Memories that Shaped an Industry : Decisions Leading to IBM System/360 (History of Computing)

Rating:
Summary: Thanks for the Memories
Review: Memories That Shaped an Industry by Emerson W. Pugh

A review by Frederick A. Ware

This was a fascinating book, covering the first 25 years of computer memory technology. The story includes technical details, legal battles, manufacturing disasters and personality clashes. It has a definite IBM-centric viewpoint, but that can probably be excused given IBM's dominance of the industry in this time period. The dominance was due in large part to IBM's mastery of ferrite core technology, as the book explains convincingly.

The early stored program computer was critically dependent upon the size and speed of its main memory. After the Eniac project demonstrated the ineffectiveness of patch cables for the (instruction) sequencing of digital calculating machines, it was generally recognized that instructions as well as data must be held in a central memory. The read and write time of this memory would effectively determine the performance of the computer, since an instruction would need to be accessed in every cycle. Several technologies were developed and discarded before the computer industry settled on a relatively robust solution - the ferrite core memory cell.

The book is chronological, with the relevant time period divided into roughly three intervals:

1945-50 Vacuum tube processors with delay line or CRT memories

1950-55 Vacuum tube processors with ferrite core memories

1955-70 Transistor processors with ferrite core memories

The ferrite core memory had a lifetime of roughly two decades before being supplanted by semiconductor memory in (about) 1970.

Chapter one covers the post-WWII period. The emerging computer industry had a number competing groups. Aiken of Harvard and IBM had cooperatively developed a series of electromechanical calculators with punched card sequencing. The Eniac team from the University of Pennsylvania formed the Univac Corporation. The Eckert mercury delay line was the basis of the main store of their early machines. Another IBM group developed the 603/604 series of vacuum tube calculators. They went on to build the IBM 701, a vacuum tube processor with Williams electrostatic CRT's for the main store.

Chapter two describes the early ferrite core experiments. The idea was developed independently by Wang of Harvard (founder of Wang Labs), Haynes of the University of Illinois, Forrester of MIT, Raschman of RCA, and Eckert of the University of Pensylvania all worked on the basic idea in the mid-to-late 1940s.

The basic idea was to build a small ring out of magnetic material which had a large magnetization threshold. Cores would be placed in a two dimensional array with independent sets of accessing wires in the x any directions. A single core is accessed by pulsing one x and one y wire simultaneously. Current pulses in a single x or y wire would be unable to flip a core's magnetization, but the single core addressed by both an x and a y wire would be magnetized in the proper direction (for writing a bit). Reading was destructive - a zero would be written into a core, and a third sense wire would have a different type of pulse depending upon whether the core originally had a zero or one.

Chapter three describes the efforts of an MIT group to build the Whirlwind computer (with core memory) for the Department of Defense. Chapter four describes the follow-on project (Sage). These were the first real-time computers. The Sage processor had a 64x64x36 core memory with a 7.5ns cycle time. It was delivered in 1953. Each 64x64 plane required 40 hours of hand labor to assemble.

Chapter five describes a number of commercial systems developed with core memory. These include: IBM 702 tape buffer in 1953

Sperry Rand 1103 computer in 1954

IBM 704 computer in 1955 (64x64x36 memory)

IBM 709 computer in 1957 (512x256x36 memory)

Initial resistance to core memory disappeared once it was demonstrated that it had a 35x lower error rate than the Williams electrostatic CRT memory. The core cost in 1955 was about one dollar per bit.

Chapter six describes the Stretch computer development at IBM. It was to have a 10MHz processor clock and a 2us memory cycle time (128x128x72). It would have about 100x the performance of the IBM 704, and was developed for the Atomic Energy Commision. It was delivered in 1956. Technology from this project went into the 7090 scientific computer and 7080 business computer. Within a year or two, Ferranti, Sperry-Rand and Control Data Corporation had developed machines comparable to Stretch.

Chapter seven covers the period in which IBM develops the 360 computer family. The goal was a family of five CPUs which had compatible instruction sets, but which spanned a performance range of 100x. They would be delivered in the 1964-66 time frame. They would consist of:

Model 360/30 0.5xIBM709 1 byte/cycle

Model 360/40 1.2xIBM709 2 byte/cycle

Model 360/50 4.5xIBM709 4 byte/cycle

Model 360/60 12.0xIBM709 8 byte/cycle w/ 2-way interleaving

Model 360/70 48.0xIBM709 8 byte/cycle w/ 4-way interleaving

The memory bandwidth spanned a32x range, approximately matching the performance target.

Chapter eight covers some of the legal battles fought over core memory technology. Since much of the early work was done by researchers at different institutions and organizations, multiple cross licenses were required by any company manufacturing core memory. IBM paid about one cent per bit in royalties to MIT and RCA in the 1960's. These continuing payments were one reason why IBM investigated many alternate memory technologies, including: monolithic magnetic films cryogenic memory dual core memory multi-hole core memory plated wire memory semiconductor memory

These other technologies were investigated because of performance and manufacturing cost issues with ferrite core technology. Eventually semiconductor memory became the obvious winner, and has dominated memory technology ever since.

Chapter nine wraps up the story, but leaves the reader ready for an accounting of the last thiry years of semiconductor memory technology.

Rating:
Summary: Thanks for the Memories
Review: Memories That Shaped an Industry by Emerson W. Pugh

A review by Frederick A. Ware

This was a fascinating book, covering the first 25 years of computer memory technology. The story includes technical details, legal battles, manufacturing disasters and personality clashes. It has a definite IBM-centric viewpoint, but that can probably be excused given IBM's dominance of the industry in this time period. The dominance was due in large part to IBM's mastery of ferrite core technology, as the book explains convincingly.

The early stored program computer was critically dependent upon the size and speed of its main memory. After the Eniac project demonstrated the ineffectiveness of patch cables for the (instruction) sequencing of digital calculating machines, it was generally recognized that instructions as well as data must be held in a central memory. The read and write time of this memory would effectively determine the performance of the computer, since an instruction would need to be accessed in every cycle. Several technologies were developed and discarded before the computer industry settled on a relatively robust solution - the ferrite core memory cell.

The book is chronological, with the relevant time period divided into roughly three intervals:

1945-50 Vacuum tube processors with delay line or CRT memories

1950-55 Vacuum tube processors with ferrite core memories

1955-70 Transistor processors with ferrite core memories

The ferrite core memory had a lifetime of roughly two decades before being supplanted by semiconductor memory in (about) 1970.

Chapter one covers the post-WWII period. The emerging computer industry had a number competing groups. Aiken of Harvard and IBM had cooperatively developed a series of electromechanical calculators with punched card sequencing. The Eniac team from the University of Pennsylvania formed the Univac Corporation. The Eckert mercury delay line was the basis of the main store of their early machines. Another IBM group developed the 603/604 series of vacuum tube calculators. They went on to build the IBM 701, a vacuum tube processor with Williams electrostatic CRT's for the main store.

Chapter two describes the early ferrite core experiments. The idea was developed independently by Wang of Harvard (founder of Wang Labs), Haynes of the University of Illinois, Forrester of MIT, Raschman of RCA, and Eckert of the University of Pensylvania all worked on the basic idea in the mid-to-late 1940s.

The basic idea was to build a small ring out of magnetic material which had a large magnetization threshold. Cores would be placed in a two dimensional array with independent sets of accessing wires in the x any directions. A single core is accessed by pulsing one x and one y wire simultaneously. Current pulses in a single x or y wire would be unable to flip a core's magnetization, but the single core addressed by both an x and a y wire would be magnetized in the proper direction (for writing a bit). Reading was destructive - a zero would be written into a core, and a third sense wire would have a different type of pulse depending upon whether the core originally had a zero or one.

Chapter three describes the efforts of an MIT group to build the Whirlwind computer (with core memory) for the Department of Defense. Chapter four describes the follow-on project (Sage). These were the first real-time computers. The Sage processor had a 64x64x36 core memory with a 7.5ns cycle time. It was delivered in 1953. Each 64x64 plane required 40 hours of hand labor to assemble.

Chapter five describes a number of commercial systems developed with core memory. These include: IBM 702 tape buffer in 1953

Sperry Rand 1103 computer in 1954

IBM 704 computer in 1955 (64x64x36 memory)

IBM 709 computer in 1957 (512x256x36 memory)

Initial resistance to core memory disappeared once it was demonstrated that it had a 35x lower error rate than the Williams electrostatic CRT memory. The core cost in 1955 was about one dollar per bit.

Chapter six describes the Stretch computer development at IBM. It was to have a 10MHz processor clock and a 2us memory cycle time (128x128x72). It would have about 100x the performance of the IBM 704, and was developed for the Atomic Energy Commision. It was delivered in 1956. Technology from this project went into the 7090 scientific computer and 7080 business computer. Within a year or two, Ferranti, Sperry-Rand and Control Data Corporation had developed machines comparable to Stretch.

Chapter seven covers the period in which IBM develops the 360 computer family. The goal was a family of five CPUs which had compatible instruction sets, but which spanned a performance range of 100x. They would be delivered in the 1964-66 time frame. They would consist of:

Model 360/30 0.5xIBM709 1 byte/cycle

Model 360/40 1.2xIBM709 2 byte/cycle

Model 360/50 4.5xIBM709 4 byte/cycle

Model 360/60 12.0xIBM709 8 byte/cycle w/ 2-way interleaving

Model 360/70 48.0xIBM709 8 byte/cycle w/ 4-way interleaving

The memory bandwidth spanned a32x range, approximately matching the performance target.

Chapter eight covers some of the legal battles fought over core memory technology. Since much of the early work was done by researchers at different institutions and organizations, multiple cross licenses were required by any company manufacturing core memory. IBM paid about one cent per bit in royalties to MIT and RCA in the 1960's. These continuing payments were one reason why IBM investigated many alternate memory technologies, including: monolithic magnetic films cryogenic memory dual core memory multi-hole core memory plated wire memory semiconductor memory

These other technologies were investigated because of performance and manufacturing cost issues with ferrite core technology. Eventually semiconductor memory became the obvious winner, and has dominated memory technology ever since.

Chapter nine wraps up the story, but leaves the reader ready for an accounting of the last thiry years of semiconductor memory technology.