The Mechanical APDL program, by default, creates only one (sometimes large) file for each of the various file types. In the past, many systems had smaller disk partitions than were required for some of the very large files, or were limited to a maximum single file size of 2 GB. To accommodate this hardware limitation and provide the flexibility to store very large files across multiple disk systems, you can specify split file sizes and divide files across multiple disks.
By default, file are never split. You can, however, set your own file sizes for split files. The split file sizes are defined in units of Mwords (1024*1024 4-byte integer words).
You can enable file splitting for jobs using the FILESPLT directive in your config.ans file. (An alternative method is to use the /CONFIG,FSPLIT command.) The default setting within Mechanical APDL is the recommended option in almost all cases. File splitting will almost always increase run time, particularly if different disks are used for various file partitions. The best way to improve I/O performance on modern systems is by using multiple drives which are configured in a RAID0 striped configuration. (See the Performance Guide for more information on improving performance.)
To use the file-splitting feature, first determine a split point for your files. Then, any file types that can be split and are larger than that split point will be split across partitions. Select a split point carefully. Split files are less efficient than single files, and to reuse split files, you must match the split points correctly. Do not set your split point so low that small files would be split unnecessarily.
Figure 7.1: Determining Split Points illustrates one way to determine an appropriate split point. In this figure, if you chose split point 1, all of the file types would be split. If you chose split point 2, however, only two of the four file types need to be split.
You can split the following files. See the Basic Analysis Guide for file descriptions.
file.full
file.emat
file.erot
file.esav
file.mass
file.sub
file.mode
file.rdsp
file.rfrq
file.rst
file.rth
file.rmg
file.dsub
file.usub
file.db
file.page
To split files across partitions, follow these steps:
Issue the df command in Linux to determine which partitions will hold the split files. To achieve the most efficiency, create as few split files as possible.
Determine the least amount of space available across those partitions. This number will be the size of each split file. Adjust this number downward to be conservative.
Calculate the split point in terms of megawords (millions of words) by dividing the maximum size of the file to be split in bytes by 4194304 (4x1024x1024).
Place the config.ans file in the current working directory. The file should contain the following line:
FILESPLT=xxxx
In the above line, xxxx is the integer value
indicating the file split point in megawords. With this config.ans
file in place, the file will be split into a new file every increment of xxxx megawords. These files are all created in the current
directory and are named Jobname.xxxnn, where XXX
indicates the file extension (see list, above), and nn is a two-digit number from 02 to 99.
To place these files in different partitions, symbolically link the name
Jobname.xxxnn to a
pathname on another partition with the ln -s command in
Linux. For example, to split a file.emat and write a second file to
a different partition:
ln -s /tmp/file.EMAT02 file.EMAT02
The above example puts a second file.emat into the /tmp partition.
When you finish creating the appropriate symbolic links, you are ready to execute the Mechanical APDL program.
You cannot transport the split files to other systems.
Caution: Unpredictable results can occur if you try to reuse a split file for substructures or restructuring and the split point has changed. Ansys, Inc. recommends discarding the split files after the run without reusing them. If you do intend to reuse the split files, the split points must match correctly. Also be aware that using split files can result in a slight performance degradation.