What 20th-century computers besides KDF9 replicated hardware for each concurrently-running program?

Question

The KDF9 (English Electric, 1963) had a hardware option for timesharing. This provided four instances of most per-program context: the nest (16-deep expression evaluation stack), the subroutine jump nest (16-deep return address stack), and the 16 Q-stores (modifier/counter registers).

The timesharing director (=operating system kernel) therefore supported up to 4, no more, concurrently executing programs. Context switching avoided the need to save and restore 48 registers. So, context switching was relatively cheap for such a sophisticated machine (unless you were paying to buy the hardware). Each of the 4 possible programs had an associated priority, and completion of I/O by the hardware was cognizant of whether the completed I/O belonged to a higher priority level than the currently-executing program.

I don't think I've ever heard of any other system that provided this level of timesharing support in hardware. The ICL 1900 system sort-of comes close, in that the resident exec had a small and fixed limit on multiprogramming, but I think that's due to exec storage limits (tables take space).

These days, of course, 4 processes is indistinguishable from zero, but 4 wasn't necessarily a huge barrier in the 1960s (KDF9 had 32Kwords memory, max, so you're not fitting too much in there).

Anyway: what other systems had hardware per-program resource replication?

(Apologies for another list-oriented question, but quite frankly, the discussions that ensue from such forms is more interesting than the "here's a question, here's the answer" format. It's in the nature of historical information that people know different corners of the subject.)

---

Wikipedia's page on the Ferranti Orion lists a handful of computers, including the Orion, that had 'hardware support for timesharing', but it's not clear to me whether that included replicated hardware. Maybe there was nothing to replicate -- the Orion accumulators were just the first 64 words of (relocated per program) address space. The minimal requirement for hardware-assistend timesharing support may be limited to provision of datum/limit registers.

Answer 1

The design you have described is very similar to Barrel processor where CPU contexts are duplicated and switched at each clock cycle.

A barrel processor is a CPU that switches between threads of execution on every cycle. This CPU design technique is also known as "interleaved" or "fine-grained" temporal multithreading. Unlike simultaneous multithreading in modern superscalar architectures, it generally does not allow execution of multiple instructions in one cycle.

Like preemptive multitasking, each thread of execution is assigned its own program counter and other hardware registers (each thread's architectural state). A barrel processor can guarantee that each thread will execute one instruction every n cycles, unlike a preemptive multitasking machine, that typically runs one thread of execution for tens of millions of cycles, while all other threads wait their turn.

Notable examples include:

• CDC_6000_series
• UltraSPARC T1
• XMOS XCore
• Most(All?) modern GPUs

Answer 2

The Honeywell 800 had a multithreading CPU with hardware registers for 8 concurrent threads. See here, an article for a brief description.

Basically the CPU itself implemented a round-robin scheduler - first it would run a single instruction from "thread 1" out of the corresponding register set 1, next it would run a single instruction from "thread 2" out of the corresponding register set 2. (Looking now at the previous answer this seems to be an example of a "barrel processor" but I don't know for sure.)

Notice that this computer was first available in 1960 - and this article claims 89 were delivered. I know where one of them was: When I worked for Honeywell's Marine Systems Division, West Covina, CA while still in college an H-800 was installed there. It did the payroll for the Marine Systems Division (I think there were two locations for that business unit). No disk: Only a bank of spinning tapes. 6 or 8 of them? Anyway, the thing to note was this was in 1977-1978! Thing had been obsolete for years yet was still running a production job for them - it's how we got paid!

(This extended retention of working assets apparently was not unusual for Honeywell. My desk at work was this large slate gray steel thing that a tank could have run over and left completely undamaged. It had a metal property tag riveted to the front. It said: "Minneapolis Regulator Company". Honeywell stopped being called the "Minneapolis Regulator Company" long before World War II ...)