What's the pitfalls when trying to connect a new SCSI hard drive to an old SCSI bus?

Question

Problem

I have a couple of old computers. Some of them have a SCSI-1 or SCSI-2 interface over an internal 50 pin IDC connector or an external DB25. Now I want to connect something to them, in particular a hard drive. One option is to get a "modern" hard drive.

I have heard that parallel SCSI is backwards compatible, so given the right adapter it should be possible to connect an Ultra-320 SCSI hard drive to a SCSI-1 platform. A comment to this question mentions that it is almost correct that SCSI is completely backwards compatible and that a new question should be asked concerning the details. This is that question.

(Please note that this question is not about SCSI-to-CF/SD adapters, it's not a generic "how to add storage to an old platform".)

Questions

• Bus width
  The old systems use an 8 bit bus, but a lot of the newer SCSI devices are listed as having a 16 bit bus. What happens there? Will the device figure out that my host can only use 8 bits?
• Electrical
  There seems to be a few different electrical standards: Single-ended, LVD, HVD. Presumably an adapter needs to convert between them. Is it possible? How do I know which is being used by the device? Wikipedia has this to say:

  Different SCSI standards use the same SCSI connectors as in HVD and LVD SCSI (High Voltage Differential and Low Voltage Differential) . HVD uses 15V while LVD uses 3.3V, so connecting an HVD device to an LVD host bus adaptor can blow the line drivers on the HBA, likewise an HVD HBA connected to an LVD device.

• Device size
  I know that a lot of file system drivers on the host side will have issues with larger disk sizes, and this question is not about that, but will the SCSI-1 command set at least be able to figure out that the device on the other end is indeed a super-mega-unheard-of large block device of several hundred gigabytes? That is - would a correctly implemented SCSI-1 hard drive driver with no arbitrary limits be able to interrogate and access the full size of a modern SCSI device?

In summary

What do I have to do, know, or find out if I want to connect an Ultra-320 SCSI hard drive to a SCSI-1 computer from the late eighties or early nineties?

Answer 1

Bus: yes should negotiate.

Device size: possibly okay. If your OS can issue (and your scsi controller supports) a read(16), then you'll get the full capacity. If it only supports read(10), then it'll look like a 2TB volume (assuming the drive is bigger!). I've heard of some random old controllers that didn't support read(16) and caused problems for big volumes, but don't know how prevalent that was. If it's 2TB or less, then no problem.

Electrical: unlikely to be a problem. So most of the other things should work (and will at least be safe to try) if you have the right electrical connections. HVD is incompatible with the other two (SE and LVD will be okay together). But, HVD was uncommon outside of special environments. (It was more expensive and the only purpose was to allow longer cable lengths). I wouldn't expect to see it on a random desktop. (And it's no longer made, so a new drive will not be HVD).

Most systems were supposed to use a different icon for HVD, the double diamond instead of the single diamond. It wasn't always true, but if you see that, you should expect HVD. See http://www.pcguide.com/ref/hdd/if/scsi/prot_Diff.htm for the images. I would look for that symbol stamped on the rear of the controller card near the port (or be embossed on the case if it's on the motherboard).

My main worry would be that that an older OS can't handle a large volume, not the SCSI config.

---

Oh, I didn't realize those links were to documentation of specific products. If the products don't mention "HVD" or "differential" anywhere (and I didn't see any), then there's no way they're HVD. They should be fine electrically.

Answer 2

Almost all SCSI devices are either 8-bit wide and have a 50-pin connector with single-ended (open collector, no differential drive) electrical signals, or are 16-bit wide, using LVD differential signalling AND having a fallback capability to work with single-ended drives if the wiring harness connects the right sense pins.

There are 68-pin data connectors (two types) standardized for the wide SCSI, as well as some 80-pin types that have data and power and configuration functions. The 80-pin SCSI SCA connector disk drives have no jumpers to be set, all that configuration being on the other side of the connection (usually an interposer circuit board adapter).

Will the device figure out that my host can only use 8 bits?

Yes, any transfer has a parity bit for each eight data bits; if parity is wrong for the 'extra' wires, the transfer is eight-bit, and if it is right, there are 16-bits (and in theory, 24-bit and 32-bit work similarly, but such hardware is uncommon). The narrowest device on the bus determines how many bits are transferred.

Now, it gets complicated.
The electrical interface can be open-collector TTL (single-ended = SE), or high-voltage differential (HVD), or low-voltage differential (LVD). This is indicated by a SCSI symbol with slightly different character SCSI symbols depending on the device, and finding the symbol will greatly facilitate understanding the result of connection. For SCSI-2 and later, pin #16 = "DIFFSENS" of the 68-pin connector will read 0.0V (or ground) on SE drives, +5V on powered HVD systems, and +2.5V on powered LVD systems; if a HVD or LVD drive senses incompatibility on that pin, it either turns output drive OFF, or converts its interface to SE if it has that capability (and it will work, but at a slower rate).

Well-designed devices do NOT blow anything out, if connected to incompatible hardwares, unless they pre-date the adoption of this convention.

What do I have to do, know, or find out if I want to connect an Ultra-320 SCSI hard drive...

You need power connection, data cable to a controller, and you need to set switches/jumpers to give each SCSI device a unique address. The addresses are 0 to 7 for the 8-bit units, 0-15 for the 16-bit ones, and usually #7 is the host controller (the computer). For a Macintosh, #0 was the factory boot disk, and #3 was the CD or DVD drive. Look for jumpers ("A0", "A1"...), or sometimes a case will have provision to plug a thumbwheel switch with numeral-in-window readout. All three jumpers OUT on a SE 8-bit bus signifies ID = 0; A0 plus A1 jumper sets ID =3.

Lastly, you must have terminators on the bus; one terminator will be OK on a bus four inches long or less (that's how a toaster-Mac is wired), but a second is required on the distal end of longer bus cables; often these are resistor packs to be drive-mounted in sockets, on modern disks they are logically switchable (so that the SE/LVD dual devices can operate) and require a jumper setting. There are feedthrough external terminators, that take power from the bus wiring, but one can overuse a bus-power source.

Fuses may blow if three or more terminators get connected. Don't do that.

The terminators may require terminator power jumpers, which do NOT disconnect terminator resistors, only select a suitable power source. If there are resistor packs on a non-termination disk, they ought to be removed. If an external disk has two connectors, you can apply an external terminator on the second connector, because terminator power can be connected through the data cable (but only in ONE PLACE, you want to know where the jumpers all are). It is common for internal cables to have one terminator ON THE CABLE to connect a number of terminator-less drives to a terminator-equipped host interface. important- terminators are different for LVD, SE, etc., and you need compatibility of the terminator as well as of the devices.

On elderly (1990s) Macintosh systems, gigabyte RAID was just fine on SCSI hardware. Other OS'es may have other features, and disk format limitations are usually present in software.

Answer 3

Personal experience: your mileage may vary

I had an 15K RPM U320 disk drive salvaged from a year 2004 Dell Precision workstation and thought I could replace the one in a year 2000 IBM Intellistation M PRO that has an integrated Adaptec 2940 with three ports.

Even though the drive had jumpers to downgrade the size/protocol and effectively showed up at boot, it actually never booted. I believe @Raffzahn is right, if not explicitly said in the docs it's already a good indicator about the compatibility.

We tend to over-estimate the 'loose' compatibility of older devices in comparison to more recent ones, which are more tolerant/capable of handling 'mostly following the standard' devices.

Here's another good example about this workstation, replacing the outdated ATI by a better Matrox G400: absolutely nowhere in the docs I could find that PCI parity must be enabled for the G400 to work correctly. (it took me a while to figure that one out)

From a practical standpoint:

• it will depend of your HBA
• it will depend of your device
• their firmware versions, there can be more recent versions bug-free, but will you be able to flash them is another story (not always possible in old hardware)

Answer 4

Very sinple answer: Check your drives documentaion and compare the facts with whatever your host interface provides

Otherwise this question is way to broad to generate any useful answer.