[cctalk] Re: Drum memory on pdp11's? Wikipedia thinks so....

Bill wrote: Remember that there was a very active third-party peripheral device market for the PDP-11 series. Could be that someone offered a head-per-track drum store that the person that made the update to Wikipedia may have spied. On a somewhat related note, many years ago, sometime in the late 1970's to perhaps no later than '81, I toured a local (Portland, OR) timesharing service that was called "Information Sciences, Inc.". They had their datacenter in a building in downtown Portland. They offered timeshare services on a well-built-out and heavily-modified DEC PDP-10 (KA-10). The machine had a bunch of add-on solid-state memory boxes, as well as a real drum memory. The drum memory unit was quite large, a box about 5 feet wide, 4 feet tall, and probably about 3-4 feet deep. The controller for the drum was another box that was about the same size as one of the memory boxes, and was cool because it had a bank of lamps at the top of the cabinet that showed the activity of the drum. It was flickering like crazy, as it was during prime-time that I was there, and they had about 180 simultaneous users on the system at the time. I was told that the drum was a head-per-track unit, and revolved at 8,000 RPM. The KA-10 had been modified to support virtual memory using a pager box of some sort, and the drum was used to support high-speed paging and swapping. The drum was kind of ominous, as you could sense the kinetic energy that was just waiting to be released if a bearing failed catastrophically. It had this low frequency resonance that felt as if it permeated the entire data center. The guy that took me on the tour said that the wall behind the drum had to be specially reinforced as if the drum exited the reinforced cabinet due to some kind of failure while at speed, it would have gone through any conventional wall like it was made of paper, and another wall which was the side of the building, and would have fallen 6 floors to the ground below, which obviously would have been disastrous. Apparently if there was a failure, due to the direction the drum rotated it'd come out the back of the cabinet rather than the front. I was also told that the drum cabinet had special mounting that was a large structure of steel beams in the mezzanine level beneath the datacenter that connected the mounts to the main support beams for the building, because the gyroscopic effects of the drum would have torn out anything else. The mounts had to be inspected every six months to look for cracks or any other sign of stress-induced problems. I don't know who made the drum unit, as I didn't see any tag on it, but the cabinetry was colored a significantly darker shade of blue than the DEC boxes were, which tells me that it was probably a third-party device. I was also told that the drum had its own UPS and genset, because it would be "bad" if it ever shut down in an unplanned way. Apparently there was a procedure that involved slowing down the drum revolution in a stepwise fashion. This apparently took something like 2 1/2 hours to perform, and was orchestrated by the controller. I never heard what "bad" meant in the context of the drum powering down in an unplanned fashion, but the tone of voice of the guy explaining it made it clear that it was something to be avoided at all cost. This was the first time that I'd ever seen a PDP-10, notably a KA-10, and I absolutely fell in love with the console of the machine. It's my favorite classic blinkenlights console. -Rick