• Back to all articles

Non-SQL Databases for Linux

The best-known databases these days are based on SQL, but are often overkill for what you need to do. This review discusses lighterweight alternatives, including xBase, DBM, and ISAM systems.

xBase Descendants

Back in the 1970s, a database system was released for the CP/M operating system called Vulcan. Ashton Tate bought the rights to it, and renamed it dBase. It went through a number of iterations and version numbers, and a number of companies produced "clones".

The architecture was interesting; it combines these features:

• Each table is represented as a file, as is each index.
• There is a "table browser" that allows viewing and editing a table.
• There is a "form builder" that allows you to fill in fields based on application control that might involve multiple tables.
• There is a somewhat BASIC-like language for constructing reports based on querying tables.

Using all of this, you can build quite powerful interactive applications to work with a set of database tables.

In the early days, the tools for building "forms" were somewhat primitive (they got more sophisticated over time). In those days, personal computers were not connected to networks, so programs were inherently single-user-oriented. When PCs started getting networked together, locking schemes were introduced to allow use by multiple concurrent users.

Much later, an ANSI committee, X319J, created a common "dialect" called xBase, trying to unify the functionality of notable commercial implementations such as dBase, Clipper, and FoxBase, with the result that it is typical to call these sorts of database systems "xBase systems".

Early implementations functioned as file-based systems, that is, programs accessed data through the OS filesystem, in which each table and index is represented as a file. This approach doesn't scale very well from the perspectives of either reliability (because any program on the system that accesses a table file has the ability to "muss it up") or scalability (because every program that accesses the database has to manage locks itself).

Some implementations have become available that have a central database manager process, as is common with SQL databases. Furthermore, SQL interpreters have sometimes been added to the set of tools, so "xBase systems" are sometimes really SQL systems.

Various xBase implementations are available for Linux:

• FlagShip -- essentially a Clipper "clone"; it compiles dBase III+ (and higher) code, and reads and writes related file formats. Free Personal FlagShip is an unlimited 2-user version similar to the commercial Personal license, but intended strictly for personal use or for development of database applications distributed for free (whether Public Domain or Open Source).
• The Harbour Project -- building a Clipper "clone" for DOS and Unix that is freely redistributable.
• PlugSys International
• Recital
• CodeBase
• DBF to other formats conversion software
• X2c -- a portable xBase compiler.
• Xbase -- a collection of specifications, programs, utilities and a C++ class library for manipulating xBase type datafiles and indices.
• XBSQL -- a wrapper library providing an SQL-like interface to the Xbase DBMS.

General xBase Documentation

• xBase File Format Documentation
• The xBase Fan Club
• whatxb
• Info about xBase
• dBase 2000 "Open Source" promotion
• The Oasis Clipper Source

Keyed Table Systems like DBM

At the "lowest level", there are quite a number of data storage systems that don't try to be terribly abstract, or to provide a complete "application environment". The characteristic example is the Unix DBM scheme, which provides a set of C function calls that allow you to store values -- associated with keys -- into a data file.

If your data storage needs are simple, it may not make sense to pull in the full sophistication of an SQL system. Furthermore, the "serious database" systems typically tend to require some administration effort, often including setting up server processes, user authentication, and the likes. If you have an application that merely needs to "store some data", a DBM-like system may be all you need.

Some SQL database systems are (or have been) based on these sorts of libraries. For instance, Informix implemented a C-based ISAM library that was often embedded in applications; the Informix SE SQL database system was implemented on top of that, as tables were represented as ISAM tables.

There are a couple of SQL databases that have been built atop DBM. One of the most interesting examples is that "transactions" were brought to MySQL when they attached it to Berkeley DB, a modern version of DBM that supports transactions and storage of multiple "tables" within a single data file.

The major families of these databases include:

DBM-like databases that allow storing "associative arrays" on disk.

These are usually thought of as involving hash tables, but sometimes use B-Trees.

ISAM databases

ISAM stands for "Indexed Sequential Access Method", an indexing system that allows rapidly seeking to appropriate locations in a data file. Since data is stored in sequential order, efficiency of use of disk space is generally quite good.

These systems tend to be highly API-oriented; while an SQL database often provides a lot of generic tools for building queries, and you tend to describe your query, these sorts of databases almost always require writing programs to "walk" through the data.

DBM-Like Databases

Linux systems almost always include some set of NDBM, the "New" DBM implementation, SDBM, ODBM, and GDBM, the "GNU" implementation.

The typical API looks like:

DBM *dbm_open(char *, int, int);
void dbm_close(DBM *);
datum dbm_fetch(DBM *, datum);
datum dbm_firstkey(DBM *);
datum dbm_nextkey(DBM *);
int dbm_delete(DBM *, datum);
int dbm_store(DBM *, datum, datum, int);

The Perl language popularized the idea of tying DBM tables to Perl associative arrays, with the result that in Perl, once you tie a name to a DBM file, you can transparently use ordinary assignments like $A["this"] = "that", rather than something like dbm_store(A, "this", "that", 4).

The Python AnyDBM_File documentation page describes some of the similarities and differences between different DBM implementations.

• The Berkeley DB Package, probably the most sophisticated such system, offers the ability for multiple hosts to access a database, multiple storage schemes (e.g., hash tables, B-Trees), distributed locking, and other pretty neat stuff.
• cdb, the "constant" database, is quite interesting; it does not cope well with updates, but provides extremely fast access to static data. The canonical use of it is as a way of storing mail routing information for the qmail mail server.
• rdbm (a reliable database) layers a DBM-like interface on top of cdb.
• bun (bundle many files together), based on the cdb format, provides something like tar, implemented atop a (tiny) database system.
• Dx

ISAM Databases

These systems are often embedded inside applications.

• Informix C-ISAM
• Open Group ISAM -- a standard based on Informix C-ISAM.
• Birdstep Raima Database Managers (formerly Raima Database Manager++, Velocis)
• Byte Designs -- one of the vendors of ISAM database software read and write compatible with Informix C-ISAM.
• ObjecTrieve -- another X/Open ISAM-based database manager.
• MixTools ISAM
• Faircom c-tree Plus
• C/BASE 4GL for Linux
• regexps.com -- the critical components of the bdb system have since been integrated into the The Hackerlab C Library.
• Bullet DBMS -- a commercial embedded DBMS that runs on Win32, Linux, and some other platforms.

Notice that many of the vendors of ISAM-like systems also sell SQL databases; once you've got the low level library to store and retrieve data, it is pretty natural to build further layers of abstraction on top of that, such as SQL interpreters. That brings us full circle back to SQL, where we started.