The Legacy Of The Floppy Still Looms Over Windows

We no longer use floppy disks on the vast majority of computers, but a recent Old New Thing blog post from Microsoft sheds light on one of their possible unexpected legacies. It seems Windows disk cache items expire after two seconds, and as the post explains this has its origin in the development of MS-DOS 2.0.

Disks, especially floppy disks, are slow compared to computer memory. A disk cache is a piece of memory into which the operating system puts frequently loaded items to speed up access and avoid its having to repeatedly access the disk. They have an expiry time to ensure that the cache doesn’t become clogged with data that hasn’t been needed for a while.

IBM PC floppy drives didn’t implement any form of notification for a disk eject, so it became quite possible for a disk to be ejected while the operating system still believed cached data from it to be valid. Thus a pair of Microsoft engineers tried their hardest to swap floppy discs as fast as they could, and it was discovered to be an impossible task in under two seconds. This became the cache expiry time for a Microsoft OS, and thus we’re told the floppy’s legacy lives on as more than just the ‘save’ icon.

As this is being written the Internet is abuzz with a viral Tweet about railroad gauges having an origin in the width of a Roman horse, that rail historians are debunking with a reference to the coal tramways of [George Stephenson’s] Northern England. It’s thus sometimes dangerous to take simple soundbite origin stories at face value, but since in this case our source is Microsoft themselves we think we can take it as being close to the horse’s mouth. Even if it isn’t a Roman horse.

IBM floppy drive image: Michael Holley [Public domain].

55 thoughts on “The Legacy Of The Floppy Still Looms Over Windows

    1. Think 5 1/4 inch floppy, they came before the 3.5 inch ones. Longer insertion distance, and flexibility slowed down the swap process. Also, the drive “doors” needed to be un/latched each time.

    2. 5-1/4″ floppies? That’s pretty fast.

      Flip the door (this was before the lever type, which was no faster) which lifts the upper frame, daisy (the plastic, fingered bowl shaped piece on a ball bearing that presses the spindle to the top surface of the cup, which is belt driven by the motor), grab and pull out the disk (Elephant Memory systems— “An Elephant never forgets!”), preferably without folding it, insert the new disk, again, preferable without folding it or hanging it up on any of the internal bits that stuck near the path, close the door, bringing the frame and daisy down to grab the disk, then the spindle needs to start.

      Two seconds? Good luck beating five without damaging something on the, then, $400 or $500 Shugart drive unit.

    3. Look at that full height IBM drive. It’s got a latch in the middle. I can imagine undoing the latch, pulling out the old floppy, picking up the new one, putting it in, and operating the latch again might take 2S.

      1. With the 8″ floppy drives on IBM banking equipment, it was possible to get two or three disks into the drive, and then close the gate – we had to be very careful to tell users to take out one disk before putting in the next.

        This is all to say, I bet you could open the latch, jam in a second disk and close the latch again all in under 2s! If everything lined up just so, you might have been able to trick the OS into writing to the double-floppy disk sandwich.

        1. At least once I put in a floppy without taking the previous one out, this was 5.25 inch. I couldn’t understand why the previous directory was still being shown. My first thought was that there was a cache. I don’t think it took too long before I realized I had two disks in there.

          They may have beeen single sided disks at that time.

          I vaguely recall something about tye need to press conntrol-C or something when changing disks to tell something that a disk had changed, but I.can’t remember if that was hatdware ir an ooerating system, and if the latter, likely CP/M.

          1. If my memory serves me right there were 5 1/4 disks that could be flipped there was a notch on the edge to tell the drive what side it was? Or was that a Write protect notch? It’s been so long since I touched a floppy lol used to have thousands of them when I was running a Wildcat BBS had a bunch of dumb headless 286’s each with a Floppy drive (to boot DOS and load the drivers for the USR Sportsters, Lan, etc) the nodes were getting power (3 mobos per PS) had to splice to set that up the the Mobos were all installed Vertical on metal shelving had plastic guides to hold the edge of the MOBOs to prevent grounding. It was a Frankenstein (but quite organized for 100 noses) in the basement of a PC store! Those were the days.

  1. Regarding railway gauges, it’s worth noting also that standard gauge only because dominant in England relatively late. Brunel’s Great Western Railway was built with a gauge of 7’1/4″ – more than 18″ wider than modern broad gauge – and stayed that way until 1892. But Stephenson won the commercial war and so when a standard gauge was chosen, the one with the greatest mileage won. This standard was only passed into law in 1846, by which time considerable amounts of track had been laid in the United States.

    1. I believe you are correct there. However I have heard that the standard was the width of the Roman Chariot, and that all of Europe followed it, except Russia. They used a different one. For example, the US can sell engines to a European company, and it can be used to pull cars. Just like a European Engine can be used here, although why would anyone want to.

      Although Jenny this does not explain what the statue in your garden is doing staring. (Don;’t Blink!)


      Interesting fact: in Rome (and other Italian cities the trams have a 10 mm wider gauge compared to trains, 1445 mm. Sometimes rails were used by freight train at slow speeds that were normally used by trams, in the photo is visible the transition between train and tramway. This was due because trans and trains have different wheels, and making the tram ones wider made possible the interconnection.
      Probably was due of the reduced nosing forces on lighter and slower trams and the different shape of the trucks.

  2. I haven’t owned a Windows machine in over a decade so I have little personal concern over the state of the OS, but I’ve always thought that Microsoft would do well to just dump the entire codebase and start over from scratch at this point.

    Obviously there’s an insane amount of legacy software they need to remain backwards compatible with, but they could use some kind of integrated emulator like Apple did with Mac OS X. It sort of goes back to the talk about completely dropping 32 bit support in Linux: at some point we’re going to have to cut the cord on decades-old hardware/software, but when?

    1. Put on an integrated emulator, and most problems can be solved.

      But for a new codebase, they would need to forget a lot of their billiant ideas implemented in win10, and just return to something that works correctly and lets one use the computer.

    2. Well, with Windows adding the WSL and porting some of their Net core to Linux … It makes you wonder the direction the W$ OS is heading. I too have been W$ free (at home) for many years. Haven’t missed it either.

      This talk of floppies brings back a lot of memories loading DOS/CPM/Windows/Linux off of floppies … and then backups on floppies… and floppies with errors… :) . Those were the days…

    3. It would take too long (and cost too much) to write a new OS from scratch and MS would never, ever use and open source software that would even allow a vaguely, distant, remote almost impossible chance of “mere humans” having legal access to any of “their” code so it won’t happen. Windows forever (weather we like it or not.)

    4. Microsoft should have gone full 32 bit when the 80386 CPU was introduced. Same for 64 bit when the first 64 bit CPUs were introduced.

      At least 64bit versions of Windows don’t support running any old 16 bit software.

      1. It does not support running 16bit applications in long protected mode only because the CPU itself cannot handle that. In pro versions of windows they include an emulator of some sorts (IIRC). So they would gladly support 16bit if they could.

  3. “I don’t know if the modern two-second cache flush policy is a direct descendant of this original office competition, but I like to think there’s some connection.”

    That is a quote taken straight from that Microsoft article. So I am not sure if it’s as straight from the horse’s mouth as you wrote…

    1. And besides, it’s not official Microsoft documentation, it’s just a developer blog. But in my books, Raymond Chen’s blog is much more reliable than any Microsoft document or press release could ever be :)

  4. Okay, i see a challenge coming… it’s the floppy challenge. How fast can YOU swap a floppy?

    There are 3 possible kind of drives: 8 inch, 5 1/4 inch, 3.5 inch But since we gotta start somewhere, let’s do the 5 1/4 inch floppies. Because they are perhaps the most realistic challenge. As the 3.5 inch is to easy and the 8 inch is to rare.

    Now to keep things fair we also need a standard drive mechanism. This is going to make things difficult. As there are many kinds, some have rotating levers, push and tilt mechanisms, spring loaded push and tilt mechanisms.
    I’m pretty sure that there are drives that are easier for the left and some that are easier for the right handed. Speaking of hands, how many hands may a user use. One or both or perhaps a second person pushing the lever… no, just one person, that’s easy to agree upon.

    Then we must perhaps also agree on a special type of disk. Because well… some are more floppy then others, so that would not be fair. The color might not be an issue you might think, but the default black floppy might cause contrast problems in a default black drive because in order to achieve less then 2 second time values, everything counts. The position of the drive, should be perfectly level, otherwise the disk might even fall out by itself reducing time.
    Oh, and the disk should have no label describing it’s contents as that might obstruct the mechanism.

    let the games begin…

    1. 8″ would be longer, as the 120V induction motor that most of these used was pretty slow to spin up. Just timed one at about 4 seconds, but, as it hasn’t been used in 20 years, might have been faster in its prime. Also remember that the 8″ drives wore out disks fast since they generally spun continuously from the time the frame was engaged.

        1. The local surplus place about 25 years ago had boxes of odd flopoies. There were some that actually flopoed but had a metal hub like the 3.5inch we know and love. But I can’t remember if they were 3 or 3.5 inch. They were obviously also-rans sonething that lost out in the transition to smaller than 5.25 inch disks.


      1. Fuji 2.5″: the OFFICIAL floppy disk of the 1988 olympics. I think I still have a box of them some where. The Minisport went in about 1994. I KNOW I still have an in-box OEM battery for that somewhere.

  5. DiskChG signal was there. Exception to cheaper disk controller that may not have implemented properly. 5.25 Cheap drives too. Half height =quarter power on some.
    CrApple floppy drives were horribly slow and no DMA but peanut butter jelly resistant. Commodore should had sticker “got alignment disk¿?”. More of a software issue. Not all drives were Tandon or Shugart. Story would be more believable if inferred multiple 5.25” floppy drive and controller configurations. At best more anecdotal m$ stuff. That company had software for more than Blue.
    8in floppy drives that used AC induction motor may take longer than two seconds to stabilize and whipping around a freestanding 8 in floppy could be assault with a deadly weapon. Make awesome throwing stars. Some skill required. Evidenced by bean counters insistance of buying non reinforced hubs for drives that needed/ recommended it. Sure putting photo of ass on drywall divider and have a contest to see who can get them to stick may have been a bit much.

  6. I had a 5.25″ drive on my TI-99/4A with a spring loaded door and spring ejection of the disk. I quickly learned to position my hand just so, hit the button with my thumb and catch the disk as it snapped out.

    If I didn’t, the disk was on the floor. But for disk change speed trials, that would be the drive to use. Get the second disk ready, hit the button then as soon as the first disk is clear, pop the second disk in and sweep the door closed in one smooth motion.

  7. I recognize the drive in the image… that is out of an Osborne 1 “luggable”. I had one, briefly, a year or two ago. That drive is an MPI 5.25″ IBM-compatible drive, or, at least, the mechanicals from it. The circuit board on top is a custom in-house job by Osborne – I’m told they typically bought the standard drive, controller and all, binned the original controller, and substituted there own equipment.

      1. And still reserving A: and B: for floppy drives. I can’t say I’ve tried to see if I could reassign those letters to other drives, but I’m mildly irritated all the same.

        1. You can assign A: and B: , you can also use subst to assign a drive letter to a path, but the real absurdity is that we still have drive letters at all. There is some logic to partitioning large drives but the Linux convention of mapping them to paths in a unified file structure seems much more logical.

          1. Why?

            The physical drive is a separate unit, which may be removed from the machine, so attaching it as a “folder” in the file system hierarchy tree doesn’t reflect the reality of the machine. The Linux/Unix convention of having a fixed hierarchy that requires all these elements to be in their correct places is really cumbersome and makes many things needlessly difficult.

            Like say, “if I have X amount of space on physical drive A, and Y amount of space on physical drive B, can I install a piece of software that fits on B but not A”. With the drive-letter convention, you just install the software on drive B, whereas with the linux convention this may or may not be possible, as some files may have to be copied to device A because that’s where /foo is.

    1. I’d have to disagree. I was a tandy tech for almost 10 years & saw a LOT of drives. I’d say this is a tandon, early, discreet component, double sided.

      as for the 2 seconds, most drives would do a home (seek to trk 0 from unk trk), & then seek to/read the directory (trk 17). There is also the spin up time & wait for index mark/hole, at 300 rpm, things travel pretty slow, so theres close to a second in ‘built-in’ time delays.

      Also, the 8″ drives tandy used did spin continuously, but had a head load signal that engagaed/disengaged the pressure on the r/w heads.

  8. Surely a much more obvious legacy is that A: drive and B: drive are still ‘reserved” for floppy drives right up to and including Windows 10. In the absence of floppy drives, why hasn’t the main system drive not commandeered the A Drive assignment?

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.