The Photo Store

In the late 50’s at Livermore we began to think about on-line storage. Theretofore computer operators would bring magnetic tape to the machine to resume a simulation or to run graphics programs to produce reports on the state of a simulation resting on tape. Also there were card drawers scattered in offices where we kept the sources of programs and decks of input parameters for simulations. We began to think about a storage system that would keep this data to be retrieved automatically by application logic, or on command by someone at a timesharing terminal. Timesharing did not yet exist but it was on the horizon. Today such ideas are called ‘storage area networks’.

I do not recall what other developments influenced our thoughts but while we initially thought the storage must naturally be read-write, others had observed that write once storage had its charms, especially if it cost much less per bit.

We considered how much data we could usefully store, first without considering cost. Bandwidth to memory for the vacuum tube machines of the day suggested that our machines would produce about 1015 bits in the economic lifetime of a storage system. Considering the engineering cost of getting those bits out of the machines caused us to revise our goal down to 1014 bits.

Considering that the important transactions were to supplant people moving tapes and card decks around, we thought that access times of a few seconds would be ample, about one access per second, and perhaps 106 bits/sec would suffice.

Looking at descriptions of incipient storage technologies convinced us that 1014 would be difficult and we observed that the then current practice of keeping only one out of every 50 or so simulation steps was nearly as valuable as keeping every step. 1013 bits seemed useful and worth talking about to hardware vendors.

We talked to several computer manufactures. One of the interesting proposals was the ‘basement magnetic tape’ proposal described briefly here. ‘Basement’ had to do with the notion that operators would seldom be seen around these tape drives; they might be somewhat remote from the machines.

Ultimately IBM proposed to build a system with 1012 bits for a price that seemed worthwhile. IBM was also more credible than most. A contract was let which included development of the hardware. IBM sold the system as the ‘IBM 1360’ but it was broadly known as the Photostore. Lawrence Berkeley labs also got such a system for storing bubble chamber data. The CIA got a similar system with some modifications.

The Livermore system was indeed delivered and managed by a software system running on a DEC PDP-6. An index on a 1010 bit IBM data cell kept track of the contents of the store. A 109 bit fixed head Burroughs disk served as a data buffer between the store and the client machines. After an extended period of software development the store became very useful and its data overflowed to a substantial offline extension. It was eventually supplanted by a newer commercial offering.

Livermore’s Photostore is now on display at the Computer History Museum. This YouTube video shows many views of the complex mechanics of the photostore. There are a few unimportant errors.

The photostore recorded data on small rectangular chips of photographic film about one inch by two inches. The unexposed film was placed in a vacuum and written with an electron beam. Horizontal and vertical registration were written by the electron beam. After writing the chips went through a series of automatic development steps. Each chip held several hundred thousand bits. About 32 chips were kept in a box about the size of a pack of cigarets. These boxes were moved about pneumatically and kept in ‘drawers’. There were many moving parts and the practice was to build discrete electronics to sequence the solenoids that drove the mechanics. Of course there were many sensors that informed the electronics as well.

The following is by word-of-mouth. IBM wrote a program for the new IBM 1800 binary mini computer and this program simulated the yet unbuilt control electronics. It was fast enough to work with the real mechanical system. IBM decided that an 1800 running the simulator would be cheaper than the electronics and the 1800 became a standard part of customer systems.