Mozart2 has been tested on computing platforms ranging from a single cpu on a desktop machine to more than one hundred cpus on a state of the art supercomputer. Whereas one typically does not have to be concerned with memory or software requirements on supercomputer systems that is not the case for desktop workstations. No matter what the computing environment the following are requisites for running mozart2:
We have not tested mozart2 with any Microsoft operating system. The Apple OS X operating system is suitable for running mozart2. Fortran 90 compilers are part of any supercompter system such as those from IBM, HP, SGI, ... . For desktop systems mozart2 has been compiled and is configured to use either the Portland Group, Lahey/Fujitsu, or Intel fortran 90 or 95 compilers. All compilers produce similar execution times with normal optimizations. To date we have briefly tested an alpha version of the xlf compilers from Absoft for the G5 MacIntosh. Mozart2 is not configured to compile for the G5 but if requested we will supply such support.
In addition to the above software requirements there are hardware recommendations. For desktop systems it is strongly recommended they have at least one gigabyte of main memory and at least 50 gigabytes of hard disk storage. Mozart2 configured at standard spatial resolution of 128 x 64 with 34 vertical levels will not run in 512MB main memory. For dual cpu systems whether they are Intel or AMD or Apple we recommend one gigabyte main memory per cpu. Mozart2 has successfully run using Openmp on two cpus with the above mentioned compilers on Intel, AMD, and Apple hardware. Scaling ranged from 85% to 95% efficiency meaning that execution time was reduced by anywhere from 1.7 to 1.9 as compared with a single cpu when using two cpus with Openmp.
Mozart2 is an "offline" model in that it requires dynamical and other data to be periodically input during a simulation. The standard dynamical datasets which drive mozart2 require about 25 gigabytes of disk storage for a simulation year. As supplied there is a standard output specification that results in about 25 gigabytes of output per simulation year. This is for the standard horizontal resolution of 128 x 64 and 34 vertical levels. Output is for variables that are output on a once a day and monthly basis. Mozart2 has a preprocessor that among other capabilties allows the user to modify the "standard" output. Thus output could either be considerably more than the "standard" 25 gigabyte per simulation year measure. Again for personal systems we recommend at least 50 gigabytes disk storage with 100 to 200 gigabytes preferrable.
Processor technology is constantly changing for the better, thus the following rough performance guide should be taken as only that; a rough guide.
| Hardware | Cpu count | Wall clock days/simulation year |
|---|---|---|
| Intel Pentium IV | 1 | 12 |
| AMD Opteron | 1 | 10 |
| Apple G5 | 1 | 9 |
The above figures are only as rough guide and do not represent an endorsement of any specific vendor. Our testing is limited by availability to actual systems.
By comparison 64 IBM Power4 cpus will finish a year simulation of mozart2 in about 7.5 wall clock hours.