Thanks for that insight. Whilst it is possible to use smpsus in a parallel current sharing mode to create a monster PSU from a bunch of smaller units, I suspect that it's probably a practice best left to the specialist PSU manufacturers to create monster PSUs from arrays of smaller units kept safely hidden away within the confines of a monster "Black Box" of PSU magic rather than allow contractors to assemble and set up such monstrosities on site.
TBH, I think most modern "Main Frames" are implemented out of a grid connected array of smaller computing units, each with their own independent PSU (probably a switching converter fed off a medium voltage DC bus) where the loss of a computing unit out of a modestly sized group of, say 16, simply results in a slight degradation of performance rather than a total outage. The need to provide redundency for each individual computer unit's psu then becomes, well... redundent.
The whole point of redundnecy isn't primarily PSU availability but computing power availability and a computer system with its performance degraded to a mere 25%, let's say, is infinitely preferrable to one that's gone totally off line.
A modern "Mainframe" based on grid computing technology considerably eases the provision of truly independent PSU redundency, even to the point where the more paranoid practice of using two or more independent connections to the national grid to maintain power to the data centre becomes "A Walk in the Park" exercise to implement compared to trying to maintain a common DC bus feeding power to a single high spec computer.
I suspect the only situation where such an array of cheap smpsus have been paralelled in this fashion will be in the workshops of amateur experimentalists with a "Mad Scientist" complex and a temptingly large collection of smpsus to 'burn up' in a FrankenPSU project of their own devising.
There might even be videos on YouTube. There seems to be no shortage of such videos on other, more bizarre, projects, so why not? :-)
Really? Often in those days the power demand was so huge that there already were multiple power supplies in a single cabinet, and with a reasonable design they could just as well be made redundant.
E.g. on the Burroughs 7800 mainframe that I once worked on, the CPU cabinet has 40 paralleled power supplies. Total power consumption of the processor was about 5000 A at 4.8 and -2 V. See:
The current IBM Z series is certainly organised as a rack of heterogenous CPUs connected to disk arrays judging from what I've read, though I'm uncertain how it applies to the others such as the Honeywell ClearPath systems. But, I don't know whether IBM system boxes are any more likely to have dual power supplies than minicomputers or server racks. Historically mainframe PSUs seemed to be pretty much bullet proof: put it this way, the other stuff in a mainframe box needed TLC more often than the PSU did.
Apart from the S/88, the only IBM boxes I've used were System/38 and AS/400 (now called the i-Series) machines. These had no multiple redundancies anywhere in the box apart from all disks being part of one or more hot-swappable RAID 5 clusters. In these systems all disks and RAM are components of a single virtual memory pool, so the machines make no distinction between the filing system and the virtual memory page swap area. Indeed, persistent storage is best considered to be an in-memory structure. The only disc access operations it needs are page reads and dirty page writes, with the location each page is stored in being determined by disk load-balancing requirements.
Indeed, and using dual PSUs, UPS, etc are rather a special case because they can provide redundancy and resilience in ways that have little or no impact on the rest of the computer.
True fault tolerance, which is the next stage, requires not only that every component in the system must be at least duplicated, but in addition that all components must be hot-swappable and either the hardware or software must be capable of recognising faults as they occur and dealing with them, i.e. the faulty component must be disabled before its output affects correct operation of the rest of the system and the appropriate people notified of problem. The computer must be capable of operating without the failed component until its replacement arrives on site and the failed component has been replaced. At that point the computer must recognise that the component has been replaced, test it and, if it passes, fire it up and put it back into normal operation.
On top of that, its highly desirable that the fault tolerant system can be geographically dispersed so that it can continue operation despite physical problems, e.g. network breaks, power failures, fires & hurricanes.
The Tandem/HP NonStop machines support all of the above, though I've seen the failing member of a Tandem's dual PSU blow the computer room breaker switches during a weekend, with the result that the Tandem and a Stratus in the same computer room both stopped about three hours later when the batteries in their PSUs were drained.
That said, the Tandem/HP NonStop boxes consist of up to 16 networked units, each containing multiply redundant components. As each component can be paired with its logical backup, which may be any of the boxes on the network, the complete system can be physically spread round the world for geographic dispersal. The Stratus is only redundant within a single cabinet, so can't be geographically dispersed.
BTW, both NonStop and Stratus systems were available in the 1980s, so they aren't exactly recent technology, though the underlying hardware has obviously changed a lot since then.
--
martin@ | Martin Gregorie
gregorie. | Essex, UK
org |
Yes, really. The biggest 1900 I used was a 1904S. I goodly amount if the space in the base of its CPU cabinet was PSU (40kW nominal IIRC) but AFAIK that was a single monolithic PSU. The separate disc controllers and
4-way tape drives had their own PSUs and, again IIRC were 20kW nominal).
Different manufacturers took different approaches.
--
martin@ | Martin Gregorie
gregorie. | Essex, UK
org |
Which all depends on the efficiency of the speakers of course.
I remember years ago feeding the output of a small battery portable radio into a Wharfedale RS12DD in a 3.5 cu ft cabinet and finding it surprisingly loud whereas I have come across modern speakers requiring
10's of watts before you could even hear a whisper out of them.
Horn-based loudspeakers are among the most efficient. Fed by an early 1970s-era transistor radio, a home-built cabinet of the old Altec A-7 Voice of the Theater design made a very adequate music-listening SPL in a fairly large bedroom. I'm not sure I ever found out how loud they could go with a 30-50W amplifier, though they could extinguish a small candle placed an inch or two in front of the woofer cone.
--
Robert Riches
spamtrap42@jacob21819.net
(Yes, that is one of my email addresses.)
I did sort of qualify it. B-) It's all a cost benefit thing hinging on what is termed "decent". Working yesterday with a system that was peaking to 105 dB(C) SPL at about 50 m outside. Pretty "decent" though the bottom end was tending fart rather reproduce the notes at that level. Only four little stacks spread along 100 m or so. By little I do mean little, 2 x 350w x 700h x 500d mm units for the bass with three 150w x 600h x 300d on top for MF/HF. Probably a bit overkill for the living room and a Raspberry though...
I have trouble combining 5 V power rail, decent and PC speaker (aka cheap) into anything that makes sense. It's pretty much a "choose any two" combination. B-)
Agreed. Look for some sort of boombox with mains power and a line-in socket, since that seems to be what the 3.5mm audio out is designed to drive. This is not necessarily a cheap option, though it could be an excuse to buy a Bose Wave, should you want one, or you might get lucky on eBay and find a scruffy one going cheap.
--
martin@ | Martin Gregorie
gregorie. | Essex, UK
org |
So would a Class D amp (there are chips for this) plus some bit bashing in a Pi do a better job? Looking at the pricing of commercial audio components convinces me that it's like the price of caviar - way higher than it should be. I just wish I wouldn't have to use a PCB or wire-wrap to build such a thing; that's the kind of project I never complete unless someone twists my arm. (Yeah, I know full well that there's no such thing as a free lunch, and that sometimes I have to settle for MacDonald's).
--
Windmill, TiltNot@NoneHome.com Use t m i l l
J.R.R. Tolkien:- @ S c o t s h o m e . c o m
All that is gold does not glister / Not all who wander are lost
There's also the fact that a Class D amplifier could be designed to run a 100v. speaker system from a 5 volt supply (given enough power). It could be a 'boost' type design rather than a 'buck' design.
--
Windmill, TiltNot@NoneHome.com Use t m i l l
J.R.R. Tolkien:- @ S c o t s h o m e . c o m
All that is gold does not glister / Not all who wander are lost
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.