Ch3-2

Computers in Spaceflight: The NASA Experience

- Chapter Three -
- The Skylab Computer System -
Hardware

[68] The choice of a central processor for the Skylab computer system marked a break from NASA's previous practice. The Gemini and Apollo computer systems were custom-built processors. Apollo did have an immediate predecessor, but the number of changes necessary before flight negated most of its resemblance to the Polaris system. To the contrary, Skylab and, later, the Shuttle, used "off-the-shelf" IBM 4Pi series processors, though they both needed the addition of a customized I/O system, a simpler and necessarily idiosyncratic component. By using existing computers, NASA avoided the serious problems associated with man-rating a new system encountered during the Apollo program.
The 4Pi descended directly from the System 360 architecture IBM developed in the early 1960s. Some 4Pis were at work in aircraft by the latter part of that decade. The top-of-the-line 4Pi is the AP-101, eventually used in the F-15, B-52, and Shuttle. The version on board Skylab was the TC-l, which used a 16-bit word, in contrast to the AP-101's 32 bits. A TC-l processor, an interface controller, an I/O assembly, and a power supply made up an ATMDC¹⁰. Each flight computer had a memory of 16,384 words¹¹. This memory was a destructive readout core memory, which means that the bits were erased as they were read and that the memory location had to be refreshed with the contents of a buffer register, which saved a copy of the bits before they were passed on to the processor. The memory was in two modules of 8K words each¹². Addressing ranged from 0 to 8K, with a hardware switch determining which module was being accessed¹³. The redundant computer system was composed of two processors attached to a single Workshop Computer Interface Unit. The unit consisted of two I/O sections (one for each computer), a common section, and a power supply¹⁴. Only the I/O section connected to the active [69] computer was powered. The inactive computer and its I/O section of the interface unit were not powered. The common section contained a 64-bit transfer register and timer associated with redundancy management¹⁵^. The transfer register and timer were the only parts of Skylab that consisted of triple modular redundant (TMR) circuits¹⁶. Basically, TMR circuits sent signals in triplicate on separate channels and then voted. The single output from a TMR voter represented either two or three identical inputs.
The final component of the computer subsystem was the Memory Load Unit. The original design did not contain one, but, like the Gemini Auxiliary Tape Memory, engineers later added it. Whereas the Gemini tape unit was useful in handling memory overloads, designers included the Skylab tape unit to further increase the reliability of the system. It carried a 16K software load and an 8K load that could be written into either module of either memory of the ATDCs. If up to three modules failed, the mission could continue with reduced capabilities with an 8K program loaded into the remaining module. This raised the total reliability of the system from a factor of 0.87 to 0.97¹⁷. The tape load would take a maximum of 11 seconds¹⁸.
NASA decided to add the Memory Load Unit in the summer of 1971, when both IBM and Marshall realized that a Borg-Warner tape unit, like the two already used as telemetry recorders, could be upgraded for program storage. IBM imposed some manufacturing changes on the recorders (primarily piece part screening) to make the process more nearly match the care taken in constructing the computers¹⁹.
NASA awarded the contract for the computer system to IBM on March 5, 1969²⁰. By October, designers froze the choice of processors and their configuration, a decision heavily influenced by the concern for redundancy and reliability²¹. The first computer was delivered on December 23, 1969. IBM eventually built 10, the final 2 being the flight versions, which went to NASA on February 11, 1972, over a year before launch. Two of the ATMDCs and an interface unit were turned over to IBM for use in testing both hardware and software, ensuring that the final verification would be on actual equipment rather than simulators²².
IBM took great pride in delivering on time without sacrificing reliability. In applying Saturn development techniques to the Skylab equipment, for example, IBM required all piece parts to exceed expected stress levels²³, and prepared the ATMDC for thermal conditions, the most dangerous stress to electronic components²⁴. A number of design problems, including thermal and vibration difficulties, analog conversion inaccuracies, and interconnection failures, had to be overcome²⁵. To make up time lost handling these problems, IBM sometimes went to a 7-day, three-shift debugging cycle²⁶.

[70]

Figure 3-2. The concept of Triple Modular Redundancy.

[71]
Figure 3-3. A block diagram of the Skylab Computer System with the dual ATMDCs, tape memory, and common section shown. (From IBM, Skylab Operation Assessment, ATMDC, 1974)

Probably due to the care taken in manufacture, the computer system had no failures. A planned ground-initiated switch-over from the primary to the secondary computer occurred after 630 hours of orbital operations. The second computer then ran the remainder of the 271-day mission²⁷. On the final day, the system did another switch-over and used the tape unit for the first time, primarily to prove that it would work. A transmission of software from the ground to the computer was also practiced. IBM's reports of the performance of the hardware are quite self-congratulatory but, based on the actual record, justified.