Computer Engineering 30 Years After the IBM

ften in our hurry to create the future in computer engineering, we ignore lessons from past projects. One project that offers interesting insights, the IBM System/360 Model 91, recently celebrated the 30th anniversary of its first shipment. The largest member of the series based on IBM’s most enduring and influential computer architecture, this machine owed much to its predecessors. Together, these machines show us what the state of the art in computer processors was like in the 1960s—when IBM announced the S/360— and illustrate how far computer engineering has come in 30 years. This leads me to some reflections on how this field has changed and needs to change for the future. PREDECESSORS In addition to the competition from the CDC 6600,1 there were three important influences on the Model 91 design. The first was the IBM 7030 “Stretch” machine,2 the company’s first large, pipelined, high-speed machine. IBM began the project in the late 1950s and delivered it in the early 1960s. The 7030 embodied several imaginative machine technology and architectural concepts, including what we now know as emittercoupled logic (ECL) circuits. A fully pipelined system with highly interleaved memory, it used speculation on branch as determined by a bit set in the branch instruction. Despite these innovations, the 7030 was a financial failure because of late delivery as well as disappointing performance, which partially stemmed from the long delay in recovering from a mispredicted branch. Cy be rs qu ar e