In an ISAM system, data is organized into
records which are composed of fixed length fields, originally stored sequentially in key sequence. Secondary set(s) of records, known as
indexes, contain pointers to the location of each record, allowing individual records to be retrieved without having to search the entire data set. This differs from the contemporaneous
navigational databases, in which the pointers to other records were stored inside the records themselves. The key improvement in ISAM is that the indexes are small and can be searched quickly, possibly entirely in memory, thereby allowing the database to access only the records it needs. Additional modifications to the data do not require changes to other data, only the table and indexes in question. When an ISAM file is created, index nodes are fixed, and their pointers do not change during inserts and deletes that occur later (only content of leaf nodes change afterwards). As a consequence of this, if inserts to some leaf node exceed the node's capacity, new records are stored in overflow chains. If there are many more inserts than deletions from a table, these overflow chains can gradually become very large, and this affects the time required for retrieval of a record.
Relational databases can easily be built on an ISAM framework with the addition of logic to maintain the validity of the links between the tables. Typically the field being used as the link, the
foreign key, will be indexed for quick lookup. While this is slower than simply storing the pointer to the related data directly in the records, it also means that changes to the physical layout of the data do not require any updating of the pointers—the entry will still be valid. ISAM is simple to understand and implement, as it primarily consists of direct access to a database file. The trade-off is that each client machine must manage its own connection to each file it accesses. This, in turn, leads to the possibility of conflicting inserts into those files, leading to an inconsistent database state. To prevent this, some ISAM implementations provide whole-file or individual record
locking functionality. Locking multiple records runs the risk of
deadlock unless a
deadlock prevention scheme is strictly followed. The problems of locking, and deadlock are typically solved with the addition of a
client–server framework which marshals client requests and maintains ordering. Full
ACID transaction management systems are provided by some ISAM client–server implementations. These are the basic concepts behind a
database management system (DBMS), which is a client layer over the underlying data store. ISAM was replaced at IBM with a methodology called
VSAM (virtual storage access method). Still later, IBM developed
SQL/DS and then
Db2 which IBM promotes as their primary
database management system. VSAM is the physical access method used in Db2. ==OpenVMS==