You might look at the title and not realize what the reference is – but apparently, SCSI was supposed to be pronounced ‘sexy’ until somebody ruined it. I suppose that’s fitting when SCSI evolved from SASI, which one can only presume, was intended to be pronounced ‘sassy’.
Quick SCSI ‘101’
SCSI is another one of those old interfaces which has sort of died off. Popular in early Macintoshes and some other systems (and largely unpopular with most PCs), it was probably best remembered in the SCSI-1 incarnation of 50 pins, 8 devices. It would have been remembered as “superior” for its time in terms of speed and command set standardization. Its legacy, command-set wise, lives on in Serially-attached SCSI (SAS).
The death of SCSI is not at all surprising – increasingly, due to better signal processing and better ICs, we are managing faster than ever rates in serial communication which obviate the problems of data skew in parallel systems (which is one of the limiting factors to how fast (and long, distance wise) parallel buses can reliably run). It also solves a few issues with connectors having large numbers of pins which are vulnerable and easily damaged. Its parallel, multi-device daisy chain nonsense is gone for good. I guess that’s good news, as SCSI was (in reflection), an inconvenient bus.
Working with SCSI was like putting all your eggs in one basket. This basket includes internal and external drives – including the one you booted off. If anything “malfunctioned”, or a device got removed, the whole bus broke and data corruption was a possibility. Imagine “inadvertently” unplugging an external drive and trashing your boot disk – could happen, but more likely, you’d just have hung your system and needed a reboot, losing anything that was stuck in RAM.
There was also issues with the behaviour of the bus – certain devices didn’t get along, and while the cable length was quite generous, often if you got close to the limits, you would experience issues related to termination affecting signal quality. Definitely not a ‘plug and play’ solution on a bad combination of cables, connectors and terminators. You also needed to get the right connectors on your cables amongst a collection of commonly used connectors.
Need I remind you that you had to manually set your SCSI ID on each and every device? It’s a manual configuration chore which (luckily) we’ve gotten rid of in modern buses – either by having an auto-configuration strategy (like Cable Select in IDE, which may (depending on the cable) not work at all) or eliminating the multiple-drives-per-channel business altogether (like in SATA).
It’s not the first post of mine that’s involved SCSI – after all, my Syquest and ZIP drives do use it. But so far, all of my things have been hooked up to the Adaptec AVA-2906 50-pin PCI SCSI controller – which is a FAST SCSI-2 interface.
This was a 50-pin single-ended signalling format carrying 10MB/s. This would be familiar to many in the early days of the CD-Writer where such cards were bundled with the SCSI CD-RW drives of the day. But less popular were the wide incarnations of SCSI which took over from this point.
Wide SCSI is noted for having a 16-bit width rather than an 8-bit width, thus the pin count is increased to 68 pins. Thanks to a recent ‘windfall’ on OCAU’s forums – I managed to bag a load of SCSI gear at cost of postage.
The “wider” SCSI family
The serious SCSI drives pretty much involved the 68-pin wide format, in order to take advantage of a doubling of data transfer speed. In addition, this doubled the number of SCSI ID’s from 8 to 16, allowing for more devices on each bus. Narrow and wide formats enjoyed increases in data transfer rates through changes in the bus clock rate and signalling formats (moving from single ended to low-voltage-differential bus and double-clocking the data rate).
Thanks to Axe from OCAU forums, I ended up with a pile of controllers and drives – some working, some not (in rather surprising ways).
Wide SCSI Controllers
This is a Buslogic BT-956C Wide SCSI controller. The interface chipset is nicknamed the “multimaster”. The manual would date this card at about 1994, and it involves also a narrow-SCSI connector to allow you to connect both types of devices, noting that all three connectors can not be used simultaneously. This card supports single-ended SCSI with software controlled active termination, but not differential signalling (as the BT-956CD does). Luckily most SCSI drives which are LVD can operate in SE mode as well.
A close inspection of the card sees an AMD N80C186-25 80186 25Mhz CPU. It’s a clear sign of the x86’s presence in embedded environments as well early on, which has now moved to ARM/MIPS instead. Maybe that’s why these cards were so expensive – that’s pretty much a computer right there!
Buslogic was a big name in controller cards, so it was rather strange for me not to be able to find much about them anymore. As it turns out, they were acquired by Mylex Corporation which, itself, was acquired by IBM.
The next controller I managed to grab a hold of is the Adaptec AHA-2940UW. Adaptec was one of the biggest names in SCSI controllers and interface chips – you could almost never go wrong with Adaptec as many external devices (such as scanners) were often only qualified for use with Adaptec controllers. They definitely weren’t the first, however.
This controller claims a speed up to 40MB/s, which would have (informally) have been known as SCSI-3. This is why it’s known as an Ultra Wide SCSI (as opposed to the Wide SCSI above). It is a single ended signalling controller.
Adaptec is now owned by PMC-Sierra, but they still maintain their support pages. One of the great things is that old hardware like this tends to be supported by Linux ‘out of the box’, making for a truly plug-and-play experience. Hunting for drivers under Windows is also relatively limited as support for some popular hardware was baked into the OS.
The last controller is a LSI/Symbios Logic SYM8952 dated 1999. This controller supports both Low-Voltage-Differential (LVD) and Single Ended (SE) signalling, but eliminates the narrow SCSI connectors altogether. With a “high-9” terminator, you could still attach narrow-SCSI devices to the bus, although such specialty adapters are getting difficult to acquire.
This is an Ultra2 SCSI adapter, which has a transfer rate of up to 80MB/s. This adapter seems to have been made during the acquisition of Symbios Logic by LSI Logic (which is still present and thriving today – recently acquring Sandforce and entering the SSD market). An interesting piece of trivia is that Symbios Logic itself was a spin-off from NCR (yes, the guys that make point-of-sale cash-register equipment), who were the first to make a SCSI bus controller IC.
Wide SCSI Connectors and Terminators
As you could see from the photos of the card above, the connectors used for Wide SCSI were almost universally HD68 type.
This connector was very annoying as the pins were thin and fragile and easy to bend. A careless user, unplugging the connector by rocking it from side to side, can easily permanently damage the connector by flaying the side pins outward and possibly cracking the shroud. The pins are also vulnerable to pressure – compression of the connector from top to bottom can result in all the pins bending inward towards the center. It wasn’t sturdy compared to the 50-pin connectors before.
At the end of each bus, a terminator must be fitted to attenuate signal reflections from the ends of the bus. There were two types of terminators in SCSI1, although most SCSI2 systems had settled on “active” termination (utilizing ICs, as opposed to passive termination involving resistor packs only) for reliability reasons.
It is marked MLVD Terminator – I suppose this might mean “Multimode LVD” meaning it’s probably SE compatible as well.
Removing the cap reveals three Linfinity (now Microsemi) LX5241CPW Multi-mode SCSI terminator ICs.
Additionally, I also managed to have a pair of these shipped to me –
These are 80-pin Single Connector Attachment (SCA) to 68-pin adapters. SCA was used in high end drives intended for direct mounting into backplanes – allowing host swap and configuration to be done by the backplane. This adapter breaks out the power to a molex connector, the data to the 68-pin connector on the rear, and some of the configuration to DIL header pins. The pins include activity LED, Spindle Synchronization Signal, Delayed-Spin, Commanded-Spin, and four-bits of SCSI device ID.
SCSI Drives – Seagate Barracuda ST39173W
I received a pair of these Seagate ST39173W’s 9.1Gb 7200rpm Ultra Wide SCSI drives. From the StorageReview article linked above, it would have been $850 each in March 1998. Funnily enough, both drives work flawlessly which is a great sign.
The drive has a very interesting design – the PCB is very busy and the spindle motor is connected in a triangular fashion. We see a mix of ICs from Mitsubishi, Lucent, TI, Siemens, MX, UMC and VTC. There are also connector fingers on the side for manufacturing use. There is a set of jumpers at the front, as well as one just under the cache RAM for configuration.
SCSI Drives – IBM DFHS S4x
I purchased this drive from eBay just for fun – it’s an IBM Starfire 4.51Gb 7200rpm full height 3.5″ drive. It’s amazing how detailed the specification is for these drives, which are normally used in OEM applications. The labels on it imply that it was installed for a customer of EDGE in 1997/1998. It’s got an interesting copper-label band wrap around the drive over the top of the cover, rather than around the perimeter.
The PCB is also interesting – who knew IBM SCSI drives have a Western Digital controller on it, and another AMD 80186 CPU? It looks like each of the motor phases is handled by a discrete pair of MOSFETs as well. The drive is dated underneath as manufactured Week 30 of 1996.
SCSI Drives – Hewlett Packard DGHS
I got shipped a pair of these as well, but this one wasn’t functioning. It was completely dead, no spindle action – so it’s a good donor for a teardown. Upon looking at the top cover, it’s already obvious, despite the HP D6108A branding, it’s an IBM DGHS series drive. A quick look for a datasheet turns up this. It’s an Ultrastar 18XP series 7200RPM drive of 18.22Gb capacity. It also supports Ultra2 LVD for 80MB/s maximum interface throughput.
The drive was manufactured 20th August 1998 in Singapore and is a full height drive. Just so you know what that means …
Yep. That’s twice as tall as your ordinary 3.5″ hard drive you use today. They sneakily put a cut-out for an electrolytic cap there too. I like it. The drive has an amount of heft that isn’t in most drives today.
The serial number is on the front along with some front-access configuration jumpers.
This being a serve drive, it has the 80-pin SCA, so we need to use the adapter if you want to use it outside of a backplane.
The PCB reveals an IBM SOC and another AMD 80186 CPU, amongst the other chips (a Linear Technologies motor driver, an Intel flash chip, a TI chip, etc).
Now time to crack it open and see what’s inside:
Looks like a regular hard disk right? Not quite. First, notice the spindle clamp – there’s not a single screw on this one. I couldn’t manage to undo it with the tools I had. The next thing is …
… there are ten platters and twenty heads. Wow. This thing is a monster!
What a shame this one of the pair wasn’t working – but I managed to grab the magnets which were even stronger than the normal hard disk magnets. These are almost lethally strong.
There’s a picture of the heads, the gimbal assembly and the voice coil motor.
To test my macro skills – I decided to get a snap of the head bonding wires – where the signal read/write pairs are bonded to the little gap that forms the heads. It’s about 2mm across.
SCSI Drive – Western Digital Enterprise 4360
This is a 4.36Gb single-ended 7200rpm ultra fast wide SCSI-3 drive. What a tongue twister. But I saved the best for last – so you’d probably want to keep reading. It was Made in Singapore on 4th February 1997, and I received a pair, but both weren’t working.
The spin-up on both drives were slow, but not horribly loud. The motor spindle chip got really hot – so I tried a heatsink, then a block of ice. I figured the bearings on such an old drive might have been seizing, so I even froze one of the drives. If you see condensation on the following images – that’s because the photographs were taken of the drive that got frozen.
The underside of the PCB seems quite simple – especially for a drive of this age. There’s a piece of flash, a WDC controller, a spindle motor controller and probably a voice coil motor controller.
The side of the drive looks like the consumer Caviar series of the time. The rim is sealed with tape – of which this one seems to have shrunk. The side of the drive has four holes instead of the three which is more usual of contemporary drives.
The seam of the tape ends at the front – and there’s a jumper configuration block there too, as expected.
So why didn’t this drive work? Just take a look …
That is not normal. The platter surfaces have literally grinded away. They are no longer smooth in bands, which vary in size. Running a fingernail over the surface shows distinct changes from smooth to near-abrasive. One has to wonder how this extent of platter damage could ever happen.
I was so shocked, I took a whole stack of photos of it.
What’s clear is that every platter in the stack has suffered.
The cover, air filter and internals confirm that the cleanliness of the internal chamber has been violated. This dark dust is almost like carbon – and I suspect it’s a sign of how the drive may have failed. The platters are often coated with a hard carbon diamond-like coating to reduce the changes of damage from when there is a head crash.
Maybe the heads were idling or being read/written to on very specific tracks for long periods of time, and vibration/humidity/dirty air may have caused head crashes which caused debris to be generated, making it more likely to have more head crashes until it entirely snowballed.
Then, in an effort to recover the data, someone ran a utility that kept retrying (e.g. ddrescue) and that eventually did everything in (maybe). Despite all of this, the drive didn’t make any awful noises, just a slightly harsher bearing sound …
Some platters were more affected than others – some of the dark matter may have been the plastic former that the head sliders are made of being “abraded” off, literally sanded away by the rough surfaces and never really “flying” in the first place. There are sections where the media colour is concentrically lighter – indicating possible wear and developing roughness.
I am astounded. I never thought I’d ever meet this severe failure.
The rear of the PCB shows several components – mainly an Intel MG80C196 Microcontroller, Adaptec SCSI interface controller and some cache RAMs.
Thanks to Axe, I’ve managed to grow a collection of valuable SCSI gear. While most people may not see the value of such things, for a “hobbyist” which might see some gear arrive one day which needs the controllers/cables/terminators – this could be the difference between recovery and none at all.
It was interesting to test all the SCSI disks I’ve managed to come into possession of – and the WD certainly showed a very interesting failure mode that must have taken a long time to develop. But that wasn’t all – we managed to see that the use of x86 processors in embedded situations isn’t so strange after all – drives and controllers have made use of them.
No matter what, it’s definitely nostalgic playing with old equipment. Would have cost a fortune back when they were new …