[cctalk] Re: Odd IBM mass storage systems

The main storage area of the ElectroData/Burroughs Datatron 205 was 20 tracks of 200 words each for a total of 4000 words. The drum rotated at 3570 RPM, so the average access time was about 8.4ms. The four quick-access tracks (or "loops" as they were called) were 20 words each and worked as a delay line, much like the Bendix G-15's drum. These tracks had separate read and write heads. When writing was not taking place, the digits from the read head were simply copied to the write head 36 degrees behind on the surface of the drum. So as the last digit of the track was written, the first digit of the track was coming under the read head, yielding an average access time of 0.84ms. When writing, digits from the processor were shunted to the write head in lieu of those coming from the read head. When power was removed from the system, the 4000-word main memory was preserved, but the data in the high-speed bands was lost. The 205 had instructions to transfer 20-word blocks between the main memory and the high-speed tracks. There were even a couple of instructions to move a block to one of the high-speed tracks and branch to a word in that high-speed track. The high-speed tracks were addressed modulo 20 (i.e., word 4005 was the same as 4025, 4045, 4065, ... 4985). You had to get good at dealing with addresses that were congruent modulo 20. The high-speed tracks were so much faster than main storage that most programmers went to a lot of effort to "block" 20-word segments of the program to the high-speed tracks and execute at least the most active code from the faster storage.