Posts by Michael H.F. Wilkinson

After we fix that, how about we also accidentally break something important?

Let me just add this one optimisation to the code ...

I was working on implementing an imaging filtering algorithm by a competing research group to compare to our own new shiny algorithm. The authors of the competing algorithm weren't allowed to share their implementation due to IP issues, so in the interest of fairness, we decided to do our level best to squeeze every last drop of performance out of our implementation of the competitor's algorithm (as much as we did to our own algorithm) before comparing them. We had successfully reduced computing time by a factor of three on both methods, when I thought of another optimization for our competitor. It boiled down to reusing earlier results rather than recomputing them, at no extra memory cost. Easily done. It knocked another 7% off the wall-clock time. Not much, but nice to have, and it took just ten minutes to implement.

It worked like a charm on all images in our test set. Except for one, where it seemed to enter an infinite loop, never terminating until we killed it. I spent hours and hours trying to work out what was wrong, but ultimately had to give up. We left out this last optimization in the final test. The 7% didn't make much difference anyway, as our algorithm was far faster, and had a better worst-case computational complexity: O(N log N) vs O(N² log N).

Revenge for being fired is best served profitably

The only hardware I "disposed of" was my work PC, which was going to be replaced by a more powerful one. My old desktop machine was no slouch, but it was equipped with fast, but slightly esoteric RD-RAM (2 banks of 1 GB each). Our admin wanted these machine replaced with machines with more common types of RAM. As my machine at home was considerably less powerful, I asked whether they minded me disposing of the item. "No problem" was their answer. I then asked if I could "liberate" the RD-RAM of another machine that was heading for the dumpster, and the answer was yes once more. I ended up with a nifty machine with 4GB of RD-RAM, which served me well for quite a few years. No direct profit, but quite a good savings.

Watch your mirrors: Tesla Cybertrucks have 'Full' 'Self Driving' now

Your mileage may vary

Had they tested this "Full Self Driving" technology in Naples, I do not think an average of one intervention every 13 miles would have been achieved. Once every 13 seconds in rush hour seems like a better estimate. Lovely place, absolutely great pizza (and other delicious food), but driving there poses ... interesting challenges, even to a seasoned driver.

Once self-driving cars navigate Neapolitan or similar chaotic traffic without incident, we can really speak of FSD.

Personalized pop-up was funny for about a second, until it felt like stalking

Clearly not in the category "Mostly harmless"

I would much prefer using messages like:

“Hi there! This is Eddie, your shipboard computer, and I’m feeling just great! "

"I know I’m just going to get a bundle of kicks out of any program you care to run through me.”

In a prank like that.

Doffs hat to the late, great Douglas Adams

BOFH: AI consultant rapidly transitioned to new role as automotive surface consultant

"In fact, I've helped a number of people take that first step into the space program."

It could only end one way. Clearly there has been a tragic Database Normalization Warning

The boss's car didn't stand a chance. Hilarious episode!

OS/2 expert channeled a higher power to dispel digital doom vortex

Wizardry

Was Jim really Gandalf coming to the aid of Rohan?

I'll get me coat. The one with the LotR trilogy paperbacks in the pocketses

Starfish Space to tackle orbital junk for NASA with SSPICY Otter

Not likely an Ubuntu release

Obstreperous Otter could be, or Spicy Starfish, but Spicy Otter? I don't think so.

Did you hear the one about the help desk chap who abused privileges to prank his mate?

Re: Early days

A Dutch Funding agency is name "NWO" I spent a lot of time getting frustrated that Word would constantly replace it with "NOW", even though I had autocorrect switched off. Then I found the "Replace as you type" feature. Killing that made live a little more bearable. Thank goodness said funding agency now also releases its application forms in LaTeX.

NIST: New smoke alarms are better at detecting fires, but still go off for bacon

Re: Heat detector

I have a carbon dioxide detector in the kitchen, smoke alarms elsewhere in the house. So far no false alarms (and no fires). Adding a heat detector might well be a good idea.

Crack coder wasn't allowed to meet clients due to his other talent: Blisteringly inappropriate insults

Re: One place I was at...

I once had to indicate that the (wired) mouse worked a lot better if you pointed its "tail" away from you. Said user was complaining that my latest software update had caused the mouse to work in reverse. To her credit, she did blush in embarrassment when corrected.

How about automating insults

In my early programming days (late 80s, early 90s, even before the first BOFH episode ever), I was developing a UI for an image processing system, which made life much more bearable for the PhD students and doctors trying to do their research. For many, a mouse was just something you spotted in the lab, rather than an input device for a computer, so I ensured there was an extensive help menu system and many (in my view) clear instructions on the screen. There was a robust error handling system in place to ensure input errors from users were handled, and clear error messages were displayed.

Despite these efforts, some users tended to get things wrong very frequently, and then asked me what they should do. Almost invariably, I replied with: "Please read the instructions on the screen". They would proceed to do so, and then sheepishly perform the instruction, after which the system worked neatly (as a rule). This got so annoying at times that I toyed with the idea of entering a "blood_pressure" or "rudeness_index" variable in the system that rose whenever a user made an error, and slowly decreased in time as the user gave syntactically correct input. This would be linked to an extensive table of error messages, which would progressively become more rude as the blood_pressure variable increased. A bit along the lines of the BOFH excuse generator, but then for insults.

Sadly (or perhaps fortunately) I never got round to implement it, but a man can dream.

I don't know what pressing Delete will do, but it seems safe enough!

Ouch!!!

There should really have been a dialogue box popping up with the warning

DO YOU REALLY WANT TO SERIOUSLY TICK OFF ALL THE OTHER ADMINS?

Y/N??

REALLY??

Y/N?

OK, IT'S YOUR FUNERAL

More seriously, I am sort of surprised there wasn't any warning along the lines of "are you absolutely sure you want to remove these entire accounts from the system". Or maybe Pat ignored this, as you all too often get superfluous warning messages popping up that you just click away because you are tired of them (been there, done that). Sometimes that has dire consequences.

NASA engineers play space surgeon in bid to unclog Voyager 1's arteries

Re: Stunning engineering....

It's not just having anything built today last so long, but also in a harsh environment with no way to service it. I do not doubt there will be, e.g., Volvos built today which will still be working in 2071, along with many other old-timers, but only because they are stored safely, and lovingly maintained, not blasted by micro meteorites, harsh radiation, extreme cold. NASA's engineers are nothing short of miracle workers.

BOFH: The Boss is right, the applications of AI are truly staggering

Indeed, we speak of Database Normalization Warnings

Muppet broke the datacenter every day, in its own weighty way

This kind of semi-random, intermittent error is such a pain to diagnose

A friend of mine once worked on a 68020-based computer (quite a beast in its day, I think it had a whole 16 MB of RAM), which would be on the fritz from time to time. It turned out, there was a tiny break in one of the connections on the motherboard, which sort of closed when the board was slightly bent one way, but opened if the board was seated slightly differently. Some hawk-eyed technician found it, put a little solder on the break, and the board worked flawlessly.

Python script saw students booted off the mainframe for sending one insult too many

OPERATOR != BOFH

Evidently. Quicklime and carpets weren't involved in the sanction, for starters, nor was a cattle prod.

OpenAI co-founder's Safe Superintelligence startup inhales $1B in funding

Re: Alright, boys and girls

Could I be a "Non-linear neural network layer expert"?

"This is to harness the power of non-linear dynamics within the general framework of deep convolutional layers in current neural network architectures"

Not sure what this will do, but I think it sounds cool

No grey beard, but (thinning) grey scalp with complementary grey stubble when I cannot be arsed to shave in the morning

To patch this server, we need to get someone drunk

Re: Reminds me of a claim by a sysadmin

NT didn't have the limit 95 and 98 had, but I seem to recall one kernel patch for NT claimed to repair 60 out of 300 known memory leaks in the kernel. If it did have this many memory leaks, I would guess uptime was limited due to memory slowly clogging up.

Re: Reminds me of a claim by a sysadmin

True, but I tended to take his claims with a pinch of salt afterwards

Reminds me of a claim by a sysadmin

He claimed his Windows NT servers had zero downtime, and each and every of his servers had run for more than a year continuously. Now I had a little experience in Windows NT administration, as I ran a Windows NT 4.0 machine at home (mainly to prevent the missus from performing "spring cleaning" activity, wiping out "unwanted files" such as config.sys). I dutifully downloaded and installed patches as they arrived (almost on a weekly basis), and in my experience, each and every patch required a reboot. I wondered if he had some trick I wasn't aware of to patch the systems without needing a reboot, or whether he had several redundant servers, allowing the services to remain available 24/7. I first asked if he applied every security patch, which he said he did. I then asked whether these didn't require a reboot, and he answered that of course he rebooted the systems after patching. So I suggested that his servers had not run continuously for a year. His answer was he didn't count the reboots for security patches as downtime, because he didn't physically power down the computers.

This was some strange usage of the phrase "zero downtime" I was not previously aware of.

Techie made a biblical boo-boo when trying to spread the word

Book of Armaments, Chapter 2, Verses 9-21

might have been a better choice for the test message

A nice cup of tea rewired the datacenter and got things working again

Of course tea makes things work better

Especially when used as a strong Brownian-motion generator with an atomic vector plotter suspended in it. Fire up the old Bambleweeny 57 Submeson brain!

The one with the HHGTTG radio play cassette tapes in the pocket please.

Re: There is a proper way to do most things

Darjeeling is my preferred tea in the afternoon, but at work I stick to Keemun black tea. No matter how strong you make it, it never turns bitter

Astronomers back call for review of bonkers rule that means satellite swarms fly without environment checks

Re: Now just a cotton-picking minute, there

I thought the shortest possible unit of time was the New York second, i.e. the time between the traffic light turning green and the yellow cab behind you honking his horn.

Doffs (grey Tilley today) hat to the late, great Terry Pratchett. I'd better be going. The coat with "Lords and Ladies" in the pocket please

Microsoft security tools questioned for treating employees as threats

I would think the BOFH fully agrees ...

that all employees are a risk, and I am sure he happily agrees he and the PFY are a grave risk as well (grave as in "shallow")

Bargain-hunting boss saw his bonus go up in a puff of self-inflicted smoke

Not blown board, just one chip

I remember our then supplier of computer equipment for our image processing systems installing a 80287 math co-processor for our 12 MHz 80286 PC-AT compatible machine (the faster of the two machines we were using at the time). That worked fine for a brief while, until it didn't, and I could smell something fishy (and I don't mean the content of Baldrick's apple crumble). I opened the case, and sure enough, the 80287 chip was not looking happy, with a slight bulge and discoloration of the housing. On closer inspection, the part was labelled 80287 - 10. I checked the other, 8 MHz machine, and it had a co-processor labelled 80287 - 8. I checked some data sheets, and indeed we had been supplied with the cheaper 10 MHz version. It apparently did not like being overclocked. The company claimed I was wrong, but did supply a new 12 MHz one free of charge.

Future machines were sourced elsewhere. Can't think why.

Tech support chap solved knotty disk failure problem by staring at the floor

Everything was earthed properly, but the Leica power supply was simply a beast which when switched on caused lights to dim briefly. It should never have been certified for use in general, let alone in a hospital, in my opinion. The Olympus and Zeiss power supplies for exactly the same type of mercury UV lamp caused no problems whatsoever.

But luckily, Emmett got to the bottom of the problem quickly

The strangest problem I encountered was when the code of an image processing system which worked happily on a (for the day) powerful PC (486 @ 66 MHz, 8MB RAM, and a Matrox MVP-AT/NP image capture and processing board) in the microbiology lab where I worked next to an Olympus fluorescence microscope, borked when running on essentially identical hardware, and same MS-DOS version, in the department of dermatology, next to their very fancy Leica fluorescence microscope. Mysteriously, a previous version of my code worked fine on both. The bad behaviour only showed up when the mercury UV lamp used for fluorescence microscopy was switched on, and only on the Leica microscope. I suspected an RFI problem, but the mystery was why it only hit the new code. Switching back to the old release solved the problem, but the old release was for a Matrox MVP-AT board. The MVP-AT/NP version had an additional "Neighborhood Processor" (hence the additional NP) which was essentially a hardware accelerator for image processing tasks. Not using it caused a significant slow-down in processing speed. The "solution" was to replace the image capture part of the code (when the fluorescence microscope had to be switched on) with the MVP-AT version of the code, and do the offline processing of the data with the faster MVP-AT/NP version of the code.

This was far from ideal, as switching the Leica UV lamp on and off could cause the computer to crash anyway, so the protocol became: switch the computer off before switching the fluorescence lamp on, start the computer up, do your image capture, switch the computer off, switch the UV lamp off, switch the computer on and do the offline analysis.

BOFH: Videoconferencing for special dummies

I trust the hammer in question ...

was a hefty lump hammer

Developer tried to dress for success, but ended up attired for an expensive outage

At least the problem was quickly contained

Deary me, and it's just Monday

Stargazing with the Beaverlab Finder TW2

Re: Ask an astronomer

Planets are bright enough that light pollution doesn't matter much. Heat rising from buildings is a much bigger problem.

Re: I'm not an experienced star-gazer

That's odd. A 6" scope should easily resolve Saturn's rings (even an 80mm does that). Maybe your C6 is out of collimation, or the tube wasn't cooled down to ambient temperature. The latter causes thermal currents within the optical tube, and these completely ruin the view. I need to leave my C8 out for about 45 minutes before the view stabilizes.

BOFH: The true gravity of the Boss and the 3-coffee problem

Cafeine overdose incoming!

Armed with a chainsaw.

This is why I only drink tea. My colleagues might get very nervous if they thought I was drinking large quantities of coffee.

BOFH: Well, we did tell you to keep the BitLocker keys safe

Superb episode once again, letting the Boss dig himself deeper and deeper into hole

and then charging him for the use of a ladder to get out

"Because I need to recover the authentication server. But you made me stop that and start on finance."

"What?"

"The authentication server. Active Directory. Comes after accounting."

"You should have told me that!" he snaps.

"You wouldn't let me."

Hilarious! Classic BOFH

Re: recovery workshops in, uh... Belgium?"

Plus extra points for slipping in the worst rude word in the galaxy

Doffs hat to the late, great Douglas Adams

Facebook prank sent techie straight to Excel hell

I had a little bit of fun putting a program named CRASH.EXE in the directory of the executables of an image processing system I was developing back in the days of MS-DOS. The more inquisitive users of course wondered what this program did. What it did was redirect the keyboard interrupt to my own interrupt handler, which did nothing whatsoever with any key press. It then caused the screen to flash and put a message in the centre stating the system had crashed. No key press could stop the program, so a hard reset was the only option.

On a single-user, single-tasking OS this could be done safely because no other program could be running. It did cause some consternation with people sheepishly admitting they had crashed the system, and could I, being both admin and developer, set things right. Generally I got very evasive answers when I asked them what they had done to cause the system to crash. Some owned up they had run CRASH, whereupon I would ask them what they had expected such a command to do. Some mumbled they thought it might be a game. Some bolder ones asked why such a program was on the system in the first place, and I said it was there for testing purposes.

Testing users, that is, not the system.

Dangerous sandwiches delayed hardware installation

Sandwiches can be dangerous ...

especially if prepared with Archchancelor Ridcully's Wow-Wow sauce,

or of course if Bergholt Stuttley Johnson had anything to do with the recipe.

I'll get me coat

There is no honor among RAM thieves – but sometimes there is karma

Nice one

Excellent revenge on the overzealous manager.

It reminds me of the time I got a replacement machine at work. As I was into fairly hefty image processing work, my old machine was not considered fast enough. It was, however quite a bit more powerful than my home desktop, so I asked the admins what they were going to do with the old one. The answer was that the would have to dispose of the old machine, but if I would want to "volunteer" in the disposal task this, that was OK. The machine I liberated was unusual in that it used RDRAM, and only 2 out of 4 slots were in use. I noticed an identical machine was slated for disposal, and asked the admins if I could liberate that RAM as well, which they kindly allowed. The machine worked neatly for many a year at home.

It really helps to be good friends with the admins

BOFH: It's not generative AI at all, it's degenerate AI

I am seriously tempted ...

To make degenerate AI a mandatory part of our machine learning course (and the relevant BOFH episodes mandatory reading material)

Great episode

Stop installing that software – you may have just died

No strange reason to stop work but I did spot an old box still in action

But I did spot an ancient HP-1000 mini-computer from 1977 or so chugging away processing data from the Westerbork Radio Synthesis Telescope (at one time one of the largest radio telescopes in the world) in 1996. It basically had been chugging away since the commissioning of the telescope. By the time of my visit they had set up a Linux PC to process the same data, and wanted it running 24/7 for a year before considering decommissioning the HP-1000. Some old boxes are very hard to kill.

Big Tech's eventual response to my LLM-crasher bug report was dire

Re: Well free the prompt, then

++++ OUT OF CHEESE ERROR ++++

++++ REINSTALL UNIVERSE ++++

++++ REDO FROM START ++++

Users rage as Microsoft announces retirement of Office 365 connectors within Teams

Re: A shower of shite

I tend to compare it to poetry. Vogon poetry, that is (or perhaps even Azgoth of Kria poetry)

Outback shocker left Aussie techie with a secret not worth sharing

Ouch!!

I was sort-of bracing myself to some deadly spider or snake hidden in the bowels of the telephone exchange causing trouble, but no assistance from the local fauna was needed for this mishap.

It does remind me of the time I was installing a new, far more robust dimmer system in a cafe I used to frequent. Before attaching the new dimmer system to the mains supply, something prompted me to check the wiring. My suspicions were quickly confirmed when (amongst various other horrors, I found a green/yellow (earth) cable actually being used as the live 230V 16 A wire. Not something I would recommend. I corrected some of the worst transgressions before hooking up my dimmers. I also told the owner to get an electrician to sort out the other horrors I found. I am not sure if he did, but my dimmers worked flawlessly for several years, until the cafe was sold and completely renovated.

Windows: Insecure by design

Re: how much punishment are you willing to take?

Thumbs up for the Nac Mac Feegle battle cry

An arc welder in the datacenter: What could possibly go wrong?

Re: well...

ale awarded

all alliteration always appreciated

Blame-shifting gone mad

Shifting the blame for a cock-up is of course par for the course in these situations. It's a bit rich to shift the blame to someone who did a similar job competently and wasn't anywhere near.

The story also brings back memories of the time I was teaching someone to solder electronic components onto a circuit board, but they kept referring to it as welding. I would certainly not recommend welding to attach even the more robust triacs (380V, 32 A, TO-48 housing) used for a theatre dimmer to any circuitry, so just to drive the terminology home, next time they said they were going to weld a component to a practice circuit board, I couldn't resist going "KZZEERRT" as the soldering tip touched the component. Gave them quite a shock, but they did find it funny. Their soldering was actually pretty neat.

For the record: You just ordered me to cause a very expensive outage

I don't like saying "I told you so", but ...

I told you so!

Those would be an appropriate parting shot from Norman before storming out

Admin took out a call center – and almost their career – with a cut and paste error

It would of course not do ...

to blame the upper echelons in management for skimping on a test environment, which was ultimately the root cause of the issue. I do feel a more experienced BOFH might have hinted that said "grown-up humour" apparently came from somewhere in the system, possibly from management or beancounters, but he would need to spend some serious (paid) overtime researching who he was going to pin it on was responsible for this problem.

BOFH: Why's the network so slow?

"spanning tree, bayesian binary caching, flapping vlans, combination crypto signing errors"

Brilliant stuff. Sounds like the BOFH excuse generator has gone through some updates