I believe that a floppy-based raid 0 configuration was referenced in Dante's Inferno as an object one would encounter in the 5th circle of hell (wrath), if I'm not mistaken.
@@timmooney7528 I did it with 3 1Gb Iomega Jaz SCSI dived in Raid 0 for backups on my Linux home server. Worked fine until it stopped working. :-/ Back than i wished you could make CD-Rom RAID for backups. I tried it, but my knowledge how to make such a thing, and make it with my available hard drives wasn't there. I later made a Raid 6 with 7 1Gb Iomega Jaz SCSI drives. I had to put them in their own enclosure with an dedicated SCSI card. It worked nicely, until i ran out of disk space ( 5Gb isn't that much anymore :-/ ).
This is reminding me of that old video by "The iBookGuy". He put several USB flash drives in RAID, then later changed his name to "8-Bit Guy" to hide the shame.
Possibly the reason the RAID array failed, the heads on those drives may need to be cleaned. They used to sell "floppy drive cleaners" .. basically a floppy disk with a cleaning pad instead of a magnetic storage medium. Put a couple drops of alcohol on the pad, put the "disk" in and try to mount it in OSX. Do this a couple times and it should clean the heads of the floppy drive possibly increasing your success rate.
those cleaning disks are the worste. the cleaning pad is also inside the floppydisk itself to protect the magnetic disk. and will collect crap during its life.. back when where still booting of disks it whas usualy the most dirty and teh first one to go. even if you use fluids it not going tobe enough and doing it while powered on is also not such a asuring idea. just popoff teh front plastic bit of the drive and clean it with a cotton tip and some alcohol. alcohol will just do you dont need to desolve annything else but the crap However i think its had more todo with the controler type of the usb floppy it whas distict a sertian type that failed. but seem to work. like its 2.8mb able or even zipdrive formats like 120mb using difrent drivers that dont have the same hardware function i think the same fallback legacy driver for all the drives will make them behave the sameway and will allow raid
There's another problem, that doesn't have to do with any one drive, but trying to use them all at once. which is going to be the case when your using raid. so I looked up a usb floppy drive and it stated that they use 500ma (or 0.5a) at 5v and your powering them with two strips. so to see how much power these drives are going to use when they are all working in tandem take the power use of a single one and multiply it by 17 (half of 34 since the other half is on a different hub). when you do the math that comes out to 8.5 amps per hub. Convert that to watts and that hub needs to be able to supply over 40 watts of power (42.5w) now a quick search of powered usb hubs showed various levels that max out at though 36watt was common. if that's the case with your hubs then while you'll get the drives to all connect when you start them up slowly like that as only one drive is active at a time and the rest are in standby so not using near as much power and thus not overloading the hubs power supply. but when they all go to active read or worse write mode at the same time some of them are not going to get enough power and thus fail and like you said if one of the drives fails it brings down the whole ship. I was wondering about transfer rate because older computer. As it would overwhelm a 1.0 usb port. But turns out even 1.1 should have the capacity as we are looking between 5-8mb/s and 1.1 can handle 12mb/s. I think that computer you mentioned has 1.1 so while the data side should be okay. I say should because it looks like the max practical speed of a floppy is 100-250kb/s though they are rated at 500kb/s-1mb/s but even so it would more likely just run slower then it could. though that would depend on how well the raid controller is able to handle the likely out of order nature of the writes and reads coming in if it was to get up to the full speed and not just 250kb/s speed. Still make sure that the hubs are rated for the power drain that's going to hit it as that 8.5 amps or 42.5watts is only from 17 of those 34 drives doing a write cycle at the same time. this is what I would check first, because if your powered usb hubs ac adapter is not outputting at least 43watts or 8.6 amps (as the hub itself is also going to need some power) the rest doesn't matter there will always be at least one drive that is going to fail due to lack of power during the read/write cycles even if all the drives and disks are in perfect working order. And like you said it only takes one to crash the array.
@@CrossRoadsOfTime this is software RAID, so the OS is the controller, but aside from that yeah you’re exactly on the money. Needs more hubs probably. Could also be a mismatching head seek latency or something between a few models of drive. (Or even the former power issue directly causing the latter issue.)
@@kaitlyn__L never said it was a hardware raid setup. even with software raid, there is still a Controller managing the RAID setup. it's just that in his case the controller is a program running in the OS. vs a program running on dedicated hardware.
Just sad that Retro forgot to take the opportunity to name his floppy raid: Shadow Legends. Also love how the chess game takes so long to load, that it feels like he managed to ressurect the Cassete tape era of games, where games would take alledgedly 113827091327312907321987 hours to load.... because they were stored in a freaking Cassete tape...
the Guinness world record should add a record for "Most number of floppy drives RAIDed". they already do have a lot of ridiculous records so this one should slide easily through...
You have to pay to get your records witnessed, which isn't cheap at all. They might not even find it worthwhile to certify, since only nerds would be interested not the people who buy the Guiness book.
Since each floppy disk sector is 512bytes, the maximum amount of drives would be 2880 for peak throughput. That's reading the entire contents of each drive in one go anywhere in the raid. However, seek time is quite high, at over 500ms. As you add more drives, the seek time is eventually going to out weigh the throughput transfer. Given the fastest transfer would be around 250Kb/s from the inner track of a high density disk, this means we're looking at exponential loses from the raid after around 7.5 drives at 300rpm standard speed. It would be interesting to see you bench mark the drives as more are added/removed to see how that plays out.
I think you're mistaken on some things. I don't see why the number of sectors per drive would have anything to do with sequential throughput. The only limiting factors would be the bus and computer speed.
@@eDoc2020 There was a typo in that message. They meant with 512b sectors, a raid array of 2880 floppies would mean you could theoretically read an entire floppy's worth of data anywhere on the array every cycle.
I don’t get this either. What does sector size have to do with anything? The advantage to RAID-0 is having I/O operations in flight simultaneously. At some point, the I/O ops will complete before the next can be scheduled, so there’s no point adding any more drives. Is that at 2880? I don’t know. The latency doesn’t quite make sense to me either. For sequential reads, the latency of the initial seek will overlap as each drive navigates to the start of the data chunk, but you don’t need to wait for the first to complete before asking the next to start seeking, so it just ends up being a fraction longer than a single seek to have them all positioned, at which point data is already being read by the first drive, so you’re really only limited to the seek time of one drive. Random non-consecutive sectors would be a different matter, but so would they be on a single drive.
Edit: Sorry, that reply was written on my phone so I wanted to keep it short. I'll explain in a bit more detail now. Sector size is how one calculates the maximum theoretical disks that can be added to an array before you yield zero gains.(There are other factors that play into this so you obviously won't reach anywhere near that amount before you encounter 0 or even potentially negative returns, so don't take this at face value) "At some point, the I/O ops will complete before the next can be scheduled, so there’s no point adding any more drives. Is that at 2880? I don’t know." The access time; in this case, the time it takes for the track to make an entire rotation would be the limiting factor, not the speed at which one can process the data. Seek and Access times are what limit you from theoretical max throughput which was what the OP was factoring into his/her theoretical calculations. Reading 1 sector per disk per drive per rotation of a track sequentially would be the ideal condition and thus would be limited to a drive's RPM and to a much lesser degree, the time it takes for the read head to move over a single track(like you said, the added time added per additional drive is a tiny fraction above what it would take one drive); I think at this RPM the "seek" time is a non-factor as access time is already the bottleneck. @@nickwallette6201 The 2880 drives was maximum theoretical throughput if seek time and physics weren't a thing. Seek(or read head travel and disk RPM) time becomes the limiting factor for fragmented data, and to a lesser degree for sequentially written data.... the laws of physics are still a thing here. Just because data is written sequentially doesn't mean it will all be read in a single pass of the disk. In fact data is not read that way from a spinning disk which is why RPM of a disk matters. I hope this clears things up a bit
@@kefler187 Not yet, but I think I may have to think upon this for a little bit. I am making some assumptions that may not be true, such as that a drive could read an entire _track_ in one go, without having to wait for the the disk to rotate through an entire revolution between sector 1 and sector 2 of a given track. It is my understanding that _not_ being able to do this is why interleave was a thing back in the early days of hard disks, so I assume (?) that it is not a problem with floppy drives at a rather pedestrian 300 RPM. But I could be wrong. My other assumption was that, once you have the first drive seek to the appropriate track (seek time) and then wait for the disk to rotate to the desired sector (access time), you can then essentially start streaming that entire track to the host, assuming the host wants enough sectors of consecutive data that even with a one-sector-per-drive stripe, the data request would require reading adjacent sectors on every drive. If that's the case, then with a sufficient number of drives, you could, in theory, paper over the seek and access times of having to jump to the next track, if there are enough drives that you are still receiving data from one drive while another is seeking to the next track. At some point, the CPU load of dividing the chunk into individual I/O requests and allocating buffers for the incoming data will overcome the throughput advantage of adding another drive, but I don't yet understand how that has anything to do with sector sizes (other than smaller sectors require more overhead, but it wouldn't be a fixed number of drives that saturates a host's ability to process them in a timely way), nor how the RPM is really a factor, other than of course being a limiting factor in the individual drive throughput, and of course, access time, which can also be covered up by staggering I/O requests (which a single-threaded host would have to do anyway, to process them in serial fashion; and a multi-threaded host could only do per number of cores / CPUs.) I'll chew on this, though. It's interesting to think about there being some calculated limit, but I don't yet see how that is.
Reminds me of the time I asked my high school electronics teacher if it was possible to use a floppy disk as RAM, to which he answered: "only if you want to have the world's slowest computer!"
Yep, but it works.. even today. Some bluray-drives could work softram. It's not even "worsest" idea what you could find. Tape drives aren't so easy peacy put in ride and need whole custom own raid program. And cassette library with few drives and hundreds of LTO's.. no no no.. only way if you modify drives and made endless tape between drives and could buffer it somehow (like old ink casettes) and run both way ofc.. that must be weirdest "raid", with one media and many raid drives.. ;)) So, think about that.. USB sticks raid with good sticks isn't even stupid idea if you have enough usb3 channels to cpu.. you don't want use usb3 splitters with flash, it's turns same fast as this floppedraid... (:
@@jannejohansson3383 Can't use LTO for RAM without destroying it by shoe polishing. Tape is meant for streaming data in one go, not for random access memory where the tape needs to go back and forth constantly, it would break within days. USB sticks as 'half RAM' never properly worked, at best it can give a placebo effect in my experience, either the flash chip is too slow to make a difference or the CPU spends way too much time running it... Take the original Commodore 64, ~1Mhz, will require RAM capable of just over 2MB/s at 8 bits at a time and that was 40 years ago.
When I was in high school about 1984 one of my friend's father was a software engineer for some company. I asked him "Why don't they use CDs for storage?" He answered "That'll never happen. They might be okay for archival work, but that's pretty specialized. You'll never see them in a home computer."😅😂🤣😂😅😆
@@kaitlyn__L ReadyBoost, yeah that was an interesting concept, though really missing it's intended goal by a mile sadly... You needed a very fast USB flashdrive (expensive) for any potential benefit, at least back then well over $100 at which point you could just buy more real RAM for that kind of money... Even if you bought a high-end flashdrive the potential benefit was immediately wasted by the CPU spending a lot of time handling ReadyBoost... Funny concept, useless in reality unfortunately...
I suspect that if you look at those hubs in System Profiler, you'll probably find that each 'hub' is probably a 4-port hub internally with a bunch of other hubs daisy-chained off it, because 4-port USB hub controllers are cheap but bigger hubs controllers are not. That's why 7-port hubs are more common than 8-port ones. Your Mac may have some of its ports on a hub rather than directly connected to the root complex, depending on if you're using front or rear panel ports. Which means you could be three or four hubs deep. Complex USB trees with deeply nested hubs don't work very well at all, and soft RAID often has trouble telling the difference between a USB dropped packet and a catastrophic IO error because of how fragile soft RAID ends up being. So, what you probably really need, is USB PCI cards, not USB hubs.
Yep. And chances are that mac is running USB 1.1 and once the bus and IO overhead is added on he might be saturating the link speed with all of those drives. Splitting the drives onto a bunch of USB 2 root hubs would be ideal.
Interesting to see how far we have advanced in storage technology. 13 floppy drives barley scrape 1mb/s while nowadays a standard 20 dollar SSD can reach speeds up to 300-500 mb/s.
This is hilarious. I always wondered about this when I was younger. I was just getting into Linux and one of the fun projects young me had undertaking was doing a raid0 with both of my internal floppy drives together into a "megafloppy" and mounting it as a single drive. I thought it was the coolest thing watching them work in tandem.
@@charliekahn4205 raid0 is striping meaning your 1 virtual disk is written across all real disks. Removing one would be the equivalent to having a HDD and just taking a chunk out of it. That part of the data is gone. Some of the files might still be there if they're lucky enough to be only on the parts that are still there
You chose the correct file system, hah, I don't think floppies with their insanely slow seek speeds, like anything other than the simplest of file systems. Journaling on that would have been brutal.
While floppies are slow, I wonder how much overhead was in bus contention. I don't remember the average speed of reading from a floppy disk or what the performance would be for single usb floppy drives, but with so many drives on 1 port, there might be a lot of cases where drives are sitting on data waiting on the computer to get to them while fetching data from other drives, etc.
I don’t think that’s an issue. I still use floppies sometimes for text documents and it can take several seconds to copy a document that is only a few kilobytes
Wow. It's pretty rare that I learn something new from a Retro RUclips video (not your personal Action Retro). But this is useful everywhere - even in modern systems. Well Done and Continued Success!
Reminds me of an office experiment we did in a startup back in 2001. We each created an NBD (network block device) on our Linux desktops and shared to a central PC that created RAID5 array from them. That was then mounted and shared as an NFS drive 😂
There was a product in the late 90s called "Mango Medley" that did this for PCs. You installed the app on a bunch of networked PCs and it made the pooled spare space appear as a single network drive. They never really got NT compatibility going though.
Made a huge mistake and tried to watch this during lunch. I was laughing so much I couldn't eat. If you get all 34 working you should try setting up a BBS on them. Also there's no such thing as too many floppy disks. Just ask the Floppotron!
One possible reason they didn't all work together is differences in chipsets between the drives. They might be fine individually, but whatever setup MacOS uses for software RAID support might not like some chipset combinations.
RAID only in the sense that the data is striped, but there doesn't seem to be much speed improvement, at least on writes. Perhaps due to the way the USB bus handles the drives, and their slow response times (like, each drive may be holding the bus while it takes action). Those speeds really seem to be about the same as a single drive, or maybe a couple of drives. If you had multiple drives that were definitely on separate root ports, then you could be assured no one drive was holding up communication with another. If those hubs were all on the same root port, or only on two, then that could explain the speeds. Also as you write more data to each disk, throughput will go down as the head moves to the inner tracks. It would have been good to compare speed of a single drive to the RAID, and cool if there was an update with each drive plugged directly into a port on the Mac instead of using the hub. Then you'd see if there's a tradeoff for speed versus capacity. The suggestion of ZIP drives would be fun too.
Some drives may have failed due to the response times. The software RAID driver may have timed out particular drives that didn't spin up fast enough or move the heads fast enough or something like that, just like if a hard drive failed to respond in time. A full stack of matched drives would be the best test. Stick a bunch of multi-port PCIe USB controllers into the computer so every drive has its own dedicated root port (have to make sure of how many real root ports there are on each card). Man, I have time for stupid things like this now, but no money to do them.
12:15 *Queens a pawn* "Oh maybe I'm great at chess! Heheheh" *Rook immediately takes new queen* "... ... ..." "Alright welll it *does* work" *Exits the game* AMAZING
IIRC a standard floppy drive is 500 kilobits per second maximum transfer, so theoretically a USB 1 bus wouldn't be the bottleneck until 24 drives. Given actual transfer limitations and stuff the bus and the drives themselves are probably about equally loaded in the 15 drive setup.
@Sean B You're thinking that the hubs will communicate with each drive through USB 2 and then communicate all that data back to the computer over USB 3? Sorry to say that's not how it works. A USB 3 port actually carries USB 3 and USB 2 at the same time, electrically isolated from eachother. Whatever is sent over USB 2 lines will continue along USB 2 lines all the way back to the computer.
Even with USB1 devices on a USB2 hub, you need a hub with multiple transaction translators, some are single-TT and aggregate all USB1 traffic as a single USB1 stream before translating, where a multi-TT hub can pass USB1 streams in tandem
The fact that you drive an XJ is the sole reason I am subscribing. Mine is the same color just a lot more modified. And the fact that I once did something similar with a compact flash card, a USB drive and a rewritable CD, don’t know how we got it to work but it was just for fun.
I literally had a thought earlier this week of how crazy it would be to have a fileserver/webserver raid using floppies. I love seeing this concept working!
It's dosen't mean, that you have to do it still "for server".. At least you heard when NSA searching your rid "what the f.. is this, it broke our tools" And if you have r0 and you took one disk with you, no one go play your doom on that raid..
Can you imagine if floppy tech was still standard and Google's data centers just emitted insane dbs of clicking noises. 🤣 I would be interested in seeing a raid array of those modern tape drives that places use for massive backups
Creating RAID 0 works fine of course. The real fun begins when you configure RAID 5, RAID 6, or maybe RAID 10. And then remove one of the drives and see how it behaves. And readd the drive and let it automatically rebuild the array. I did this with a bunch of USB stick and software RAID in Linux to learn how mdadm works. Easy to setup and great fun.
Since the PowerMac G4 has a separate USB bus for each port I would be interested to see if putting half on one port and half on another would help with IO contention any.
i cannot believe that a floppy raid became more popular in a month than running the best-selling game of all time on one of the most unique computers from one of the most popular computer hardware and software companies over the span of 21 months. ya did good sean, ya did good.
I love seeing these experiments. It's just the kind of stuff I liked trying out myself. You're living the dream, man! Keep being you and having fun with anything you do.
You're the reason I bought the usb hub. and god damn, I though an $80 usb hub is a bit expensive for cable management under the desk, but I will never go back. by far the best investment for my computer. Great video. came back to like it.
I have been using RAID 0, 1, and 5 for 2 decades in various applications with hard disks and solid-state storage, but it never occurred to me to RAID floppies.
My thoughts too. Avoiding that source of failure would require formatting each floppy individually before starting the experiment, but the video did not say anything about that.
Laugh all we want, but the cool thing about RAIDing all of those floppies is the cool, robotic 60/70s computer sound they make - almost reminds me of the Lost in Space 60s robot. One time I had to duplicate dozens of USB sticks for a trade show and the bank of red USB sticks lit up like a Christmas tree, but I always wanted to RAID USB sticks, especially the kind that light up brightly when accessed.
This is a perfect example of adhering to the ethos of the Weird Al song, “Dare to be Stupid” in retro tech fashion! Perhaps I missed where you stated how you ended up with that many USB floppy disk drives: pure madness! 🤣🤣🤣
Back in the early 1990s I briefly worked with a guy Ken Ouchi who patented RAID back in 1976. I got hired in the early 90s to write microcode for an enterprise-level rack-mounted storage subsystem. About 60 days after I was hired, our project was cancelled. I and others were moved to the competing project within the company. Mr. Ouchi quit to become CTO at some smaller company. I do not think he ever intended RAID to be used on floppy disks.
I remember that article from 20 years ago! I tried doing the same thing myself, first with 6 x LS120 SCSI drives and then with USB floppy disks. I could only manage 4 though...
Have you thought of windows storage spaces? In theory USB drives can be added to the pool too once the drive is converted from "removable" type to "fixed".
![ARGENTINA vs. PERÚ [1-0] | RESUMEN | ELIMINATORIAS SUDAMERICANAS | FECHA 12](http://i.ytimg.com/vi/xtI25ENDDSY/mqdefault.jpg)
I believe that a floppy-based raid 0 configuration was referenced in Dante's Inferno as an object one would encounter in the 5th circle of hell (wrath), if I'm not mistaken.
@@BrianMoore-uk6js R
Well, it might be at least 2 circles, as gluttony is kind of synonymous to 34 USB FDD drives :p
If I remember from Office Space the correct peripheral for the circle of Wrath is a laser printer.
@@Inject0r At least USB FDD can reliably load.
@@TimLesher "PC LOAD LETTER"? The fuck does that mean?
Now do it with ZIP disks! I did it once back in college on an eMac while bored at the help desk, worked great with five zip drives!
Zip disks would be perfect for this.
Jazz disks would also be cool
@@kirishima638 perfect for the cursed mac!!!
Or how about lto-drives, isn't too easy job at least..
@@timmooney7528 I did it with 3 1Gb Iomega Jaz SCSI dived in Raid 0 for backups on my Linux home server. Worked fine until it stopped working. :-/
Back than i wished you could make CD-Rom RAID for backups. I tried it, but my knowledge how to make such a thing, and make it with my available hard drives wasn't there.
I later made a Raid 6 with 7 1Gb Iomega Jaz SCSI drives. I had to put them in their own enclosure with an dedicated SCSI card. It worked nicely, until i ran out of disk space ( 5Gb isn't that much anymore :-/ ).
This is reminding me of that old video by "The iBookGuy". He put several USB flash drives in RAID, then later changed his name to "8-Bit Guy" to hide the shame.
"to hide the shame" 🤣
if only that was the reason he changed his name.
Possibly the reason the RAID array failed, the heads on those drives may need to be cleaned. They used to sell "floppy drive cleaners" .. basically a floppy disk with a cleaning pad instead of a magnetic storage medium. Put a couple drops of alcohol on the pad, put the "disk" in and try to mount it in OSX. Do this a couple times and it should clean the heads of the floppy drive possibly increasing your success rate.
I feel a redo upcoming.🤭
those cleaning disks are the worste.
the cleaning pad is also inside the floppydisk itself to protect the magnetic disk.
and will collect crap during its life..
back when where still booting of disks it whas usualy the most dirty and teh first one to go.
even if you use fluids it not going tobe enough and doing it while powered on is also not such a asuring idea.
just popoff teh front plastic bit of the drive and clean it with a cotton tip and some alcohol. alcohol will just do you dont need to desolve annything else but the crap
However i think its had more todo with the controler type of the usb floppy
it whas distict a sertian type that failed. but seem to work.
like its 2.8mb able or even zipdrive formats like 120mb using difrent drivers that dont have the same hardware function
i think the same fallback legacy driver for all the drives will make them behave the sameway and will allow raid
There's another problem, that doesn't have to do with any one drive, but trying to use them all at once. which is going to be the case when your using raid. so I looked up a usb floppy drive and it stated that they use 500ma (or 0.5a) at 5v and your powering them with two strips. so to see how much power these drives are going to use when they are all working in tandem take the power use of a single one and multiply it by 17 (half of 34 since the other half is on a different hub). when you do the math that comes out to 8.5 amps per hub. Convert that to watts and that hub needs to be able to supply over 40 watts of power (42.5w) now a quick search of powered usb hubs showed various levels that max out at though 36watt was common. if that's the case with your hubs then while you'll get the drives to all connect when you start them up slowly like that as only one drive is active at a time and the rest are in standby so not using near as much power and thus not overloading the hubs power supply. but when they all go to active read or worse write mode at the same time some of them are not going to get enough power and thus fail and like you said if one of the drives fails it brings down the whole ship. I was wondering about transfer rate because older computer. As it would overwhelm a 1.0 usb port. But turns out even 1.1 should have the capacity as we are looking between 5-8mb/s and 1.1 can handle 12mb/s. I think that computer you mentioned has 1.1 so while the data side should be okay. I say should because it looks like the max practical speed of a floppy is 100-250kb/s though they are rated at 500kb/s-1mb/s but even so it would more likely just run slower then it could. though that would depend on how well the raid controller is able to handle the likely out of order nature of the writes and reads coming in if it was to get up to the full speed and not just 250kb/s speed. Still make sure that the hubs are rated for the power drain that's going to hit it as that 8.5 amps or 42.5watts is only from 17 of those 34 drives doing a write cycle at the same time. this is what I would check first, because if your powered usb hubs ac adapter is not outputting at least 43watts or 8.6 amps (as the hub itself is also going to need some power) the rest doesn't matter there will always be at least one drive that is going to fail due to lack of power during the read/write cycles even if all the drives and disks are in perfect working order. And like you said it only takes one to crash the array.
@@CrossRoadsOfTime this is software RAID, so the OS is the controller, but aside from that yeah you’re exactly on the money. Needs more hubs probably. Could also be a mismatching head seek latency or something between a few models of drive. (Or even the former power issue directly causing the latter issue.)
@@kaitlyn__L never said it was a hardware raid setup. even with software raid, there is still a Controller managing the RAID setup. it's just that in his case the controller is a program running in the OS. vs a program running on dedicated hardware.
Just sad that Retro forgot to take the opportunity to name his floppy raid: Shadow Legends.
Also love how the chess game takes so long to load, that it feels like he managed to ressurect the Cassete tape era of games, where games would take alledgedly 113827091327312907321987 hours to load.... because they were stored in a freaking Cassete tape...
Imagine if Paweł Zadrożniak did this with his 512-disk musical instrument the Floppotron.
Old video
@@DonMr He’s made a new one actually.
@@aswarmofdeadinsects when?????
@@DonMr few weeks ago
Old one was only 64
the Guinness world record should add a record for "Most number of floppy drives RAIDed". they already do have a lot of ridiculous records so this one should slide easily through...
You have to pay to get your records witnessed, which isn't cheap at all. They might not even find it worthwhile to certify, since only nerds would be interested not the people who buy the Guiness book.
People who buy those books are not too different from us
@@3rdalbum This guy. Other than nerds, who buys Guinness record books? That was unintentionally hilarious!
Since each floppy disk sector is 512bytes, the maximum amount of drives would be 2880 for peak throughput. That's reading the entire contents of each drive in one go anywhere in the raid. However, seek time is quite high, at over 500ms. As you add more drives, the seek time is eventually going to out weigh the throughput transfer. Given the fastest transfer would be around 250Kb/s from the inner track of a high density disk, this means we're looking at exponential loses from the raid after around 7.5 drives at 300rpm standard speed. It would be interesting to see you bench mark the drives as more are added/removed to see how that plays out.
I think you're mistaken on some things. I don't see why the number of sectors per drive would have anything to do with sequential throughput. The only limiting factors would be the bus and computer speed.
@@eDoc2020 There was a typo in that message. They meant with 512b sectors, a raid array of 2880 floppies would mean you could theoretically read an entire floppy's worth of data anywhere on the array every cycle.
I don’t get this either. What does sector size have to do with anything? The advantage to RAID-0 is having I/O operations in flight simultaneously. At some point, the I/O ops will complete before the next can be scheduled, so there’s no point adding any more drives. Is that at 2880? I don’t know.
The latency doesn’t quite make sense to me either. For sequential reads, the latency of the initial seek will overlap as each drive navigates to the start of the data chunk, but you don’t need to wait for the first to complete before asking the next to start seeking, so it just ends up being a fraction longer than a single seek to have them all positioned, at which point data is already being read by the first drive, so you’re really only limited to the seek time of one drive.
Random non-consecutive sectors would be a different matter, but so would they be on a single drive.
Edit: Sorry, that reply was written on my phone so I wanted to keep it short. I'll explain in a bit more detail now.
Sector size is how one calculates the maximum theoretical disks that can be added to an array before you yield zero gains.(There are other factors that play into this so you obviously won't reach anywhere near that amount before you encounter 0 or even potentially negative returns, so don't take this at face value)
"At some point, the I/O ops will complete before the next can be scheduled, so there’s no point adding any more drives. Is that at 2880? I don’t know."
The access time; in this case, the time it takes for the track to make an entire rotation would be the limiting factor, not the speed at which one can process the data.
Seek and Access times are what limit you from theoretical max throughput which was what the OP was factoring into his/her theoretical calculations.
Reading 1 sector per disk per drive per rotation of a track sequentially would be the ideal condition and thus would be limited to a drive's RPM and to a much lesser degree, the time it takes for the read head to move over a single track(like you said, the added time added per additional drive is a tiny fraction above what it would take one drive); I think at this RPM the "seek" time is a non-factor as access time is already the bottleneck.
@@nickwallette6201 The 2880 drives was maximum theoretical throughput if seek time and physics weren't a thing. Seek(or read head travel and disk RPM) time becomes the limiting factor for fragmented data, and to a lesser degree for sequentially written data.... the laws of physics are still a thing here.
Just because data is written sequentially doesn't mean it will all be read in a single pass of the disk. In fact data is not read that way from a spinning disk which is why RPM of a disk matters.
I hope this clears things up a bit
@@kefler187 Not yet, but I think I may have to think upon this for a little bit.
I am making some assumptions that may not be true, such as that a drive could read an entire _track_ in one go, without having to wait for the the disk to rotate through an entire revolution between sector 1 and sector 2 of a given track. It is my understanding that _not_ being able to do this is why interleave was a thing back in the early days of hard disks, so I assume (?) that it is not a problem with floppy drives at a rather pedestrian 300 RPM. But I could be wrong.
My other assumption was that, once you have the first drive seek to the appropriate track (seek time) and then wait for the disk to rotate to the desired sector (access time), you can then essentially start streaming that entire track to the host, assuming the host wants enough sectors of consecutive data that even with a one-sector-per-drive stripe, the data request would require reading adjacent sectors on every drive.
If that's the case, then with a sufficient number of drives, you could, in theory, paper over the seek and access times of having to jump to the next track, if there are enough drives that you are still receiving data from one drive while another is seeking to the next track.
At some point, the CPU load of dividing the chunk into individual I/O requests and allocating buffers for the incoming data will overcome the throughput advantage of adding another drive, but I don't yet understand how that has anything to do with sector sizes (other than smaller sectors require more overhead, but it wouldn't be a fixed number of drives that saturates a host's ability to process them in a timely way), nor how the RPM is really a factor, other than of course being a limiting factor in the individual drive throughput, and of course, access time, which can also be covered up by staggering I/O requests (which a single-threaded host would have to do anyway, to process them in serial fashion; and a multi-threaded host could only do per number of cores / CPUs.)
I'll chew on this, though. It's interesting to think about there being some calculated limit, but I don't yet see how that is.
"Should be fine" "I'm not an expert" he says sitting next to 30+ floppy drives
So sitting next to the drives makes him an expert? Niiice.
@@eriklarson9137 no, it's the "should be fine" part that's problematic here
Reminds me of the time I asked my high school electronics teacher if it was possible to use a floppy disk as RAM, to which he answered: "only if you want to have the world's slowest computer!"
Yep, but it works.. even today. Some bluray-drives could work softram. It's not even "worsest" idea what you could find.
Tape drives aren't so easy peacy put in ride and need whole custom own raid program. And cassette library with few drives and hundreds of LTO's.. no no no.. only way if you modify drives and made endless tape between drives and could buffer it somehow (like old ink casettes) and run both way ofc.. that must be weirdest "raid", with one media and many raid drives.. ;))
So, think about that.. USB sticks raid with good sticks isn't even stupid idea if you have enough usb3 channels to cpu.. you don't want use usb3 splitters with flash, it's turns same fast as this floppedraid... (:
@@jannejohansson3383 Can't use LTO for RAM without destroying it by shoe polishing. Tape is meant for streaming data in one go, not for random access memory where the tape needs to go back and forth constantly, it would break within days.
USB sticks as 'half RAM' never properly worked, at best it can give a placebo effect in my experience, either the flash chip is too slow to make a difference or the CPU spends way too much time running it...
Take the original Commodore 64, ~1Mhz, will require RAM capable of just over 2MB/s at 8 bits at a time and that was 40 years ago.
When I was in high school about 1984 one of my friend's father was a software engineer for some company. I asked him "Why don't they use CDs for storage?" He answered "That'll never happen. They might be okay for archival work, but that's pretty specialized. You'll never see them in a home computer."😅😂🤣😂😅😆
@@someguy4915 yeah, USB sticks for “Vista Speedup” (or whatever it was called) never gave me performance benefit back in the day.
@@kaitlyn__L ReadyBoost, yeah that was an interesting concept, though really missing it's intended goal by a mile sadly...
You needed a very fast USB flashdrive (expensive) for any potential benefit, at least back then well over $100 at which point you could just buy more real RAM for that kind of money...
Even if you bought a high-end flashdrive the potential benefit was immediately wasted by the CPU spending a lot of time handling ReadyBoost...
Funny concept, useless in reality unfortunately...
Well done! I am the guy you mentioned who made the original FDD USB Raid. So funny and awesome to see this video.
I'd love to see a RAID 5 or 6 with these - it would be so easy to replace a failed disk!
See how many magnets you can attach to the system before they can't recover
Easy, but do you realize how much time you would spend replacing failed floppy disks if they got used much?
I suspect that if you look at those hubs in System Profiler, you'll probably find that each 'hub' is probably a 4-port hub internally with a bunch of other hubs daisy-chained off it, because 4-port USB hub controllers are cheap but bigger hubs controllers are not. That's why 7-port hubs are more common than 8-port ones. Your Mac may have some of its ports on a hub rather than directly connected to the root complex, depending on if you're using front or rear panel ports. Which means you could be three or four hubs deep. Complex USB trees with deeply nested hubs don't work very well at all, and soft RAID often has trouble telling the difference between a USB dropped packet and a catastrophic IO error because of how fragile soft RAID ends up being.
So, what you probably really need, is USB PCI cards, not USB hubs.
well, that answers a question i had concerning USB hubs, generally...
Yep. And chances are that mac is running USB 1.1 and once the bus and IO overhead is added on he might be saturating the link speed with all of those drives. Splitting the drives onto a bunch of USB 2 root hubs would be ideal.
You should try a DVD-RAM raid. It would store more data and still be utterly ridiculous.
Interesting to see how far we have advanced in storage technology. 13 floppy drives barley scrape 1mb/s while nowadays a standard 20 dollar SSD can reach speeds up to 300-500 mb/s.
This is hilarious. I always wondered about this when I was younger. I was just getting into Linux and one of the fun projects young me had undertaking was doing a raid0 with both of my internal floppy drives together into a "megafloppy" and mounting it as a single drive. I thought it was the coolest thing watching them work in tandem.
What happens when you eject one of the disks? Is RAID formatted only to the disk, or does it always RAID together whatever's in each drive?
@@charliekahn4205 raid0 is striping meaning your 1 virtual disk is written across all real disks. Removing one would be the equivalent to having a HDD and just taking a chunk out of it. That part of the data is gone. Some of the files might still be there if they're lucky enough to be only on the parts that are still there
You chose the correct file system, hah, I don't think floppies with their insanely slow seek speeds, like anything other than the simplest of file systems. Journaling on that would have been brutal.
While floppies are slow, I wonder how much overhead was in bus contention. I don't remember the average speed of reading from a floppy disk or what the performance would be for single usb floppy drives, but with so many drives on 1 port, there might be a lot of cases where drives are sitting on data waiting on the computer to get to them while fetching data from other drives, etc.
A USB port is still much faster than all of these floppies combined
Oh wait this isn't usb2 yet, yeah it might be bottlenecking it a bit
@@max_uaminecraft1827 the issue wouldn't be with total theoretical throughput but with very inefficient use of the bus.
Same thought. This would be more interesting on something like a mining board with a floppy per USB card per PCIe port.
Add to that, make them LS120s.
I don’t think that’s an issue. I still use floppies sometimes for text documents and it can take several seconds to copy a document that is only a few kilobytes
Wow. It's pretty rare that I learn something new from a Retro RUclips video (not your personal Action Retro). But this is useful everywhere - even in modern systems. Well Done and Continued Success!
Just at work today we were discussing how to connect a load of SD cards to act like a backup tape set.
Raid 0, sounds great! Use unreliable media to make a fault prone storage mechanism. Super fun
Dude did what a point and click adventure protag does and tried one item on a bunch of stuff before finding the solution
If he keeps doing stuff like this it’s only a matter of time before he takes over the world with a sentient Macintosh
It has to be a 68k tho
Skynet has to start somewhere .
@@jackcrawford304 no a apple ll
developes a ghost in the shell...
@@halnor777 didn't the terminators have 6504 chip in them somewhere?
Y'know, they used to call me megafloppy back in college
"and if you enjoy the sound of scores of floppy drives screaming in unison" lmao floppy drives are literally a musical instrument!
Attached them to an Unraid server using the 'unassigned devices' plugin, and you can easily build an array out of them.
Awesome, Came just to hear them all writing and reading
Reminds me of an office experiment we did in a startup back in 2001. We each created an NBD (network block device) on our Linux desktops and shared to a central PC that created RAID5 array from them. That was then mounted and shared as an NFS drive 😂
There was a product in the late 90s called "Mango Medley" that did this for PCs. You installed the app on a bunch of networked PCs and it made the pooled spare space appear as a single network drive.
They never really got NT compatibility going though.
@@Insightfill oh, that’s very cool
Made a huge mistake and tried to watch this during lunch. I was laughing so much I couldn't eat. If you get all 34 working you should try setting up a BBS on them.
Also there's no such thing as too many floppy disks. Just ask the Floppotron!
One possible reason they didn't all work together is differences in chipsets between the drives. They might be fine individually, but whatever setup MacOS uses for software RAID support might not like some chipset combinations.
not the content we wanted, but the content we needed
RAID only in the sense that the data is striped, but there doesn't seem to be much speed improvement, at least on writes. Perhaps due to the way the USB bus handles the drives, and their slow response times (like, each drive may be holding the bus while it takes action). Those speeds really seem to be about the same as a single drive, or maybe a couple of drives. If you had multiple drives that were definitely on separate root ports, then you could be assured no one drive was holding up communication with another. If those hubs were all on the same root port, or only on two, then that could explain the speeds. Also as you write more data to each disk, throughput will go down as the head moves to the inner tracks. It would have been good to compare speed of a single drive to the RAID, and cool if there was an update with each drive plugged directly into a port on the Mac instead of using the hub. Then you'd see if there's a tradeoff for speed versus capacity. The suggestion of ZIP drives would be fun too.
Some drives may have failed due to the response times. The software RAID driver may have timed out particular drives that didn't spin up fast enough or move the heads fast enough or something like that, just like if a hard drive failed to respond in time. A full stack of matched drives would be the best test. Stick a bunch of multi-port PCIe USB controllers into the computer so every drive has its own dedicated root port (have to make sure of how many real root ports there are on each card). Man, I have time for stupid things like this now, but no money to do them.
5:40 beep - the 34 usb floppy drives themselves
YOU ARE MY HERO !
How do I install an actuall physical cassette drive to a modern PC ?
It might also be the case that the drive is fine but the disk is bad. A lot of my used floppies have problems nowadays.
12:15 *Queens a pawn*
"Oh maybe I'm great at chess! Heheheh"
*Rook immediately takes new queen*
"... ... ..."
"Alright welll it *does* work"
*Exits the game*
AMAZING
love to see how it affects marathon
Kind of a weird homage (or pre-sequel as Borderlands would have it) to the Floppotron 3.0...
Your silly side is creeping in to these videos and I couldn't be happier!
Since you got up around 1MB/s moving Chess, I wonder if USB 1.1 was a limiting factor? Should try this on a G5 with USB 2.0!
IIRC a standard floppy drive is 500 kilobits per second maximum transfer, so theoretically a USB 1 bus wouldn't be the bottleneck until 24 drives. Given actual transfer limitations and stuff the bus and the drives themselves are probably about equally loaded in the 15 drive setup.
@Sean B You're thinking that the hubs will communicate with each drive through USB 2 and then communicate all that data back to the computer over USB 3?
Sorry to say that's not how it works. A USB 3 port actually carries USB 3 and USB 2 at the same time, electrically isolated from eachother. Whatever is sent over USB 2 lines will continue along USB 2 lines all the way back to the computer.
Even with USB1 devices on a USB2 hub, you need a hub with multiple transaction translators, some are single-TT and aggregate all USB1 traffic as a single USB1 stream before translating, where a multi-TT hub can pass USB1 streams in tandem
Is it just a coincedence this video popped up on my recomendations? I bought 4 (sadly only 3 worked) floppy disks and they came less than a week ago
How funny. Ive got a 5 drive 30 tb raid in the works
The fact that you drive an XJ is the sole reason I am subscribing. Mine is the same color just a lot more modified. And the fact that I once did something similar with a compact flash card, a USB drive and a rewritable CD, don’t know how we got it to work but it was just for fun.
I literally had a thought earlier this week of how crazy it would be to have a fileserver/webserver raid using floppies. I love seeing this concept working!
It's dosen't mean, that you have to do it still "for server"..
At least you heard when NSA searching your rid "what the f.. is this, it broke our tools"
And if you have r0 and you took one disk with you, no one go play your doom on that raid..
Can you imagine if floppy tech was still standard and Google's data centers just emitted insane dbs of clicking noises. 🤣
I would be interested in seeing a raid array of those modern tape drives that places use for massive backups
You're describing the Air Force
you should stream a video file from the floppy, this is something that would have never worked from a single floppy
First thought "Why?". And then realised which channel I was watching. "Because Action Retro....."
Alright, now do this with CD-RW drives. I dare you.
now THIS is real journalism
This is so friggen ridiculous. I love it. LOL
Idea: Try RAIDing all 30 drives again but with JBOD. Maybe the ones that wouldn't play nice with RAID0 would work then?
Creating RAID 0 works fine of course. The real fun begins when you configure RAID 5, RAID 6, or maybe RAID 10. And then remove one of the drives and see how it behaves. And readd the drive and let it automatically rebuild the array. I did this with a bunch of USB stick and software RAID in Linux to learn how mdadm works. Easy to setup and great fun.
i do this with those cheapo flash drives they give out at promos
.... so. A ' floppy centipede '
The only shame i have is not finding this kind of tech vore before. You are exceptionally creative and it's awesome.
”Sponsored by RAID Shadow Legends”
Now question is if you can move all that floppy raid to another pc and reassemble it there without data loss.
Since the PowerMac G4 has a separate USB bus for each port I would be interested to see if putting half on one port and half on another would help with IO contention any.
i cannot believe that a floppy raid became more popular in a month than running the best-selling game of all time on one of the most unique computers from one of the most popular computer hardware and software companies over the span of 21 months. ya did good sean, ya did good.
I love seeing these experiments. It's just the kind of stuff I liked trying out myself. You're living the dream, man! Keep being you and having fun with anything you do.
This is my kind of computer shenanigans, almost like age old question "how many Tandies in a node would it take to run Crysis?"
This is the mad science level stuff. Love it
You're the reason I bought the usb hub. and god damn, I though an $80 usb hub is a bit expensive for cable management under the desk, but I will never go back. by far the best investment for my computer. Great video. came back to like it.
I have been using RAID 0, 1, and 5 for 2 decades in various applications with hard disks and solid-state storage, but it never occurred to me to RAID floppies.
I'm sure it's crossed your mind that some of the drives were failing due to bad diskettes and not the drives themselves, nevertheless great video.
My thoughts too. Avoiding that source of failure would require formatting each floppy individually before starting the experiment, but the video did not say anything about that.
You should use a bunch of gotek floppy emulators with external ssd instead of a usb drive🎉
Do you have an explanation why so many floppy drives refused to raid?
> Ridiculous Array of Inexpensive Floppies
> R.A.I.F.
Brain: He found da raif.
Me: Shut.
This man is terrifyingly powerful. He put an RTX in a G5, he built a RAID of Floppy Disks, and so much more
XJ Cherokee is the Powerbook of Cars
Thank you for having a perfectly designed and short intro sequence.
Even though this is a retro theme, I still can't but help think of the Mac G4 as being futuristic. I mean.. what a stylish beast.
Laugh all we want, but the cool thing about RAIDing all of those floppies is the cool, robotic 60/70s computer sound they make - almost reminds me of the Lost in Space 60s robot. One time I had to duplicate dozens of USB sticks for a trade show and the bank of red USB sticks lit up like a Christmas tree, but I always wanted to RAID USB sticks, especially the kind that light up brightly when accessed.
Why does this put a big fat happy smile on my face? 😂
I miss playing with my old Macs. I love them.
Zip disks be like: Look what they need to mimic a fraction of our power
This is a perfect example of adhering to the ethos of the Weird Al song, “Dare to be Stupid” in retro tech fashion!
Perhaps I missed where you stated how you ended up with that many USB floppy disk drives: pure madness! 🤣🤣🤣
Back in the early 1990s I briefly worked with a guy Ken Ouchi who patented RAID back in 1976. I got hired in the early 90s to write microcode for an enterprise-level rack-mounted storage subsystem. About 60 days after I was hired, our project was cancelled. I and others were moved to the competing project within the company. Mr. Ouchi quit to become CTO at some smaller company. I do not think he ever intended RAID to be used on floppy disks.
1GB zip drive RAID next time??
Nice XJ :)
Hahaha thanks
I didn’t realise quite how much I would enjoy this!
Looks awesome using black floppy drives with colorful disks
the sound of that thing is magical. it's like bladerunner
"Totally normal *chuckle*" sums this up perfectly xD
This has to be one of the most amazing things I have ever seen done with a computer.
We did it, guys. We found the Data Floppotron.
Interesting. I think you should try it with lots of 240 Megabyte Floppy's all of the same brand. :-)
Thanks for doing this so I didn’t have to.
If you’re French and English bilingual, the phrase Floppy Raid is funny… in French raide means stiff.
this is the kind of stuff i’d probably do if i had the money and had the idea. this is a terrifying idea and i love it
5:56 That Blue and Gray Translucent looking Floppy looks pretty cool
Absolutely useless outside the fact that this amused me to no end 🤣. Keep of the nerdy fun. You got a new subscriber.
I remember that article from 20 years ago! I tried doing the same thing myself, first with 6 x LS120 SCSI drives and then with USB floppy disks. I could only manage 4 though...
Completely unnecessary... I LOVE IT!!!
Have you thought of windows storage spaces? In theory USB drives can be added to the pool too once the drive is converted from "removable" type to "fixed".
Ah yes, the beauty of RAID.
Next thing you know someone's gonna RAID zipdisks together.
...Shit. I just gave out the idea.
that would be the most unreliable RAID on Earth and most annoying one once many drives start to perform the click of death
@@alerey4363 Even better.
Great! Now do it all again with those floppy-replacement options. Zip, LS-120, and CD-RW.
this is the type of shenanigans I signed up for!
Mother of God just the title. Dude - I subbed this week and your channel is now my new fav tech. The 2GS VM, the G5 streamer.. my god now this
It’s a real shame that you don’t have 100k subs, LETS GETY HIM THERE
The video in itself isn't what made me subscribe but it certainly helped. Your enthusiasm got me to sub lol.
Sometimes the things that result when the answer to the question "Why?" Is "Because I can" are a ton of useless fun!