sjs@spectral.ctt.bellcore.com (Stan Switzer) (04/21/89)
In article <530@aurora.AthabascaU.CA> rwa@cs.AthabascaU.CA (Ross Alexander) writes: > My understanding of cores is that as one runs them faster, they get > hotter. Indeed, I've seen reports in this group to the effect that on > some (faster) machines, if you knew the stack geometry and were a tad > antisocial, you could damage core planes. Ah, the good old days.... As I understood it, each access to a core heated it up a bit. The trick is to make sure that the core dissipates heat faster than it accumulates it--at some point before the core is damaged. The Honeywell 716 minicomputer (I believe these were once used as ARPA IMPs, but we used one as an RJE station) was designed so that memory would cool fast enough that you could access a core every other cycle without damaging it. The machine manuals were very explicit in their warnings that getting the machine into a one-instruction loop would fry the instructions core plane (as would an indirect cell pointing to itself). One day the system just locked up with the front panel lights on solid. Now, on the 716 the front panel displayed the memory address register and the memory data register. I was on top of that halt switch in no time flat. After taking a system dump, I breathed a sigh of relief when the memory diagnostics found no problems. It took me several days to realize what had happened. The machine had gotten into a 2-cycle indirection loop popping back and forth between the accumulator (address 0) and a memory cell. This is as close as we ever came to blowing out a core (something of a miracle, considering the amount of work we did on that system). Anyway, I just smile whenever people say that you can't hurt the computer with software. Obviously, these people never heard of the Honeywell 716. Stan Switzer sjs@ctt.bellcore.com
jbuck@epimass.EPI.COM (Joe Buck) (04/22/89)
In article <15465@bellcore.bellcore.com> sjs@ctt.bellcore.com (Stan Switzer) writes: >The Honeywell 716 minicomputer (I believe these were once used as ARPA >IMPs, but we used one as an RJE station) was designed so that memory >would cool fast enough that you could access a core every other cycle >without damaging it. The machine manuals were very explicit in their >warnings that getting the machine into a one-instruction loop would >fry the instructions core plane (as would an indirect cell pointing to >itself). Ah, the infamous HCF (halt and catch fire) instruction! :-) -- -- Joe Buck jbuck@epimass.epi.com, uunet!epimass.epi.com!jbuck