FITS Files and Physical Media

FITS is defined as a logical structure, independent of the medium on which the data appear. FITS files are regarded as bit/byte streams with a 2880-byte logical record length. Data written in FITS format can, for example, be on magnetic tape, magnetic disk, optical disk, or simply on-line. However, the reader of a FITS file must know its physical structure in order to unpack it.

Historically, magnetic tape was the original storage medium for FITS formats. The initial conventions for $1/2$-inch 9-track magnetic tape were established in the original FITS paper (Wells, Greisen, and Harten 1981) and modifications to these conventions were presented at the same time as the rules for extensions (Grosbøl et al. 1988). The IAUFWG has since endorsed an additional set of formal rules to relate physical and logical structure for other media. Each medium has its own particular rules. The complete rules are as follows:

• For $1/2$-inch 9-track magnetic tape, physical blocks may contain one to ten logical records. Unlabeled tapes are strongly recommended; however, tapes with ANSI labels are permitted. Data reduction and analysis packages reading labeled tapes must count the tape label as the first file.

• For fixed block sequential media with a physical block size that is not an integral multiple or fraction of 2880 (for example, 512), the 2880-byte logical records are not tied to the physical block size but are written independently, over multiple blocks if necessary. The end of the last logical record of the FITS data set will probably not correspond to the end of a physical block. A short logical record extending to the end of the physical block will follow. Readers should regard a record of less than 2880 bytes as the equivalent of an end-of-file mark. Examples of media covered by these rules are optical disks accessed as a sequential set of records, QIC format $1/4$-inch cartridge tapes, and local area networks.

• For bitstream devices, FITS files are written with a blocking factor of one, i.e., with a physical block size of 23040 bits, the same as the logical record length.

• For variable block length sequential media, such as DDS/DAT 4-mm and 8-mm Exabyte cartridge tape, physical blocks may contain an integral number of logical records from one to ten, the same rule as for $1/2$-inch 9-track magnetic tape. In the case where the data system is able to generate only fixed blocks, the fixed block convention is used.

Except where a logical record of fewer than 23040 bits serves as an end-of-file indicator, a file should be terminated by the structure for the given medium that corresponds to the tape end-of-file mark. If more than one FITS file appears on a physical structure, the end-of-file indicator appropriate to the medium should separate the files, appearing immediately before the start of the primary header of the next file.

In the original FITS specification, the physical block size was set equal to the logical record length for simplicity. As time passed, however, it became clear that a significant number of the major institutes producing data regarded this physical block size as an inefficient use of tape. On a 6250-bpi tape, only 55% would contain data and 45% would be inter-record gaps; consequently, at the same time as the rules for extensions were developed, it was agreed that physical blocks on $1/2$-inch 9-track magnetic tape could contain one to ten logical records. Blocks containing more than one record were called long blocks. At that time, because the multirecord blocks were new and might not be expected by some FITS readers, the BLOCKED keyword was introduced as an indicator that there might be more than one logical record per physical block on a tapes. However, usage was not universal; therefore, absence of the BLOCKED keyword therefore did not guarantee single record blocks, and by its definition, its presence did not guarantee multiple-record blocks. It could thus not be used effectively as intended. As the multiple record block became established as standard, and with the rise of the concept of FITS as a purely logical structure, the need for a BLOCKED keyword has disappeared. Information describing how a tape is blocked should be sent with it, and users should ask for blocking information when receiving tapes. The keyword itself should no longer be used as a blocking indicator, but it remains formally reserved for its original purpose and should also not be used for any other.

(The rules for the BLOCKED keyword also were in conflict with other rules in FITS. This keyword was required to be in the first logical record; however, if the value of NAXIS was greater than 33--or 32 when EXTEND was present--it could not be put in the first record without violating the rule that the NAXIS$n$ keywords must be consecutive.)

Conventions for writing a FITS file on more than one physical unit (tape, disk) are only now being developed. In the original FITS paper, Wells, Greisen, and Harten (1979) suggested a method for handling multivolume unlabeled tape, but this proposal has not been put into practice widely. Many current programs that read FITS files will be unable to handle a multiple-unit file. The grouping proposal, discused in section 5.3, uses tables to associate physically separated FITS HDUs into a logical unit.