Posts by Peter Gathercole 4859 publicly visible posts • joined 15 Jun 2007
IMHO, a 'Service Charge' should be part of the meal. A tip should be discretionary, based on how well you were served. Whether a tip should count as an allowable expense is a completely different question.
I feel it's the same problem that is creeping in with US style reviews on service when anything less than 100%, 10/10 or similar is regarded as a negative, whereas I want to be able to indicate when it was outstanding and exceptional, not just ordinary.
Whilst I agree about not allowing the most valuable assets to be sold to foreign companies, this strategy also has to be aligned with an investment policy to allow British companies to obtain the resources they need to grow to compete in a world market.
We're great at innovating, but not so good at turning that innovation into a viable business as a result of a lack of investment options. Hence, British companies with valuable IP that don't grow fast enough become targets for buyouts. If you just block the purchases, then they are likely to wither and die as other companies outside of Britain decide to re-invent the concepts and then out-compete the one that did the original innovation.
OK Patent law should prevent out-and-out copying, but even defending patents is expensive. The owners of a company, when faced with defending a patent may just decide to sell out, take the money and run to the next opportunity.
Not just mains. Charged electrolytic capacitors, especially big ones can still give you a belt, even after mains is removed. If there is sharp metal around, the involuntary reaction tends to cause you to gash your hand as well.
And HT from a tube TV or monitor can be even worse.
Still bear the scars of both of these when learning this from 50 years ago. HT burns don't heal well. Faint, but still there.
Aren't they already a way down this route?
Once they differentiate between the cloud and perpetual licenses (probably by rolling out regressive patches to the on-prem versions of Office), they'll just stop selling them, killing the practice of selling surplus older licences.
What they may be most worried about is the recent rulings that allow some customers to move their existing on-prem licences into the cloud for reduced or no cost.
Ah. I do remember that some tape players had low voltage inputs. I've never really seen them in use much, except for maybe in a car. But if you had mains input, that is generally what I saw being used.
I don't think I've ever seen a tape player that could run on mains being shipped with a power brick.
Unfortunately, all of the big chain charity shops now cherry-pick their donated LPs before they get to the shop floor. They have lists of artists that they know to pick out. These then appear on places like ebay, where quite a lot of charity shops do business. This behaviour coincides with the vinyl renascence.
I used to pick up LP mix bundles of 10 or more online, and used to regularly find interesting disks, but recently, not so much (and I get left with a lot of dross, does anybody want to buy some Val Doonican or James last LPs, or MFP compilations? I have plenty.)
What you can still find is niche artists (like your Bonzo Dog Doo Dah band LP - I found their "Best Of..." a while back) which are not on their list, and slip through. Can be good if you are looking for oddities, but not so much otherwise.
Sometimes junk (rather than charity) shops are a better bet to find LPs nowadays.
That was normally either because of a bad/dirty tape movement, which sometimes wrapped the tape around either the capstan or the pinch roller, or a toddler in the back seat getting their hands on a casette and trying to see how long the tape was!
I lost several of the tapes in my collection because of this (mostly children, I used to clean the tape players quite frequently).
Not at all. Get a PC with a line (not microphone) in connector (which most desktop PCs used to have, although you may have to look inside and connect it up yourself on some more modern systems), connect this to some analogue signal, and load up Audacity. Once you have a digital copy on disk, work out how to convert it to MP3 or how to write it to a CD.
The problem is that most laptops only have a microphone in, and more recent ones expect digital microphones connected by USB.
But you can get USB audio converters that provide line level inputs for a few 10's of pounds if you look around. Also, nowadays, many low end (and some not so low end) turntables (if you're wanting to rip LPs) have built-in A-D converters that attach to a computer by USB.
20 years ago, I ripped quite a lot of LPs using the built-in sound card of an ordinary (original) Celeron PC, although I used a commercial Windows product to do it (the PC was my wife's, and it was conveniently located across the room from my HiFi, and the software did automatic track recognition and allowed you to name the tracks). More recently, I use Audacity and one of the aforementioned USB sound devices attached to my laptop running Linux to rip some 50's swing LPs I had knocking around, for my son who likes swing (quite why, I don't know!)
There was a strange phenomena with choosing the cassette player to use on a computer. The thing was that more expensive did not always mean better.
Rule 1. Only use a mono cassette recorder. If you had to use a stereo one, only use one channel, and you may find exchanging tapes with other people difficult.
Rule 2. Get one with a pause button
Rule 3. If using a BBC Micro (or some others that copied it), get one with the microphone start/stop 2.5mm jack
Rule 4. Cue and Review capability was very useful
Rule 5. Get one that allows easy access to adjust the azimuth of the record/playback head
Rule 6. Speed stability and distortion (like clipping), and mains hum will cause you problems, so get one that sounds OK when playing music
Additional factors
Automatic gain control (auto recording level) can be either a good or bad thing. When implemented properly, it can be good. If you can turn AGC off, and adjust the recording level manually, that's even better.
If you can find one that allows you to listen to the signal while playing back, or when recording, that is tremendously helpful.
I bought a reasonable quality Panasonic slim casette recorder back in around 1982. It was around 15 quid, more expensive than most cheap ones, but worth the extra money. It didn't have the 2.5mm jack, so I added one into the power line to the motor. I also drilled a hole to get access to the screw that adjusted the azimuth while playing. I used this very successfully with ZX81, Spectrum and BBC Micro, and it's still knocking around somewhere even now. The WH Smiths Data Recorder was a very safe purchase, having almost everything described above.
I've mentioned azimuth several times. If you couldn't read a tape recorded on someone else's recorder, adjust the azimuth until it sounds as bright as you can, and try again. It normally helped (if you had a tone control as well, turning up to it's highest treble setting could also help). If you changed it, reset it in a similar fashion while listening to a tape recorded on it before you fiddled. If you screw it up, adjust against a commercially created tape. These were normally OK when it came to azimuth.
Also, the moon is a vacuum, whereas Mars has at least some atmosphere. Probably makes a difference to any lubricants.
The type of dust that the rovers need to cope with will also be a lot different. Moon dust is generally more abrasive and settles differently when kicked up, as there is no atmosphere to wear it down or carry/scatter it.
Plus, gravity is stronger on Mars.
Generally speaking, prey animals have their eyes on either side of their heads. The example used in my biology text book at school was rabbits, who can see not far off a full 360 degrees without moving their heads, but have almost no stereo vision except straight ahead, and not for much of that. Useful to spot predators around you.
Predators have their eyes on the front of the head, as being able to judge distance through stereo vision is very useful when trying to catch your prey. But they can't see that much from either side without turning their heads, and nothing from behind. This is taken to extreme in some raptors and other hunting birds, who have their eyes locked into fixed directions relative to their heads for maximum stereo vision (ever wondered why some owls can twist their heads almost fully around? This is the reason).
Horses are a bit in-between. They are mostly prey animals, and their eyes are towards the sides of their heads but not so much as rabbits. They do have a degree of stereo vision as a result. But very often, they will look at you from one side of their head or the other (this can be very disconcerting when a one-eyed pony I know looks right at you with their missing eye, not sure what is happening then, but I suspect it's listening rather than looking). They do also have a large degree of surround sight, which is why working horses used to wear blinkers to stop them reacting to what was going on around them.
The existing master socket is where I want it. But as I said, at least 5 90 degree bends to follow the existing route of the copper cable.
I've never talked directly to an installation engineer, but from what I am told by other people who have been through the installation, where they can't run a fibre to the existing master socket, they'll suggest a different place to install the ONT and router, and this is the main issue. There really isn't a good place to put it, because there's no power anywhere on the wall where the fibre would be run to.
I suspect strongly that they would cancel the install, which is why I want a survey.
I have a couple of problems with the way that FTTP is rolled out where I live.
Where we are, all of the copper connections are over ground, on poles. No real problem there, because they are putting fibre over overhead cables.
Only problem with this is that currently, our copper line runs across the back gardens of 4 other properties between us and the pole we're connected to. This is not the closest pole, but the closest one has trees and very (very!) tall shrubs between us and the pole.
There should be no problem with this, apart from the issues of running overhead cables over several properties, but I guess that they would use the existing copper wire to guide the fibre.
But the next problem is that the back wall of our house, which is where the existing copper line is connected to has absolutely no power on it (house built before 1920, power mainly constrained to the middle of the house when last re-wired). Currently, the copper wire runs to the back wall, down the wall, around the kitchen (not an extension, built like that originally) involving going up and down the apex of the wall, round a corner and down the side of the house, where it enters at a window, and where there is power. Something like 5 90 degree bends. Openreach won't do that. Too many bends.
I want to get a survey done suggesting the best way of overcoming problems. Openreach won't do that either, even if I offer to pay for a call-out.
I've heard too many stories about people having an FTTP install scheduled, the engineer turning up and saying they can't do it, but for the FTTC service to be turned off anyway and for it to remain off for days/weeks while the problem is sorted out. My neighbour is currently fighting that one.
I work from home because of a medical condition. I need a broadband service to work.
Plus, of course, my current ISP does not have any plans with an IP phone service included (I'm happy with mobile, my wife isn't), which means I'm trying to negotiate with a new ISP as well as trying to work out how the service would be installed anyway.
I talk to people and watch videos where the fibre comes to the front of the house underground, through the front wall, and can then run straight to a suitable power socket. It seems so simple for these installs. I envy them their hassle free switch-over.
I've been told that existing FTTC installations will not be pressured quite so hard as people who still have analogue lines to the exchange, which is really what Openreach are trying to eliminate.
It's a while since I read the T&C's on a Microsoft EULA, but I'm certain that they used to explicitly exclude using the licence shipped with a system in a VM (I'm sure it's in the Windows 10 EULA, but I believe it was also in the Windows 7 EULA).
IIRC, the way that they did this was by making the only version of Windows that you could officially run in a VM the most expensive tier of a WIndows retail licence, not any of the OEM ones shipped with a PC.
So it was always excluded by licence, they're just enforcing it by key management. If you manage to get it authenticated, then you've been lucky. They did make a previous attempt even as early as keys being printed on labels attached to the system by checking the licence key against some magic number in the BIOS tied to a manufacturer, such that if you tried using a vendor-specific key on a system or VM that did not have this identifier, the it would not be accepted.
Some gulls are very large and quite heavy, and within a few miles of the coast can often be seen around roadkill.
I once hit a pigeon full on with a car, and the explosion of feathers that ensued scared me half to death.
Mind you, the worst experience I had was hitting a swarm of bees while driving on a motorway. Completely covered the windscreen with gooey sticky mess that the wipers wouldn't clear. Was the only time I stopped on the hard sholder when I hadn't broken down.
I was working on a contract, living about 90 minutes away from the office (and commuting daily!) with no remote access, so I didn't really want to be on the call-out rota. One week there was nobody else available, so I reluctantly agreed to do it for just that week.
On the way back home, I had the radio on, knowing that I would recognise the beep of the pager. I heard nothing.
Next day, my name was mud. 20 minutes after I left, someone paged me to get me back before I had gone too far. Looking at the pager, it had gone off, but I hadn't heard it. Thinking the pager might be faulty, we had it paged. Imagine my surprise when instead of the recognisable, and difficult to ignore beeping, we got a tune played by the pager! Turns out that the previous person with the pager had changed the alert from the default to the tune, which had disappeared into the music playing on the radio. I was still blamed, but some of the fault was deflected to the person changing the alert.
Management added an instruction to the standby procedures for the person taking the pager to do a test page to check the audible alert, and also that the pager was working. Fortunately, I was never put on the rota again at this contract, and I had a mobile phone for all the roles afterwards, not that I did much standby.
Yes again.
But as an aside POWER (original RIOS systems) != Power (as in Power 10 et. al.) as well! There's been so many changes in architecture and branding across the years that sometimes one makes mistakes.
I remember the confusion when the the single-ship RSC POWER implementation, as in the 220 and C10 were launched, and people called them PowerPC by mistake.
They were discontinued long before 64 bit chips. IIRC they had Power 603e processors in them, and the latest version of AIX that would run was AIX 4.1.5.
At the AIX Systems Support Centre in the mid-80s, we had a number of Thinkpad Power Series 850 systems that went home with the out-of-hours support people so that they had access to AIX documentation on CD (this was before IBM put the documentation on-line, and the format of the documents on CD only had an AIX reader), They were quite good for the time, although they were much larger and heavier than their Intel counterparts. I remember getting about 2.5-3 hours of use out of the internal battery. The one thing I did find was that the PCMCIA modems that IBM sold used to fail very quickly, so we had to resort to external modems to dial back into the centre's computers to get access to other resources.
For S and L models, the clue is in the name. The last digit is the number of U's of rack space it occupies. The one before that indicates the number of sockets, and anything before that is the power generation. The letter in the front indicates Scale-Out (S), Linux (L) or Enterprise (E). The numbers for the E systems do not indicate the number of sockets or rack space, but align to older system sizes such as E1050 and E1080.
Scale-Out systems have significant CPU and memory resource, but limited I/O slots for adapters. They are intended to be used where a service scales sideways over many systems, such as Cloud type deployments. The Enterprise systems have more I/O slots and can have more memory and CPUs, and the largest can be built up in a modular fashion to create very much larger single systems.
So S1112 is a Scale-Out Power 11 system with one socket in a 2U enclosure.
Generally speaking there are equivalent S and L models, with different firmware which dictates how you control and configure virtual machines and resources. The L models are a little cheaper, but significantly limit the number of AIX and IBM i partitions you can run.
In my first job after university, a hateful one writing RPG/2 in a batch environment (after using PL/1, APL and C at university) I had a problem with the behaviour of some code that I just cold not tie down.
Although I can't remember what the code did, or exactly what the symptoms was, I was embarrassed to have it pointed out (by the most prudish of old-school data analysts who regarded any other languages than RPG and COBOL as being abominations) that I had a letter O rather than a numeral 0 in one of the character field tests. I blame the poor font and a fabric ink ribbon that needed changing on the band printer that printed out the diagnostics for the lack of visibility of the problem.
Did not stick that job for more than a year.
I have a two sided view of this.
I absolutly agree that current LLM "AI" is not artificual intelligence. They're statistical models that select outcomes based on their training data.
However...
How do we, as humans, learn? At the heart of it, we're memory machines that hoover up experience and information as we grow, which conditions how we act, and then at some point we are judged to be "intelligent" (at least some of us are, I'm not so sure about some MAGA supporters!)
Our memories remember by association. Inputs trigger responses. If you think about it, very little of our lives are not triggered by what is going on around us and our previously history. Even creative types are conditioned by their surroundings and experiences (I've just re-acquainted myself with the story of 'He's so Fine", and "My Sweet Lord", which was ruled on in court as being inadvertent plagiarism triggered from previous memories, which I think sort of proves my point).
Maybe the inference and association systems we have in our wet-ware is more complicated than any current LLM, and able to take in more nuanced and wide ranging sets of inputs to apply to our learned experiences to control our behaviour.
The only thing that I can think of that is different between us and LLMs is that we generally have aversion pathways in our learning experiences. We learn that some things that we do result in pain, punishment or other unwelcome results, and I don't think that LLMs as they currently are have this learning mode. I *think* that this is what generates our morals, in that we learn that some things we do hurt ourselves, and some hurt those around us. To give this to a LLM, we have to make them aware that some things that they produce will result in some undesirable, possibly existential effects, although current'y I can't think of what we could use to do that in something that is not self-aware, and able to understand what being turned off would mean.
Without some form of identifying results that generate harm, I'm afraid that we could end up with entities that have all the characteristics of a sociopath, that makes suggestions regardless of the results on the subjects!
That 4 million lines of code probably includes all of the drivers for the various display adapters it supports (and over time, that has become a lot).
It may also include some of the toolkits that shipped as part of the history of X11. I'd have to see the context of that number.
The actual display and communication code is probably only a fraction of the total.
Ah, that's after I dabbled on COBOL. It did not have pointer types then, and it appears that the current pointer usage was mostly standardised in COBOL-2002, a long time after I stopped. I do remember unions, but I thought that the data structures had to be the same size, checked at compile time. Could cause data corruption, possibly, but unlikely to cause dangerous memory use.
Maybe a case of modernising a language, but making it less safe!
I will qualify what I said. Legacy COBOL is inherently memory safe.
On the subject of unions, C++ and C both have a union type, and have had pretty much since the earliest days of C. Don't know about Java, though.
Changing the subject, I used to have access to a strict standards Pascal compiler. People using it hated it, because it imposed such strong typing, that it was difficult to write anything remotely complex. Many Pascals relaxed this with extensions that made Pascal easier to use (particularly pointer and casting type extensions), but much less safe. People wanted ease of use over strictness and safety.
Several generations of supercomputer ago, the Met Office was having their old NEC systems removed down a passage way with a suspended floor. I do not know the full details, but as I understand it, one of the posts holding the floor moved, and the tile carrying the weight of one corner of the rack tipped, causing the rack to fall over until it hit the wall, blocking the corridor.
There was someone who was caught by the frame falling over. They ended up in hospital, although I understand they made a full recovery.
It was after this that the Met Office and IBM decided to use load-spreaders when moving their kit down this corridor.
I also herd a story from my then manager who had been an IBM Customer Engineer (CE) about a machine room where one complete row of tiles was removed, against all good practice, and the entire suspended floor moved and fell in one piece. No idea who the customer was, or when this happened, but it would probably have been in the '70's sometime.
For IBM RS/6000 SP/2 systems, that were pretty tall, there was actually a special wheeled tool/cart called a rack tilter that could be provided to gently tip a rack over until it would fit through a low door. As I understand it, there was just one of these in the UK, and it's availability had to be checked and booked well in advance of the delivery in order to make sure that it would be available.
I heard a story about an SP/2 delivery to Exeter University, which is/was built on a hill side, where the approach to the loading bay for the datacentre was on a slope, and they could not get the angle right to unload the SP/2 frames from the lorry onto the loading bay without fouling the top of the loading bay door!
I assure you that the early generations of PDP-11 and VAX-11 were not single board computers. The CPU itself in both was split over several boards in both families, and this was augmented with memory and I/O cards. If I remember correctly, the board set for the basic VAX-11/780 was 13 cards.
It was the F11 and J11 single-chip PDP-11s that started the single board systems, and then 78032 and 78132 MicroVAX processors that did the same for VAX, although both had memory and I/O controllers on separate cards for most of the systems they were built into.
The distinction between micro-, mini- and mainframe nowadays has less to do with the CPU and it's implementation, and much more to do with the way that the systems are packaged together.
Stepping away from IBM systems for a minute, if you look at the NEC and Fujitsu ideas of mainframe systems (yes, they're still being made), you will find that most (all?) of them have parallel x86-64 processors with their mainframe architecture implemented through emulation. But it is the size and scale of them that make them 'mainframes', not the CPU type.
It used to be that mainframe systems were generally multi-rack systems, with I/O processors and storage in different racks than the CPUs, but this is no longer the case, with I/O for mainframe being integrated first into the rack with the CPU, then onto the CPU itself.
Looking at it from the other end, micro- and mini- systems often (although not exclusively) had everything in a single case/rack, but now scale out to have their storage separate. The two ends have converged, and this is true across the board, with I/O adapters and protocols becoming common across the board.
In my mind, the name 'mainframe' ought to be retired, and systems should be referred to by their architecture. IBM itself tried to do this when they invented the terms zSeries, pSeries, iSeries and xSeries for their platforms in the late '90s, followed by their re-branding System Z et. al, and now it seems IBM Z etc.
COBOL is a well-specified, mature language with phenomenal backward compatibility, and is inherently memory-safe, as it always uses bounded integers for all of it's array and character indexes, and has done almost since it's invention.
But it's not flavour-of-the-month, and inherits significant un-coolness from the fact that it's a language associated with mainframes, which must-be-bad, according to 30 years of 'modern' thinking.
In a time when mainframes were considered expensive, this un-coolness was actually an aspect of companies and their new-computing-age architects thinking they could save money by moving first to upsized mini-computers, and then to supersized microcomputers. If it can be shown that mainframe is now cost-competitive again, I think mainframes may have a resurgence, and with them, COBOL.
What was expensive with mainframes was their RAS and reliability features. As these features have crept in to modern massive hyper-converged "mainframe-class" x86-64 servers, so the cost has gone up. And this is also true when it comes to the software costs, as shown by Broadcom trying to monetise VMware.
I suspect that someone may also suggest moving back to smaller, non-virtualized discrete systems sometime in the future!
I've seen several mainframe installs over more than 30 years. They've never used forklifts.
The 30" racks that were used around 20 years ago for z13, and which were also used with 9125-F2C supercomputer systems as well as both earlier and later mainframe models had very big wheels (for a rack), and rolled quite easily. There was apparently a battery powered tractor unit to tow the racks in place, but I never saw one actually used. For some models, suspended floors needed to be covered by large wood and duraluminum sheets to spread the load across more supports as the populated racks were heavy. These racks were so tall that adding any hight over the basic rack by lifting them may well have prevented them going through most machine room doors or in lifts.
The only time forklifts, or more normally pallet lifters, were used was to lift them off the delivery vehicle, because they were delivered on pallets, but then so were most IBM 19" racks like the 7014-t42 that were used for Power, IBM i and previously xSeries systems. Normally these were de-paletted in the loading bay, then wheeled into place on their own wheels.
Another interesting point, which I've only just remembered, is that many old terminals used processors and support hardware that the MAME emulator can run. There are a bunch of original firmware ROM dumps on places like Bitsavers and Github that you can load in and get 100% accurate behaviour of these terminals in MAME. The problem is connecting them to a real serial port.
I was playing around with a Falco 5220e ROM image a while back, because that was a terminal I really liked at the time, which did a good vt220 emulation, but with the benefit of a real escape key, multi-page scrollback, dual RS232 ports allowing it to be connected to two systems at the same time, and even having a Tektronix 4014 emulation that I could drive from inside the Access Technologies 2020 spreadsheet for graphics on a serial terminal.
I am becoming very nostalgic in my old age.
For a long time after Lenovo acquired the IBM xSeries business, Lenovo contracted with IBM for field engineer support. I've always found, even in this day and age, that IBM's field engineers are a good bunch of people, who try very hard to satisfy their customers.
Not too sure about the last 5 years or so, not had much to do with them in that time.
I'm more concerned about the impact of pumping heat into the water surrounding the floating datacentres.
If they were in deep ocean, it probably wouldn't be too much of an issue, but in the shallower rivers or river estuaries, or close in to shore in the sea (for networking latency reasons), I can see these things having quite a measurable effect on the ecosystem that surrounds them.
I know that we've been using seas as cooling for a long time, for ships propulsion, and outfall from coastal cooling systems for things like power stations, but these things are rarely consequence free.
Even more so since they started exclusively using the 'countrified' version of the theme rather than the original orchestral version.
Even if I had wanted to continue listening to the Archers (I did used to listen on my drive back home because it followed the evening comedy show on Radio 4), this new theme makes me want to switch off or to another channel immediately.
DAB+ is actually more efficient than the original DAB, both in terms of the bandwidth and the power use.
This may be because the electronics are more modern, but I remember when I used to use a set of alkaline AA batteries in about 3 hours in one of the early DAB radios that I had. Early DAB radios were real power hogs. I used to get about the same time from a pocket AM radio on zinc-carbon AA batteries.
The small FM radios that I had used to use an alkaline PP3 battery in about 6 hours. But most of the battery FM radios I had used zinc-carbon PP9 batteries (big, heavy things) that used to last months.
I know I'm getting old, but I'm not that old!
I remember Radio 2 being on LW, I also remember Radio 1 being set up on 247 metres, especially for Junior Choice on Saturday and Sunday mornings. Before Radio 1, I started to listen to Radio Luxenbourg for 'pop' music (because I couldn't receive the pirates where I was in Surrey) on my little transistor radio that only did MW and LW. It was also often more convenient compared with FM radios as well, because you needed an extendible aerial for decent reception on FM (when I eventually got FM radios), whereas an internal ferrite core aerial was sufficient for MW and LW.
I don't think I remember it being called 'the Light Programme', (although maybe I do vaguely, my grandparents used to listen to the radio a lot, and I spent a lot of time with them when I was younger), and I do definitly remember that appearing on the dials on any number of radios that my family had knocking around.
They moved Radio 4 onto longwave when they felt that the majority of users would have switched to FM.
This is done all the time by many, many organisations.
They have a private DNS setup for their Intranet, which recurses out to public DNS servers for anything not in an internal domain.
Just treat the UK the way that companies operate, keep UK websites on .uk domains, and keep the authoritative servers for those domains in the UK (where they should already be).
I know it's not that simple really, but it is a model that is used and works. No need to split the internet.
But the railway bubble, like the canals, left a lot of long-lived infrastructure, like cuttings, tunnels, track beds, bridges, marshalling yards, stations et. al. which then helped drive industrialisation (and much of it still used!) even if the companies that built it no longer existed.
One really wonders whether bit-barns full of high maintenance, short-lived, specialised hardware that can't be easily re-deployed will have the same impact in the years that follow any AI downturn.
One of my concerns is that if someone does find a way of re-using some of this potentially redundant kit, it's going to have knock on effects on the further supply of new systems in an already shrunken market, causing the technology suppliers to leave, leading to a medium-to-long term supply problem of new systems.
I first used Ingres in 1979, from a BSD2.something on a PDP-11/34a running UNIX edition 7 for a database course at Uni. It was tremendously slow. One user would make everything else crawl, two would pretty much stop the system.
A few years later, I was running my 'own' PDP-11/34a, but this time, I had SMD disks and 2MB of memory (PDP-11/34a should only be able to address 256KB, but this one, modified by SYSTIME, had 22-bit memory addressing bolted on, making it a strange beast that was a non-I&D -11 that could, in theory, have the maximum 4MB of memory).
Jiggling the code a bit, and compiling up Ingres for a system with a maximum of 56KB per process required it to be broken down into 6 different processes for different, talking together through pipes (no shared memory on UNIX edition 7). But unlike my first experiences, there was sufficient memory to prevent it swapping, and I could also increase the number of system buffers by moving them out of the kernel space that allowed more files to be opened simultaneously. With this still modest system, I could get 4 or 5 basic QUEL sessions against the demo db running at acceptable speed.
The version of BSD 2 I had at that point had an overlay linker and loader on it, but I never managed to get Ingres to compile and run using this, but I reckon that if I had been able to, performance would have been better still.
