Thursday, 24 February 2011

Univac Fastrand II - what a beast of a storage device?

The FASTRAND™ II




FASTRAND II

The FASTRAND II random-access mass storage system was one of the most impressive peripherals ever attached to a commercial computer. Used with the UNIVAC 1108 computer, it provided the first permanent file storage capability in the UNIVAC 1100 series family. 

No UNIVAC programmer who ever encountered a FASTRAND is likely to forget it. It was big, heavy (it weighed about two and a quarter tons, and required special reinforcement of the raised floor it sat on), had a large window lit by fluorescent lights which let you see the two huge drums rotating in opposite directions at 880 revolutions per minute and the heads jumping back and forth as various tracks were accessed. In operation, it emitted a deep rumble accompanied by a whoosh of air much like the engine room in Star Trek. Opening the door on the end revealed a complex control panel which included, at the top, a collection of “hit detectors” which told you if one of the read/write heads had contacted the drum surface.

FASTRAND II Specifications




Storage Capacity 22,020,096 36-bit words;
132,120,576 6-bit characters,
or about 90 megabytes¹
Drum Speed 880 revolutions per minute
Read/write heads 64, on a movable boom between the two drums
Average access time 92 milliseconds
Transfer rate 26,283 36-bit words per second,
about 100 kilobytes¹ per second
Recording density 1000 bits per inch
Track density 105 tracks per inch
Price (1968 dollars) Controller: $41,680, Storage unit: $134,400
Maximum 8 storage units per controller

Twenty-eight

Another odd aspect of the FASTRAND was that it addressed data in blocks, or sectors of 28 36-bit words. Most storage peripherals on binary computers use a sector size which is a power of two, for example 32 or 256. If the sector size is a power of two, address computation, such as figuring out which sector to read and which word within it to access when presented with a word address in a file, can be done entirely by shifting and masking, which is almost always much faster than multiplying or dividing. For example, on the UNIVAC 1108, division took 10.125 microseconds, while a double-word shift of any number of bits required just 0.875 microseconds: eleven and a half times faster. 

Nobody ever seemed to know why the FASTRAND used 28 word sectors; the most common explanation was that the device had originally been designed to work with UNIVAC 490 series real-time computers with a word length of 30 bits and a FASTRAND sector size of 32. This may be the case, but the arithmetic doesn't quite work. If you take the number of bits corresponding to 28 36-bit words (1008), then divide by 30, you get a capacity of 33.6 30-bit words in the same sector. Why would they have wasted a word and a half in every sector? A device designed for 32 30-bit words would have a sector size of 26 36-bit words, with two-thirds of a word being wasted. Another purported explanation was that the FASTRAND had originally been designed for 32 word sectors, but wasn't reliable enough at the recording density that required, so the sector size was reduced to reduce bits per inch to a point where the hardware worked acceptably. 

Whatever the reason, the number 28 was never far from the mind of a UNIVAC systems programmer, since the EXEC-8 operating system made no attempt to sugar-coat the properties of the underlying hardware. When programming in assembly language, one allocated buffers as multiples of 28 words, calculated addresses by multiplying and dividing by 28 (one wag expressed surprise when the 1110 did not include special hardware for multiplying and dividing by 28), and since word-addressable drums were usually treated as high-speed FASTRAND emulators, the influence of 28 crept throughout the system, to the block sizes used on tape drives transferring data to and from the FASTRAND. 

Twenty-eight was referred to in jest as “von Fastrand's number”, after the (mythical) Baron Gustav von Fastrand, who it was said, discovered the number during a nineteenth-century expedition between the numbers 27 and 29. Even after the last FASTRAND had been long retired, its legacy, twenty-eight, lived on in UNIVAC 1100 Series software. Byte-oriented disc drive controllers had special hardware to repack 36-bit words efficiently into 8-bit bytes, and physical sector sizes, expressed in words, were always a multiple of 28.

Notes

  1. Since the UNIVAC 1100 series machines to which the FASTRAND II was connected used 36-bit words rather than 8-bit bytes, it isn't possible to precisely compare the capacity of a device such as the FASTRAND II with contemporary byte-oriented disc drives. Calculating the total capacity in bits and dividing by 8 yields a total capacity of 99 megabytes, but since UNIVAC software in the FASTRAND era used 6-bit FIELDATA character codes (by the time ASCII came into wide use, most FASTRANDs had been retired in favour of a variety of disc drives), the capacity in terms of contemporary computer documents was 132 million characters. When UNIVAC adopted ASCII in the 1100 series, they stored 4 characters in the four 9-bit fields of each 36-bit word, so the capacity of a FASTRAND II in terms of ASCII characters would be 88 megabytes. The same considerations apply to calculations of the transfer rate.
  2. The FASTRAND II was the second member of the FASTRAND family, and by far the most common. The ill-fated FASTRAND I had only one rotating drum and half the storage capacity. A single massive drum rotating almost 15 times a second acts as a powerful gyroscope which tries to stay in a fixed location with respect to the distant stars. Unfortunately, the Earth rotates, and this leads to a conflict between the Earthly imperative of motion and the FASTRAND I's desire to stay put, which resulted in the devices tending to move around the computer room. In the FASTRAND II, the two drums rotated in opposite directions, which cancelled out the gyroscopic effect. The story of the Navy ship which set sail with a spinning FASTRAND only to have it stand on end at the first course change is, as far as I can determine, apocryphal.
  3. Around 1970, the FASTRAND III was introduced. It was physically identical to the FASTRAND II but increased the recording density to from 1000 to 1500 bits per inch, thereby increasing the storage capacity and transfer rate by 50%.
  4. You could attach as many as 8 FASTRAND cabinets to each controller, and the controller could be connected to two separate I/O channels on the computer, allowing two simultaneous data transfers. In addition, each FASTRAND cabinet could position its read-write head boom independently, notifying the computer when it arrived, so all units could be seeking at the same time.

References

Sperry Rand Corporation, UNIVAC 1110 System Description, UP-7841, 1970.

FASTRAND is a trademark of Sperry Rand Corporation, since merged into Unisys Corporation

 The above article has been reproduced with the permission of the author John Walker of Fourmilab. 

Fastrand II's were used to provide some of the storage on the Univac machines in BP House in the 1970's. They were indeed absolutely massive. My calculations are that one of these provided about one-tenth of a GB of storage. The cost of a single Fastrand II in todays (2011) money would be about £750,000 - just image this the next time you use a 4GB USB stick or put an 16GB SDHC card in your digital camcorder! 


I can recall stories that the drums of the Fastrand II's were made from drainage/sewage pipes in the USA and filled with concrete before being given a coating and built into a "state of the art" device. 


I also recall it was said that if one was ever to come off its axis it would have enough momentum to exit the building through a window (whether open or not!) and travel to the end of The Marlowes (must have been at least half a mile).


Added by David F Thorn on 24 February 2011

Image courtesy of Computer History Museum

Photo of a Fastrand II more clearly showing the drum cylinders.


Added by David F Thorn on 27 February 2011

No comments:

Post a Comment

Note: only a member of this blog may post a comment.

Post a Comment