There are several methods that can be used to mitigate this effect. VxFS tunables set with the /opt/VRTS/bin/vxtunefs commands include:

 

initial_extent_size:

Changes the default size of the initial extent. VxFS determines, based on the  first  write  to  a  new file,  the  size of the first extent to allocate to the file. Typically the first extent is the smallest  power   of  2  that is larger than the size of the first write. I f that power of 2 is less than 8K,  the  first  extent allocated  is  8K.  After  the initial extent, the file   system increases the size of  subsequent  extents  (see max_seqio_extent_size) with each allocation.

Because most applications write to files using a buffer size  of  8K or less, the increasing extents start dou-bling from a small initial extent.  initial_extent_size   changes the default  initial  extent size to a larger value, so the doubling policy starts from a much larger   initial  size, and the file system won't allocate a set of small extents at the start of file.

Use this parameter only on file  systems  that  have  a   very  large  average  file  size. On such file systems,   there   are   fewer   extents   per   file   and   less fragmentation.

initial_extent_size is measured in file system blocks.

 max_seqio_extent_size:

Increases or decreases the maximum size of an extent. When the file system is following its default alloca-tion policy for sequential writes to a file, it allo-cates an initial extent that is large enough for the first write to the file. When additional extents are allocated, they are progressively larger (the algorithm tries to double the size of the file with each new extent), so each extent can hold several writes worth of data. This reduces the total number of extents in anticipation of continued sequential writes. When there are no more writes to the file, unused space is freed for other files to use.

In general, this allocation stops increasing the size of extents at 2048 blocks, which prevents one file from holding too much unused space.

max_seqio_extent_size is measured in   file system blocks.

 

These settings can help depending on the workload, but not in all cases.   Acceptable solutions were achieved by changing VxFS tuneables: “initial_extent_size” and “max_seqio_extent_size”, setting them to 32768 and then 65536.  In each case,  creating a new 500MB database file the first extent was 32MB or 64MB respectively, but there were hundreds of smaller extents after that.  The “initial_extent_size” tunable works for the first block, but the “max_seqio_extent_size” is not effective for sizing subsequent extents.

 

Another setting that can be used to manipulate the VxFS extent allocator is the ‘setext’ command. ‘setext –r 1049576 ’ will reserve 1GB space in as few VxFS extents as possible. Any allocations done within the first 1GB of such a file will not require interactions with the extent allocator and thus, no fragmentation will occur. Using such a method requires knowledge of how large this file might become so it may not be a good solution for long term dynamic Oracle database files where file size is constantly adjusted with the AUTOEXTEND feature.

 

An additional ‘setext’ command can he used that will also control the VxFS extent allocator: ‘setext –e 32m ’. This command will always attempt to allocate space in 32MB extents for any new allocations. It works best after the file has been first defragmented with commands such as ‘/opt/VRTS/bin/fsadm –ef ’ (per-file defragmentation) or ‘/opt/VRTS/bin/fsadm –e ’ (per-filesystem defragmentation). 

 

Another method of defragmenting existing files used at a recent customer escalation is to simply set the minimum extent size for the file and use the Veritas /opt/VRTS/bin/cp command to copy the fragmented file to a new filename. The following commands require the Oracle instance to be down:

/opt/VRTS/bin/setext -e 32m 

/opt/VRTS/bin/cp –e warn   

 

Then renaming the new file back to the original file name. The ‘cp’ will use 32MB extents and reduce fragmentation.

 

Setting the extent size to 32MB using ‘setext –e 32m ’ forces VxFS to always create 32MB extents (when possible) when doing allocating writes. This works quite well and minimizes extent fragmentation which results in better performance.

 

• Oracle data files can be fragmented at create time as well due to the parallel writer threads accessing the same file at different offsets.
• ODM handles AUTOEXTEND nicely
• In case ODM is not implemented, avoid using AUTOEXTEND instead add a new datafile to make extra space in TABLESPACE.
• In the event of using AUTOEXTEND and not using ODM on a VxFS filesystem, setting the minimum extent size using ‘setext –e ’ will prevent future extents from being created smaller than size, thus preventing new fragmentation.  Defragmenting the file system using 'fsadm -e ' will minimize existing fragmentation.