Sunday, November 5, 2017

Odd disk drive failure, more debugging of ethernet tool and restoration of 7970B tape drive


We met today to continue repairing the Diablo disk drives in order to return to archiving PARC cartridges. After cleaning and replacing heads last week, we had spun up one of the CE cartridges loaned by Digibarn to begin the alignment process. As you have read, the drive screeched and wouldn't spin either that cartridge or another known good one.

The edge of the platter was rubbing on the upper edge of a teflon guide bracket when it should have set centered in between the two forks. It appeared the alignment of the mechanical guides or some other part of the head assembly was off.

It took a few hours of analysis, adjustments of everything exactly to the procedures in the maintenance manual and then quite a bit of disassembly before we found the problem. Once we had the receiver mechanism cover removed and had taken a platter from a bad cartridge out of its case, we still were puzzled.

The height of the motor spindle determines the height and plane of the platter. The head assembly is anchored to the same castings as the motor. We couldn't see any way they could have become out of alignment this much so suddenly.

Finally, when we were visually comparing this drive to another Diablo drive, patiently measuring heights, gaps and so forth, we noticed something different about our drive spindle. A part of the CE pack can sheared off due to internal rust on the bolts and was stuck to our spindle!

Disk cartridges have an aluminum hub onto which the disk platter is bolted. Then, a steel ring is bolted to the bottom of the aluminum, to be held by a powerful permanent magnet ring on the top of the spindle.

This ring has held so firmly in place on the spindle that it looked normal, didn't shift at all, and thus we never noticed it. However, it acted as a shim to lift other cartridges up too high. We yanked the heads and indeed there was some oxide abrasion on the upper head from the rubbing when the platters tried to spin. No markings at all on the disk platters, fortunately.

An hour in the ultrasonic bath and some work with isopropyl alcohol soaked wipes gave us back clean heads to reinstall. They were inserted in place but we couldn't use the failed CE cartridge. Fortunately, Digibarn loaned us two of them. A careful cleaning of the platter left us ready to begin spinning it up and aligning the heads.

It was the end of the day already, thus we postponed the remainder of the alignment until the next work session. It will involve moving the disk arm out to track 105, putting the read head signal on the oscilloscope and slowly moving the heads outward until we have a balanced waveform that indicates we are directly over the track.

The head is moved outward by screwing in a setscrew that pushes against a 45 degree slant in the head mount, forcing it to move outward towards the center of the platter. The initial position for heads is about 5 cylinders too far towards the rim of the platter, thus we have to move about 5 mils forward using the setscrew. It sounds easy to do, although next week will tell us whether there are hidden complications.


Ken worked with his ethernet tool, a device which plugs directly into the multipin connector on the back of the Alto that is cabled to the driver card. Normally that multipin connector is cabled to a small box, the Ethernet adapter, which converts the digital signals into the analog modulation on the coaxial ethernet cable.

This tool avoids the need for that, instead avoiding the need for any ethernet cable or adapter boxes. It also converts the signal to use normal 10/100 Mbit ethernet on twisted pair, for routing to other machines or through the internet. The ethernet tool runs a full file server onboard, thus it can be used to network boot, copy files or disk images and use FTP or Telnet from the Alto.

Ken did enough testing to uncover three failure cases that cause the relatively long running diskcopy sessions to hang up. These occur when there is an outbound packet from the Alto to the network tool in close proximity to an inbound packet from the tool, typically a 'breath of life' packet that is broadcast regularly to all connected Altos.

Whether the inbound occurs before, overlapping, or after the outbound packet, the problem occurs if the time separation between them is less than a critical value. Ken is pursuing a new design for the buffer handling in his tool that should be immune to the race hazard failures it is experiencing with the previous code. The tool is quite usable except for these intermittent hangs on diskcopy or other long transfers from Alto to the device.


I gave Marc a 7970B tape drive, previously given to me by Al Kossow. The drive had been working other than failing to detect the reflective marker indicating beginning or end of tape. Suddenly, it stopped loading or responding to the buttons. We knew the lamp needed replacement for the BOT/EOT detection but this was a new symptom.

I found some aerospace stock of the exact lamp to install, but Marc still had to diagnose the failure to load. After checking connectors, power supply levels and all the usual starting points, he moved on to look over the circuitry involved.

He identified a small transistor that drives the Load latch and must switch on to begin the tape movement. While it appeared to work, its turn on characteristics were abnormal compared to a new stock replacement. A semiconductor expert identified this trace as a 'saturated collector' phenomenon, well know as a failure mode caused by a few defects in transistors.
Curve tracer of bad transistor
Good curves of replacement part

With a replacement transistor installed, the drive proceeded to load tape properly and appears ready to use. A simple transistor tester would have missed this defect, reporting the transistor as working properly. However, at low levels of input current to the base it would fail to sink adequate current through the collector, thus never forcing the flipflop to switch on.

No comments:

Post a Comment