Talk:Mainframe

When is a computer not a mainframe?

A not too opinionated discussion around various ideas about what constitutes a "real" mainframe, and what doesn't, might be interesting. As an example, in the DEC world, the PDP-10 family of computers are usually regarded as mainframes (probably excepting the DEC-2020). But IBM people seem to regard it as a minicomputer. That is not to say one definition is more correct than any other. The PDP-8 and 11 are usually regarded as squarely in the mini category even though some of the larger PDP-11s could rival a 10 in some respects. The VAX was often referred to as a "super mini"; maybe later larger models nudged into the mainframe category? Larsbrinkhoff (talk) 19:03, 8 April 2021 (CEST)

Hmmm. I'll have to go away and think about what 'mainframe' means. (And the meaning may have changed a bit over time, to make it a bigger problem.) One definition might include physical size ('it fills a whole room'), but that's less relevant today. Another might talk about what it does (e.g. big online system), and how it was run (e.g. secured machine room with operators). Another might talk about computing power ('the biggest thing smaller than a supercomputer'), and internal characteristics (e.g. duplication for reliability, ability to hot-swap, etc).

It can get fuzzy at the edges, though. E.g. hot-swapping - IIRC old Tandem's could do that, but I think of them as super-minis.

I don't thing even large -11's were anything more than super-minis (although a large one, with a bunch of MK11s, might be borderline). Likewise even the larger VAXen - although maybe the VAX 9000 could count. A VAXCluster of VAX 9000's would be a mainframe, to me.

There's no way a big KL10 system was a minicomputer. Jnc (talk) 22:14, 8 April 2021 (CEST)

IBM types clam one defining characteristic is that a mainframe must have subordinate I/O processors, i.e. "channels". As for supercomputer I think you're right about computing power, or specifically floating-poing throughput. Larsbrinkhoff (talk) 06:44, 9 April 2021 (CEST)

Hah. Then by their own definiton, the PDP-10 was a 'mainframe'; they had DF10 channels. If one uses an alternative definition where a 'mainframe' has associated smaller front end machines to do I/O, then again the PDP-10 counts, because of the DL10 and associated PDP-11[s]. (I think there was some earlier PDP-8 equivalent of the DL10, but I forget the details.)

There was; the DA10 (although unlike the DL10, it didn't do DMA, just programmed I/O). I've substantially upgraded the PDP-10 article, to talk about the busses, devices, and front ends. Jnc (talk) 04:09, 10 April 2021 (CEST)

Thanks! Larsbrinkhoff (talk) 09:45, 10 April 2021 (CEST)

I'm not big on using that definition, though, because it's too tied to that generation of technology. (And later PDP-11 I/O devices such as the DEUNA and DELUA and the later multi-serial line cards all had micros on them.) Now that it's 'trivial' to throw a bazillion transistors at things as small as a handheld, we need to look elsewhere than internal organizational stuff - other than things like reliability via replication, maintainability via hot-swapping, etc. I think one/some of the other external attributes might be better.

Oh, I don't think floating point comes into 'mainframe' at all (but I guess you were just talking about supercomputers, there); for 'supercomputer', absolutely, that's key; but for many mainframe uses (e.g. large online transaction systems), they aren't heavy on floating point. Overall computing crunch, yes, but not floating point. Mainframes also had a lot of I/O capability, including a lot of mass storage. Jnc (talk) 13:11, 9 April 2021 (CEST)