mode on the Pro. The ability to do so is an assumption that I made and did not think to check. Fortunately, this issue will be solved when Apple releases their next operating system. Until then I am still plenty happy to be getting very good performance numbers (see the result of IDL's TIME_TEST3 at the end of this). The remaining problem is that the 32-bit version hits a memory limit for some of the data sets. Those troublesome sets are being reviewed on other machines until the update is released.

Speaking of memory issues, a recent IDL crash resulted in a fantastic shot of memory usage from Activity Monitor (see below).

16 exabyte memory usage
The 16 exabytes of claimed real memory usage was a sign that perhaps the IDL process was stuck and should be terminated.

“Fanboyism” and Getting Help

Some people like Macs, and some people love Macs. An example entry for the LOVE category may be found in a Wired.com commentary about the wonderful details within the Mail program.

The fanboys, or fanpeople, of Apple are the lovers. They have created a wealth of helpful websites dedicated to all things Mac. It is such a huge group that one can even find commentary about the fanpeople. One benefit of all this love is that there is a huge collection of helpful hints and problem solutions online. The official Apple Discussions and Mac OSX Hints sites have been particularly valuable (especially the show all files in Finder hint).

It was easy to find help when I came in to the lab one day to find that the hard drive was full and it was increasingly difficult to run programs. There was a small problem after I installed Ghostscript on my own and it resulted in an error message being written to the error log over and over. This two line error message was repeated in the log all the way until the log file reached a size over 75 GB and completely filled up the hard drive. For future reference, if you ever want to install a unix-type program on a mac just use Fink instead of trying to do it yourself.

I am not an Apple Fanperson, at least not yet. Putting a picture of your mac on your blog is not enough to qualify.

Downsides?

The Pro comes with 1 GB of RAM standard. That is not enough to make use of the multiple processors in any intensive operation so the purchase of additional memory is practically a requirement. Apple's memory sells for up to three times what can be found on other sites (e.g., Crucial is where I bought mine and they have a form that automatically finds the right ones for your system†). This is the same story for just about any computer: if you are surprised to find that your super-new, super-fast processor computer is slow cycling through the 400 seven-megapixel photos of your most recent vacation, then check whether you can add more RAM.

carrying a mac pro
After reaching the top of the hill I realized there was no power outlet.

Another downside is that it's too heavy to be portable.

Why get One?

The Pro is making it considerably faster and easier to perform detailed power spectra analysis and produce comprehensive graphics from the results. If this is important in your work, then a Pro is likely to be a sound investment. I'll post a movie or similar result soon, but that will not provide much of an indication as to how the computer made it easier to produce.

Result from TIME_TEST3
  1. 0.0219791 Empty For loop, 2000000 times
  2. 0.0162680 Call empty procedure (1 param) 100000 times
  3. 0.0165861 Add 200000 integer scalars and store
  4. 0.0189450 50000 scalar loops each of 5 ops, 2 =, 1 if)
  5. 0.0146329 Mult 512 by 512 byte by constant and store, 30 times
  6. 0.0911329 Shift 512 by 512 byte and store, 300 times
  7. 0.0490079 Add constant to 512x512 byte array, 100 times
  8. 0.0399351 Add two 512 by 512 byte arrays and store, 80 times
  9. 0.0538051 Mult 512 by 512 floating by constant, 30 times
  10. 0.0877450 Shift 512 x 512 array, 60 times
  11. 0.0946081 Add two 512 by 512 floating images, 40 times
  12. 0.0300059 Generate 1000000 random numbers
  13. 0.0210979 Invert a 192^2 random matrix
  14. 0.00609899 LU Decomposition of a 192^2 random matrix
  15. 0.0198891 Transpose 384^2 byte, FOR loops
  16. 0.0393019 Transpose 384^2 byte, row and column ops x 10
  17. 0.0659800 Transpose 384^2 byte, TRANSPOSE function x 100
  18. 0.0222468 Log of 100000 numbers, FOR loop
  19. 0.0193858 Log of 100000 numbers, vector ops 10 times
  20. 0.0219600 131072 point forward plus inverse FFT
  21. 0.0217879 Smooth 512 by 512 byte array, 5x5 boxcar, 10 times
  22. 0.0166481 Smooth 512 by 512 floating array, 5x5 boxcar, 5 times
  23. 0.117186 Write and read 512 by 512 byte array x 40
0.906234=Total Time, 0.030142634=Geometric mean, 23 tests.

As a reference, the total time output from the same test on a nearly equivalent four processor 2.2 GHz AMD Opteron machine is 1.09592 seconds.

Tags: , , , ,

† Nothing I mention in this item is a paid endorsement. If anyone compensated me for mentioning them, then I would post an announcement claiming it.
Comments
Add NewSearchRSS
Write comment
Name:
Email: (not published)
Website:
Title:
 
Security Image
Please input the anti-spam code that you can read in the image.

Copyright (C) 2007 Alain Georgette / Copyright (C) 2006 Frantisek Hliva. All rights reserved.
Last Updated ( Friday, 02 May 2008 )
 
< Prev   Next >
[ Back ]

 

© 2000-2008 David Pace
Design by David Pace