Boosting ANSYS Mechanical Simulation Capacity with Intel® SATA Solid-State Drives

Submit New Article

May 31, 2010 12:00 AM PDT


By Jeff Beisheim, ANSYS, Inc and Arvind Amin, Intel Corp.

Simulation models are getting larger - using more computer memory and requiring more computational time - as engineers include greater geometric detail and more-realistic treatment of physical phenomena to gain deeper insight into the products they are developing.

While ANSYS has optimized its software to run these larger models in a parallel environment, one of the main factors limiting the size of simulations that can be run on workstations is the amount of physical memory (RAM) available on the machine. Even though a single workstation with two Intel® Xeon 5600 processors now provides up to 12 computational cores, 24 threads and memory configurations as large as 192 GB, high-capacity RAM is relatively expensive.

To reduce the need for more costly RAM, ANSYS equation solvers often utilize vacant space on the hard disk drive (HDD). A 250-gigabyte (GB) hard drive can be purchased for approximately US $50 today, but the drawback is an order-of-magnitude speed penalty given the relatively slow input/output (IO) speed and extremely high seek time for random access files. This speed penalty results in significantly longer run times. However,Solid State Drives (SSDs) can significantly decrease data access latencies versus traditional hard drives and they also reduce power consumption and cooling requirements. Using SSDs with software from ANSYS is straightforward. Once the working directory for the software is set to the SSD file system, all I/O requests done by the software will then utilize the SSD. The recommended configuration for multiple drives is RAID-0 (redundant array of independent disks), which distributes data equally among them. In ANSYS Mechanical software, solutions involving the sparse solver, distributed sparse solver or block Lanczos Eigen solver running in the out-of-core memory mode typically perform the most amount of I/O and, therefore, see the most benefit from SSDs.

In one case study, a simulation of a turbine using Distributed ANSYS was performed. The simulation used the distributed sparse solver in out-of-core memory mode; this resulted in a considerable number of I/O requests to be performed during solution. To measure the benefit of Intel SATA Solid-State Drives we ran the simulation on 1, 2, 4 and 8 cores as shown in Figure 1. The out-of-core solver required a set of scratch files for each parallel process spawned for the additional cores.

First, the behavior of the HDD runs was examined. As more cores were used, the number of concurrent I/O requests increased progressively as well. As a result, I/O seek time increased and I/O bandwidth requirement went up proportionately with number of cores. At 8 cores, the I/O load dominated. This severely impacted the solution scalability going from 4 cores to 8 cores.

However, when we used SSD, the low latency and high I/O bandwidth of the SSD were in much better balance with the CPU performance. This yielded a smoother scalability as we moved from 1 to 2 to 4 to 8 cores. It also provided significantly better overall scalability as the 8 core SSD result was 5.4 times faster than the 1 core SSD result. Finally, the 8 core SSD result was almost 3 times faster than the HDD result on the exact same machine.




Figure 1. The analysis required about 30 GB of disk scratch space to run the ANSYS distributed sparse solver on a Intel® Xeon® 5560 processor workstation. Reduced seek times for the SSD significantly improved I/O performance, leading to shortened solution time as more cores are utilized. Details on Intel® X25-E Extreme SATA Solid-State Drives below in Appendix A.


Summary

ANSYS software running on workstations with Intel® Xeon® processors in the 5000 series and Intel® X25-E Extreme SATA Solid-State Drives forms a powerful simulation solution. This combination rewards engineers and designers with faster simulation results that allow them to produce better products in less time. Using Intel SATA Solid-State Drives is straightforward and is economical for simulations requiring large amounts of I/O. They provide a cost efficient way to run larger ANSYS simulations on workstations and can perform up to 3 times faster than traditional hard disk drives boosting engineering productivity and improved product design cycle.



About the Authors

Jeff Beisheim joined ANSYS in 2001 and is currently a Senior Software developer in the ANSYS Physics Business Unit. His work includes adding solver functionality for the ANSYS Mechanical products along with improving the architecture and performance of these products. He holds a M.S. in mechanical engineering from Carnegie Mellon University, Pittsburgh, PA.

Arvind Amin joined Intel in 2004 and is currently an application engineer in Intel's Software and Services Group (SSG) working on optimizing HPC applications to take advantage of the Intel Architecture and hardware as well as software innovations. Previously he worked at Cray and NEC as an application engineer focusing on optimization of parallel vector applications. He holds a Ph.D. in mechanical engineering from University of Pennsylvania, Philadelphia, PA.