The Dual-Drive IDE Hell | Hacker News Reader

The Dual-Drive IDE Hell

os2museum.com

92 pointsuserbinator71 comments

Oh yeah. IDE Master/Slave was pure voodoo. With hard drives it usually wasn't too bad, but ATAPI devices (CD-ROMs) were just insane. I saw everything from drives that wouldn't be recognized unless they were slave to a HDD, to drives that refused every configuration except being a slave drive with no master--which isn't supposed to work at all.

Cable select sometimes worked, but I didn't trust it as much as the jumpers. I did try to match the jumpers with the position on the cable just to give myself a better chance of success.

The article is correct that mixing drive brands, or even mixing drive eras was asking for trouble. Another fun aspect is when someone wanted to install more than 4 drives and had to use one of the ISA (later PCI) PATA cards. Those cards _never_ supported ATAPI devices in my experience, so there were some machines that booted off of a HDD hanging off of an ISA card so the 4 burners could be on the system controller.

And don't even get me started on the godforsaken CMD640 that every damn manufacturer used for years.

By the time ATAPI was around (instead of proprietary CD-ROMs), all of my issues with 2 drives had gone away. You had to jumper them right (cable-select was always hit-or-miss for me, right up until SATA), but if you did that they worked just fine for me.

> so the 4 burners could be on the system controller

Oh those were the days, in which poor students financed their high-end IT equipment by selling "backup copies" of popular music albums, games and other software products on the school yard to their peers...

I found that, back in the day, simply buying Plextors was the simplest solution. For best results when burning, you'd want the optical drive to be the master device on the secondary IDE bus.

Of course IDE only allowed two devices on a bus, with generally only two buses. If you wanted more you had to go with (parallel) SCSI, which allowed up to 8 IDs (one taken up by the controller) with the 8-bit bus, and up to 16 IDs on the 16-bit controller:

* https://en.wikipedia.org/wiki/SCSI_high_byte_termination

Just don't forget to put the properly-ohmed termination at the correct location on the cable.

* https://en.wikipedia.org/wiki/Parallel_SCSI

There was folklore that said three terminations were needed to get things to work: two electrical, one on each end of the cable; and one termination of a rooster at midnight around five black candles.

I could be mistaken, but I seem to remember adding additional drives using ISA IDE controller cards.

Which does not contradict what I said:

> Of course IDE only allowed two devices on a bus, with generally only two buses.

Most motherboards only had two buses, which each bus have two devices. Adding another card would allow another 2x2 drives.

Adding a SCSI card would have allowed for another seven devices (later fifteen). Most folks though didn't need that many at home.

Also, SCSI drives were more expensive (but usually performed better), so most stuck with IDE’s limitations.

SCSI stuff was way too expensive for poor college kids.

Funnily enough ATAPI was just a slightly cut down version of SCSI run over ATA. There was little reason for the big SCSI price premium except that the vendors thought they could get more money out of businessmen and Mac users.

SCSI drives were faster because higher RPM devices were available. 10 kRPM and 15 kRPM (!!!) SCSI drives were a thing, but I don't think there were many (any?) IDE drives that spun that fast.

WD's 10 kRPM drive made quite the splash when it was announced, IIRC:

* https://en.wikipedia.org/wiki/Western_Digital_Raptor

Whether a home user "needed" that kind of speed was quite another thing.

Most businesses didn't need those fast drives either. Marginal disk speed increase did not contributing to the bottom line.

I mean if you're going to disk, you've already lost money.

There was also more queue depth, allowing scsi drives to better order reads.

And hot swapping was very trouesome with IDE (not to be mistaken for newer eide), where as scsi was much better at hot swapping early on.

SCSI also had more bandwith than ide, and scsi2 was around before I could get eide/dma5?(80 conductor) cables/drives.

Man, memories.

If there’s a contradiction it’s that it wasn’t bread versus cake. It was bread or nothing.

IDE cards were under $10 at times. SCSI devices were made from unobtainium.

It's also worth remembering that several sound card models used to have an extra IDE slot, although they generally supported CD-ROM drives and not hard disks.

> There was folklore that said three terminations were needed to get things to work: two electrical, one on each end of the cable; and one termination of a rooster at midnight around five black candles.

I count five terminations and three kinds of terminations.

A consultant I worked with claimed that a dog toy looking like a chicken would suffice - at least for early VMware upgrades but I suspect this has been going on for longer.

They'd even stitch it together and put it back on the shelf for next time.

He was very competent in a number of other things (database design and Java) so I have just taken his word for it.

The deity I worship cares less about chickens and more about behavior these days[1] so I cannot verify.

[1]: actually never actually did if my understanding is correct)

Maybe I started to use computers too late (my 1st HDD supported UDMA/33 or 66 AFAIR), but I don’t remember any problems with master/slave. Just set jumpers and it works.

Yeah, I am not doubting the author at all, but I built a lot of computers for myself and others in the 90s, and maybe I just don't remember, or maybe I always had two cables, one for each drive, but I distinctly remember setting jumpers and it just worked. That said, I was using mostly Linux, sometimes Windows, so maybe this was an OS/2 thing?

Something I do remember though, was having to get the blocks, cylinders, etc... manually when partitioning/formatting disks.

I guess I got lucky with the drives I used, which back then would have mostly been Western Digital with a splash of Seagate, maybe because I didn't typically mix vendors in an initial build I just didn't run into these issues.

Just a couple years earlier and it was more fraught, for sure. UMDA/33 was probably late '97 or early '98. I got a couple of oddball IDE drives from a friend back in about '94-- a 100MB-ish Xebec and a similarly-sized KALOK drive. I recall having difficult getting them working along side my Western Digital 250MB. I had a SCSI CD-ROM so I didn't have that mess to deal with and ended up running one of the oddball drives at a time on my secondary IDE channel.

The fun I don't miss was "disk overlays" and finally adding a Promise-branded ISA card with an option ROM to exceed the >1,024 cylinder (504MiB) "barrier". I still have that card knocking around somewhere, I think. It'd be fun to dump that ROM and see what it looked like.

The problem I remember was reconfiguring computers. There were the jumpers and changes to BIOS.

And drivers in Autoexec.bat and Config.sys for CDROM's.

If I wasn't systematic and taking notes, I could spend time thrashing round in circles through reboots and partial hardware disassembly.

Oh gosh, once I was upgrading a mid-tower clone with a second drive (170mb). It had a Cyrix 803087 chip that didn't seat tight in the motherboard and it would work fine on its side from gravity so I would button it up and the chip would loosen when the case was standing up and the system didn't boot and my notes didn't help and it was a production machine.

Figured it out around two in the morning.

Yeah, i never trusted cable select either - jumpers every time.

the worst was trying to find one when you inevitable lost a jumper....

Your comment brought back the nostalgia of often changing the jumper position and saving the day at friends place during computer troubles.

I was building PCs during this time. One thing I didn't see mentioned in TFA was that the 'master' (drive 0) MUST be located at the end of the cable. The 'slave' had to be in the middle.

The jumpers were easy to figure out and not as complicated as the author makes out. Contrary to the experience that na85 had; "Cable Select" _never_ worked for me.

Yes it was a time of chaos and frustration, but I learned more in that period then I did since then. Hard core hardware lessons.

Note: most IRQ jumpers and motherboard jumpers were the same size as the drive jumpers. I was never at a loss for jumpers, but I did manage to lose them a lot.

> The jumpers were easy to figure out and not as complicated as the author makes out. Contrary to the experience that na85 had; "Cable Select" _never_ worked for me.

I think every drive I encountered that was manufactured after the early 90s either had a printed diagram for the jumpers somewhere on it, or else had labels molded right in to the plastic case around the pins themselves. Only time it was hard, back when I used to tinker with computers a lot, was when I'd be messing with a junked 286 or something. Further, by the late 90s IIRC drives had begun to converge on a standard jumper layout.

Same, cable select was not worth messing with as it so rarely worked. Just set master/slave and be done with it. I agree with the other poster also that said CDROMS were the worst full stop.

> One thing I didn't see mentioned in TFA was that the 'master' (drive 0) MUST be located at the end of the cable.

If that's not proof that Stack Overflow has helped me learn arcana, I don't know what is. In fifteen-plus years of working with IDE drives, I never figured out this pattern and typically just fumbled jumpers into some combination of master, slave, or cable select until all of the drives worked.

This definitely isn't my recollection of the '90s. IDE just worked. Yes, you had to get the cable and jumpers correct. But it wasn't rocket science. It pales in comparison to the headache of MFM. I remember having to use the DOS DEBUG.COM and communicate directly with the controller card to get a low-level format. And two ribbon cables per drive. I had two MFM drives in an XT. Lord. Ribbon cables every goddamn where.

> one of the most failure-prone and frustrating endeavors was trying to get two IDE drives working together on a single cable as a master/slave pair

Before the advent of pendrives and broadband internet this was how me and most of my friends shared files: we got our hard drives and took them with us to the home of one another. Setting up master/slave was easily done with jumpers on the hard drive and NEVER failed.

No iomega zip drive? CDR?

These were expensive for us.

I built a lot of PCs and never once had a problem with this. Set the jumpers, use an appropriate cable for the interface bandwidth, plug drives (HDD, ATAPI, whatever) in, enter BIOS, auto detect, done.

I am baffled that this gave others the slightest bit of difficulty.

It doesn't seem to be what this article is about, but 25 years ago I needed to have a hdd and a cd writer on the same IDE cable and I couldn't get it to work without windows blue-screening. It forced me to switch to linux and I never switched back (except for work and various other annoyances).

That is so bizarre, I built and tinkered with way too many machines in that era, and I only remember one combination that didn't work, two early ATAPI CDROMs on the same chain. Everything else, just get the jumpers right and it was good to go.

This was mostly in the era of ISA and then VLB "Super-I/O" cards, so maybe mobos with integrated IDE are pickier?

ATAPI was the third generation after EIDE. Though in fairness IDE was a lot easier than ST-506 or ESDI.

I built a 2000-era PC a few years ago, and it brought back memories of how much I hated using/configuring IDE back in the day. I've had IDE drives become flaky for no obvious reason.

SATA is such an advancement. For me, the only bad part about it is the mechanics of the cable itself (comes out too easy or too hard).

I was using IDE drives on x86 PCs from 1992 to around 2005.

I never had any real problem configuring them. I probably used about 20 different drives across 10 different machines.

Put the master at the end of the cable and the slave in the middle.

After the CS (Cable Select) jumpers and cables were popular it was even easier.

SCSI was more of a pain because even though you could have 7 devices (the 8th being the controller card) it became unreliable with more than 4 drives. Sometimes I would try to add a 5th and stuff would get erratic.

I also had problems when a SCSI tape drive was on the same channel as hard drives. We got 3 SCSI cards each with 2 channels for our Sun server. We put 4 drives on each channel except for one that had our 4 tape drives (QIC-something, 4mm, 8mm, DLT library)

I have never had a problem with this, and I was building and fixing machines through the entire life of the ide standard.

There were always quirks of one sort or another with any 2 pieces of hardware, including 2 ide drives, but I don't remember this being any more of a thing than say, some particular drive with some particular controller or motherboard, or any drive and any motherboard actually supporting their purported channel speed reliabley, etc.

Read the manuals, set the jumpers and cable placement and cable type appropriately, and if it ever didn't work, it was no more likely than any other mystery hardware problem like the mouse not working because of a shared irq or something.

This wasn't nearly as big of a problem as this makes it seem! There were 3 jumpers on IDE HDDs and CD-ROMs: master, slave, and cable select. Sometimes, there were also jumpers on the motherboard. And finally there were BIOS settings. If all of them are matched correctly, it just worked. Cable select wasn't a big problem if you actually RTFM from the HDD, CD-ROM, or even the PC. I obviously can't speak for everyone, but for the normal PC user upgrading their HDD or adding a new CD-ROM drive, it was not hard. When it didn't work, you could easily remove one drive and troubleshoot by removing or fiddling with the jumpers.

I've generally had few issues with IDE hard drives and CDROM drives, from the Pentium era forward. The "fun" has been dealing with hard drive replacements, whether SD card or Compact flash.

Many of the comments here casting doubt on the complexity of IDE have me wondering how many of you mainly worked with drives you bought new. The author touches on it briefly, but jumper settings were not obviously nor consistently labelled. (Same goes for geometry which was also needed.) If you didn't have the documentation that came with the drive then you had to go hunting on BBSes or FTP sites (if you had access to the Internet) for the information. Resorting to trial-and-error when you couldn't find that particular drive model in any of the text files. The author also doesn't go into how some jumper blocks were not merely bridging pairs of pins, but had configurations with jumpers at right angles to each other. Which I assume may have been a resister array, but it could also have been dummy positions used to prevent you from losing the jumper you might need later.

ATAPI solved a lot of problems of course, but came with some of its own. Notably that if the drive, BIOS, and OS couldn't all agree the wouldn't do DMA. I recall Windows 95 bug that erroneously dropped the hard drive into PIO mode and made everything slow to a crawl.

I'm not sure I agree with this.

I worked for a large west-coast computer store, in the early 2000s, selling and then later building custom PCs for walk-in customers, as well as doing walk-in computer repairs on all manner of different systems of various vintage, and I don't recall having too many problems getting IDE master/slave configuration to work... as long as we had some spare jumpers lying around.

> Eventually, by the late 1990s, things did work reasonably well.

Used to work in a computer store in the early 2000's and 99% of the time when someone came in to return an IDE drive it was due to this problem.

Ah yes the Customer Select jumper, fun times.

Anybody remember ESDI drives ???. Glad they only lasted a few years before the industry decided on IDE.

I remember it being about as messy as ST506. I think IBM used them in some of their PS/2's. I had a Model 80 that I rescued from the dumpster. It had cost about $7000 new.

In 1990, the company where I worked bought some IBM PC/AT compatibles with 80386DX @ 25 MHz and a pair of 150 Mbyte Maxtor ESDI HDDs.

At that time, having 300 Mbyte of HDDs in your PC was HUGE compared to more typical PCs, which had just a 30 Mbyte MFM HDD, usually from Seagate.

That was the only time when I have used ESDI drives, but those were great for that time.

When I left that company, six years later, all the ESDI drives from the 20 or so PCs bought in 1990 were still in use, without any failure.

We used them with read caching controllers - some fancy board with lots of RAM with a Motorola 68020 onboard but they caused more issues than the performance advantage.

Also remember ESDI was a pain to setup - it was a hack on ST506 so there was startup software to patch the BIOS you had to run to do the proper translation from head/cylinders to blocks done.

You are right, but for a few years, until IDE/ATA became available, the ESDI drives offered faster and bigger HDDs than anything available on the MFM Seagate interface, at still decent prices.

Sure, I didn't find Cable Select reliable, but I never ran into any issue just manually setting the Master / Slave jumpers.

Also didn't mind planning and configuring IRQ jumpers.

I actually found the old convention more deterministically reliable than the Plug-N-Play one which came afterward. To this day I still have random PCIe devices that, once in a blue moon, fail to aquire the resources they need in the OS, requiring a reboot (usually on machines with a lot of expansion cards and integrated devices).

I also liked that floppy drives were just media, and didn't require vendor-specific drivers or have capabilities of emulating other hardware or installing rootkits, the way USB keys do.

I guess I'm just old.

IDE Drives seemed plenty simple to me, coming from mostly using SCSI drives during that time.

I was on a bit of a nostalgia trip during the pandemic and put together an old PC last year to play some games I remember from the 90's. (Ones that don't work on Windows 2000 or newer).

Took me forever to remember that putting a CD drive as the IDE master and a hard drive as the slave was a recipe for pain.

Side note - I was able to get MechWarrior 4 Mercs running on Win 10 64 bit through a ridiculous amount of hacking and dgvoodoo. Was pretty hilarious seeing a DX8 game working with shadow play.

Was fine if you didn't leave drive jumper at auto/default and changed it to master of slave.

Must say I find it odd that what was kinda common knowledge for those of that time in tech for today to be a hell, believe me there are far better hells of that era, IRQ sharing and what PCI/ISA slots also shared with onboard devices like various ports. Now that if you didn't have the documentation would be emense fun.

Yes, relying on auto/default or cable-select was basically rolling the dice, so far as whether it would work reliably. The only way to get things to not-fall-over was to manually set the master/slave jumpers. And as was mentioned, you had to track down the correct size jumper for the drive.

Things got really fun when you added an IDE tape drive like the Seagate Travan series to your system.

IDE Tape, oh I can imagine. Dealth with Zip drives and other chancer short lived formats - though some just attached via LPT port. I'd imagine you would want that on it's own controller as could easily see the slow transfer quirks causing issues with any other attached device - there again having an HD on same IDE run as your HD was never ideal - more so if a burner. So a slow writting device upon IDE - oh yeah, you was setup for sure with that combo for many head scratchings and quirks.

By the time I started building gaming PCs in the 90s, many drives had Cable Select settings on the jumper. It pretty much "just worked" for me.

Weird, I built hundreds of PCs all through that decades, from 286s upwards, and I do not remember this ever being a problem. I would just check the PCB on the bottom of the drive to see what the jumper configuration on any given unit was, and it was fine. Built/maintained a lot of Compaqs, IBM, some Dell, and a lot of builds from parts.

I've been building PCs since 1993 or so, never ran into this (though I remember losing a jumper a couple times!) Maybe the bugs were worked out by then. Generally I bought hard drives of the same brand, so that was probably it: no mixing and matching.

Getting flashbacks. I'm glad SATA is a single device per cable/socket.

> But there are also drives which cleverly combined the two standards for 2.54mm pitch and 0.5mm pins

These would be 2.5mm pitch which is close enough to be semi-interoperable with 0.1".

Conner - wow - a name that I have not heard in decades.

Bent and twisted IDE cables was just as big a problem.

Jumpers worked reasonably well. Unless you somehow dropped one and didn't have a spare.

As a side note, the ‘Master/Slave’ notion has been declared verboten by the woke movement.

The master/slave quip, in combination with the recent post about Google’s newspeak, made me see the controversy in new light.

What if we are being told to not use the word ”slave” in an effort to make us unable to think about slavery?

Does anyone chain SATAs in their home PC builds? Like IDEs it's been sold as a feature to SATA since the inception, but I always assumed that it was like IDE: a part of the protocol no SATA chip provider bothered to debug, or even implement.

I think there were raid/NAS boxes that actually did it...

This is all from memory, so treat with caution.

The S in SATA is serial. The older IDE jobs are PATA - parallel. Two PATA drive share the available bandwidth (for want of a better word) and multiple SATAs have all the bandwidth available simultaneously to each device.

Copying from one PATA disk to another on the same bus is horrible - in out, in out shake it all about at a few 10s of MBs-1. Copying from one SATA to another runs at full speed.

For real speed you need SSD and hardware RAID with whopping caches and a price to match. For huge storage, reliability and a fairly eye watering price, why not use SSD to cache fast SAS 2.5" spinning rust with a decent hardware RAID. Oh I've just started to specify a SAN/NAS!

SAS allows "breakouts" and I think they may have updated SATA to allow something similar.

SATA has always supported port multipliers. Backblaze used them in the early days for their storage pods.

OK, so you kids know what you are on about. Am I roughly correct in my summation?

I think you might be confusing chaining with some other feature. SATA has always been a point-to-point link, not a multidrop bus.

SATA port multipliers are a thing, but they're active devices with multiple SATA transceivers, not a passive splitter. Support for port multipliers exists in most operating systems.

Port multipliers exist, and do work, but they are incredibly finicky. The precise set of combinations of HBA, multiplier, kernel driver, and drives can be a nightmare - once USB 3 became available and reliable I largely lost interest.

Reply on news.ycombinator.com