[email protected] (Daniel Skwire) writes: > Multiprocessing support earlier than TSO? > > It was before my time, but I read and heard plenty about "MVT/MP65", which > predates TSO's rollout by a couple years, I think. > > MP65 had challenges: 'sympathy sickness" where a CPU problem took down > both CPUs in the complex, kinda sorta anti-redundancy, if you > will. Cured by MVS Alternate CPU recovery. > > Serialization? > Not so good with TEST and SET, I think it was called. Much better with > Compare and Swap, etc.
re: http://www.garlic.com/~lynn/2013o.html#54 Curiosity: TCB mapping macro name - why IKJTCB? lots of problems with 360/65 mp which did have shared memory ... but no shared i/o ... dedicated processor channels simulated multiprocessor i/o by connecting processor-specific channels to different "tails" on multi-tail control unit. also 360/65 os/360 mp support used test&set for very gross level spin-lock (significantly exacerbating any recovery scenarios). at the time, ibm definition of mp was that the system could be (manually) partitioned into two independently running systems (when 3081 came out the new term was dyadic, since it was not possible to run the two processors independently) 360/67 mp not only had shared memory but all processors could access all channels. also mp hardware configuration settings were visible to software in control registers and even some 360/67 were built where RAS software could change the hardware configuration by updating the control register values ... a little more like some of the FAA 360 RAS. 360/67 http://bitsavers.trailing-edge.com/pdf/ibm/360/funcChar/GA27-2719-2_360-67_funcChar.pdf charlie invented compare and swap (mnemonic chosen because CAS are his initials) when he was doing fine-grain multiprocessor locking for cp67 ... some past posts http://www.garlic.com/~lynn/subtopic.html#smp at the science center ... some past posts http://www.garlic.com/~lynn/subtopic.html#545tech initial attempts to get compare&swap included in 370 were rebuffed, the 370 architecture owners saying that the POK favorite son operating system people claimed that test&set was more than sufficient. the 370 architecture owners said to get it included, needed justification based on uses other than multiprocessor locking. thus was born the examples for multiprogramming/multithreading serialization by large applications that might be enabled for interrupts (useful whether or not application was running in multiprocessor environment) ... examples still found in principles of operation http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/DZ9ZR003/A.6?SHELF=DZ9ZBK03&DT=20040504121320 -- virtualization experience starting Jan1968, online at home since Mar1970 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
