Database structures.


Most databases have a structure with a header record. The header record
is at the beginning of the datafile and contains information about the
size of the header record and size of the database records. When a data
file is opened for reading the first thing to read the header record.
The field recsize contains the size of the datarecords and the field
hdrsize the offset to the first datarecord in the file.

If in the structure the size of the datarecords changes (grows), we can
allways read the old format in the correct way.

When a datafile is changed the datafile has to be rewritten completly.
Of course the new format is used then, and the new size must be stored in
the header.

The advantage of this technique is that updates can be performed automatic.
There is no need for free space for future use in the datarecords, the files
are thus smaller.


One other important thing, with some DOS based bbs'es, mail/tic processors
are using index files together with the data files to speed up the search in
the databases. Also some of them use internal memory cache for the data records.
I choose not to do this for two reasons, Linux like other Unices handles
file I/O very fast and when your system is not low on memory the kernel will
buffer all disk I/O in memory. Also Linux disks are very low fragmented due to
the design of the ext2fs. Whith all this in mind, using index files is only
extra overhead.
However, because of this you should not put the data files on a msdos
dos partition or on a nfs server.

The only exeption that uses index files are the nodelists.