analysis tools

Total Elapsed Time (sec)

Boosting Memory Capacity with SSDs

45000 40000 35000 30000 25000 20000 15000
HDD SSD
Total Elapsed Time (sec)

By Jeff Beisheim, Senior Software Developer, ANSYS, Inc.

Using 1 core

40000 30000 35000 25000 30000 20000 25000 15000 20000 10000 150005000 10000 5000 0 Using 1 core
HDD SSD HDD SSD

One of the main factors limiting the size of simulations that can be run on a workstation is the amount of physical memory (or RAM) available on the machine. More RAM can be added, but only so many slots are available on the machine, and highcapacity RAM is relatively expensive. To help alleviate this difficulty, equation solvers often utilize vacant space on the hard disk drive (HDD) the spinning medium that provides a fairly inexpensive means for long-term storage of large amounts of data. A 250-gigabyte (GB) hard drive about as small as you can get these days can be purchased for only about $50 (U.S.), for example. The drawback is an order-of-magnitude speed penalty, with the relatively slow input/output (I/O) speed of the HDD generally below 100 MB/second, often resulting in significantly longer run times.
45000 40000 40000 30000

0 SSD Scalability of Distributed ANSYS Simulation Using 1 core

4000 4000
4000

3500 3500
Total Elapsed Time (sec)

Total Elapsed Time (sec)

10000 Solid-state drives are a cost-effective way to add 5000 fast memory to workstations. 0

45000 40000 45000 35000

Total Elapsed Time (sec)

Total Elapsed Time (sec)

3000 3000 2500 2500 2000 2000 1500 1500 1000 1000 500 500 0 0

3500 4000 3000 3500 2500 3000 2000 2500 1500 2000 1000 1500 500 1000 0 500

Total Elapsed Time (sec)

HDD

HDD HDD

SSD HDD SSD

SSD SSD

1 core 1 core

2 cores 2 cores

4 cores 4 cores

8 cores 8 cores

11 core Core

22 cores Cores

4 cores 4 Cores

8 cores 8 Cores

Considerable I/O was performed in an analysis requiring about 30 GB of disk space to run the ANSYS distributed sparse solver on a workstation containing only 24 GB of RAM. The reduced seek times for the SSD significantly improved I/O performance, thus helping to shorten solution time as more cores are involved.

Total Elapsed Time (sec)

HDD 35000 25000 Solution Time for ANSYS Mechanical Modal SSD 30000 20000 Analysis with Block Lanczos Eigensolver HDD 25000 15000

Total Elapsed Time (sec)

45000 35000

45000 45000 40000 40000

20000 10000 150005000 10000 5000 0 Using 1 core 0

SSD

Using 1 core

35000 35000 30000 30000

Total Elapsed Time (sec)

Total Elapsed Time (sec)

4000 3500 4000 3000 3500 2500 3000 2000 2500 1500 2000 1000 1500 500 1000 0 500

Total Elapsed Time (sec)

25000 25000 20000 20000 15000 15000 10000 10000 5000 5000

HDD HDD SSD SSD

1 core 1 core

2 cores 2 cores

00

For this study, approximately 1 million degrees of freedom (DOF) were analyzed for 200 frequencies. Elapsed times are compared for simulations on a workstation having two file systems: one with a single SCSI 10k rpm hard disk drive, another with four Intel X25-E 64 GB SATA SSDs.

Using 1 core

Using 1 Core

This is where relatively inexpensive solid-state drives (SSDs) with minimal seek times in reading files can be utilized to dramatically quicken solution times, particularly with simulations involving high levels of I/O to be performed. While expensive relative to HDDs, SSDs are considerably faster than HDDs and are approximately two to three times less expensive than RAM for the same number of gigabytes. SSDs also offer several other advantages over HDDs, such as no noise and high mechanical reliability due to the lack of moving parts although SSDs do consume more electricity than HDDs and have a useful life limited by the number of times data can be written and rewritten at a given location. Using SSDs with software from ANSYS is fairly 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
HDD SSD HDD SSD

4 cores

8 cores

4 cores

8 cores

(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 eigensolver 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 such modal analysis, transferring 3 terabytes (TB) of data to and from the disk took nearly 40,000 seconds on a single HDD versus about 5,000 seconds using SSDs in a RAID-0 configuration. In another case, a turbine analysis with the ANSYS distributed sparse solver requiring large amounts of I/O on an eight-core workstation was completed in less than 600 seconds with SSDs, compared with more than 1,700 seconds on HDDs. For this simulation, each Distributed ANSYS process had to read/write its own set of files, resulting in a considerable n u m b e r o f I /O re q u e s t s t o b e performed during solution. n
37

Total Elapsed Time (sec)

4000 3500

www.ansys.com

ANSYS Advantage Volume IV, Issue 1, 2010