The Functional Database Before OLAP
Part 1 of 5The standalone-engine era of TM/1, 1983–1990.
This series examines TM/1 in its first form: a standalone, in-memory calculation engine sold for the IBM PC between 1983 and 1990, before OLAP had a name and before the acquisitions that eventually brought the product to IBM.
Reading time: 6–8 minutes
In Why TM1 Matters, I argued that TM1 deserves historical attention because it survived every major transformation in business computing. To understand that survival, however, we first need to understand the problem TM/1 was originally designed to solve.
The Analysis Gap
To understand why someone would build TM/1, you have to recover a problem that the early 1980s felt acutely and that we have largely forgotten: there was no good way to analyse business data on a computer.
By 1980 the relational model was ascendant. Edgar Codd’s 1970 paper had given the industry a rigorous foundation for storing and retrieving records, and through the decade relational databases became the orthodoxy for transaction processing — orders, accounts, inventory, payroll. They were superb at recording what happened. They were poor at answering the questions managers actually asked: what if margins move three points across these regions and quarters? The relational model organised the world into rows and tuples; it had no native sense of a budget laid out by product, by month, by scenario, consolidated upward.
The Spreadsheet Ceiling
Into that gap walked the spreadsheet. VisiCalc, in 1979, was a genuine revolution — it gave a non-programmer a live, recalculating grid and a feeling of direct manipulation that mainframe reporting never offered. Lotus 1-2-3 followed in 1983 and became the killer application of the IBM PC. But the spreadsheet had a ceiling baked into its shape: it was two-dimensional. The moment a real business model needed a third axis — the same statement across twelve months, or five regions, or three scenarios — the analyst was reduced to copying the grid onto page after page and stitching the totals together by hand. Interactivity without a data model.
The Forgotten Multidimensional Tradition
There was, in fact, an older tradition that had already confronted multidimensionality. Array-oriented languages such as APL had treated the multidimensional array as a first-class object since the 1960s, and on mainframes a product called Express had offered multidimensional analytical computing since the early 1970s. But these lived in the world of expensive hardware and specialist users. They were not something a finance manager ran on the machine on their desk.
So the gap was specific and real: the relational world stored data it could not analyse; the spreadsheet world analysed data it could not model; and the multidimensional tradition that could do both was locked on the mainframe. TM/1 is best understood as an attempt to put a small piece of that mainframe capability — a true multidimensional, calculation-first engine — onto a personal computer, behind a spreadsheet that ordinary users already understood.
A Warning About OLAP
Now the methodological warning that will recur throughout this series, because it matters most here at the start. It is tempting to call TM/1 an “early OLAP engine.” Resist it. The term OLAP — Online Analytical Processing — was not coined until 1993, in a white paper by Edgar Codd, his colleague Sharon Codd, and Clarke Salley, a paper whose commercial sponsorship by a competing vendor made it controversial almost immediately. A more neutral framing, the FASMI definition (Fast Analysis of Shared Multidimensional Information), came later still. To describe a 1984 product with a 1993 vocabulary is to flatten the very thing worth noticing: TM/1 arrived a decade before the category that would later claim it had a name. Its designers were not building “OLAP.” They were solving the analysis gap with the concepts and the language available to them in 1983. Reading the standalone-engine era on its own terms — without the OLAP label, without the Planning Analytics endpoint — is the whole purpose of what follows.
Sources
This post is intentionally light on archival citation; it sets the intellectual scene, and the primary evidence arrives in later posts. The historical claims about the relational model (Codd, 1970), VisiCalc (1979), Lotus 1-2-3 (1983), the APL array tradition, and Express are standard, widely documented history of computing. On the coining of “OLAP,” see the 1993 Codd, Codd & Salley white paper Providing OLAP to User-Analysts and the subsequent FASMI framing associated with Nigel Pendse and Richard Creeth’s OLAP Report. The framing of TM/1 as a response to this gap is my own, supported by the primary material discussed in Posts 2–4.
Series Navigation
→ Origins and the Initial Release: A Problem of Dates