IBM Hub/Switch Installation Manual page 96

Chapter 2 HPSS Planning
2.9.1.5 Blocks Between Tape Marks Selection
Blocks between tape marks is the number of physical media blocks written before a tape mark is
generated. The tape marks are generated for two reasons: (1) To force tape controller buffers to flush
so that the Mover can better determine what was actually written to tape, and (2) To quicken
positioning for partial file accesses. Care must be taken, however in setting this value too low, as it
can have a negative impact on performance. For recommended values for various media types, see
Section 2.9.1.12.
2.9.1.6 Storage Segment Size Selection (disk only)
The Bitfile Server maps files into a series of storage segments. The size of the storage segments is
controlled by the Minimum Storage Segment Size parameter, the Maximum Storage Segment
Size parameter, and the Average Number of Segments parameter. The smallest amount of disk
storage that can be allocated to a file is determined by the Minimum Storage Segment Size
parameter. This parameter should be chosen with disk space utilization in mind. For example, if
writing a 4 KB file into a storage class where the storage segment size is 1,024 KB, 1,020 KB of the
space will be wasted. At the other extreme, each file can have only 10,000 disk storage segments, so
it wouldn't even be possible to completely write a terabyte file to a disk storage class with a
maximum storage segment size below 128 megabytes. When file size information is available the
Bitfile Server will attempt to choose an optimal storage segment size between Minimum Storage
Segment Size and Maximum Storage Segment Size with the goal of creating Average Number of
Segments for the bitfile. The storage segment size will also be chosen as a power of 2 multiple of
the Minimum Storage Segment Size parameter.
The smallest value that can be selected for the Minimum Storage Segment Size is the Cluster
Length. Cluster Length is the size, in bytes, of the disk allocation unit for a given VV. Cluster Length
is calculated when the VV is created using the PV Size, the VV Block Size and the Stripe Width.
Once the Cluster Length for a VV has been established, it cannot be changed. If the characteristics
of the Storage Class change, the Cluster Lengths of existing VVs remain the same.
The Cluster Length of a VV is always a multiple of the stripe length of the VV. If the VV has a stripe
width of one, the stripe length is the same as the VV block size, and the Cluster Length will be an
integer multiple of the VV block size. If the VV has a stripe width greater than one, the stripe length
is the product of the VV block size and the stripe width, and the Cluster Length will be a multiple
of the Stripe Length.
The number of whole stripes per Cluster is selected such that the number of Clusters in the VV is
less than or equal to 16384. This means that any disk VV can contain no more than 16384 Clusters,
which means it can contain no more than 16384 Disk Storage Segments. Since a user file on disk
must be composed of at least one Storage Segment, there can be no more than 16384 user files on
any given disk VV.
As the size of a disk virtual volume increases, its Cluster Length increases accordingly. This means
that the Minimum Storage Segment Size also increases as the disk VV increases in size. These
relationships are calculated and enforced by SSM and the Disk Storage Server automatically and
cannot be overridden.
Guideline: When a large range of file sizes are to be stored on disk, define multiple disk storage
classes with appropriate storage segment sizes for the sizes of the files that are expected to be stored
in each storage class.
96 September 2002 HPSS Installation Guide
Release 4.5, Revision 2