Mira Computational Readiness Assessment
Assess your project's computational readiness for Mira
A review of the following computational readiness points in relation to scaling, porting, I/O, memory and vectorization may help you to determine if the INCITE Program is an appropriate resource for your research efforts.
- For current large allocations, (100M-300M) can the science runs scale to 32-racks of Intrepid (or 100,000 cores on other systems)?
- For current allocations up to 100M, can the science runs scale to 8 or 16 racks of Intrepid (or 32,000-64,000 cores on other systems)?
- Do you have a good plan for scaling, either through scaling the code or ensembles?
- Has the code been ported to a Blue Gene/P system?
- Has the code been ported to a Blue Gene/Q system? Most codes that run on Intrepid will run on Mira with no changes, but tuning yields better performance.
- Do you understand your I/O requirements?
- Do you have the ability to adjust your method for I/O if required?
- Do you have methods to exploit node-level parallelism and your memory footprint? The more this is understood, the better the code will be at exploiting many lessons learned for Mira.
- Do you use BLAS/FFT/etc., for a large percentage of computation?
- Do you have explicitly vectorized code on any platform (e.g., SSE, double hummer, etc.)?
- If you haven't pursued vectorization yet, is your code otherwise amenable to it?