High Throughput Computing: A Solution for Scientific Analysis
HTC Computing Times
FORT is using High Throughput Computing (HTC) for numerous projects, and the figure illustrated below compares the relative computing times for each of these projects.
||Maximum Entropy (MaxEnt)
||2D/3D Hydrodynamic Numerical Modeling
||Species Trends Analysis
||Bioclimatic Variables and Continuous Surface Drought Indices
For all case studies, we have provided three estimated computing times:
- Single Machine describes the estimated amount of time the project would require running on one workstation with one core.
- FORT HTC describes the amount of time the project took to run using FORT’s HTC system (note that machine availability varies throughout the day, and is therefore not constant across studies or throughout the day).
- Perfect HTC represents the estimated time to complete a project if all pool resources were available and everything was equal (e.g., memory, CPUs, and software), with no failed jobs and no constraints other than the number of processing cores.
Specific terms for the above cases:
- Case study 2 encountered additional limitations because the software used for the analysis required user interaction. We were able to overcome this limitation but at a cost of increasing computing time.
- Case study 3 provides information on processing the bioclimatic data for the 2-km climate data and not for the entire scope of the project. We randomly selected 40 jobs and averaged the processing time (single machine measurement) because we were unable to successfully complete this project on a single machine. Case study 3 could not have been completed without HTC.
- Case study 4 encountered limitations due to an inadequate number of processing cores available on the relational database server. In this case, we could not run more than 50 concurrent jobs (100 concurrent database connections).