Written by Salvador Garcia
In 1987 IBM introduced a new standard for personal computers. While the ISA standard remained standard on most of the PC sold at that time, IBM introduced its line IBM PS/2 computer, boasting a new bus architecture known as Micro Channel Architecture, or MCA. MCA improved several aspects of configuration that was somewhat tedious before MCA came along. These included sharing interrupt lines and easier configuration of the adapters by using configuration files.
As this new bus architecture was introduced, the Local Area Network had taken off. At that time the predominant topology was the ARCNet standard, followed closely by Ethernet and Token Ring. Every company that offered LAN solutions had an ARCNet controller and most also an Ethernet controller.
It was in May of 1987 when Micron Computers (Not related to Micron Technology), or simply “Micron” decided to develop an ARCNet controller for the MCA. This new architecture posed some controversy. Some said that IBM was not happy with the ISA standard because many vendors had compatible adapters for it without IBM benefiting in a monetary way. So now IBM had come up with a proprietary architecture that they would license to all card developers. This policy was not set in stone, so Micron decided to go ahead with the development of the ARCNet board.
Micron had already signed up to NetWorld 87, a trade show dedicated to connectivity. It was going to take place in Dallas in October, so a race to have a functional prototype started in May. Micron already had previous experience developing LAN boards, having ISA products for both Ethernet and ARCNet. The development team consisted of four people:
1. The electronics engineer responsible for developing the hardware interface.
2. The software architect responsible for the software
3. The PCB engineer responsible for developing the PCB
4. The technical writer responsible for user documentation and component logistics
The software architect was also the Product Manager for this project.
The Project Description
The first step was defining the goals of the ARCNet MCA board:
a. Use the ARCNet Standard Microsystems Corp (SMC) chip
b. Take advantage of MCA’s new features
c. Compatibility with Netware 2.x and MS Networks.
While the company began the process to procure a few samples of IBM’s computers, specifically models 50, 60 and 80, the development team started on the design of the board.
As the first prototype of the board began to take shape, component documentation became a priority. The board used approximately 20 integrated circuit components plus a variety of passive components such as capacitors and resistors.
The task at hand was getting part numbers of each of the components from various vendors so that the purchasing department could have a choice and not be bound to just one component vendor. Datasheets littered the R&D department. These datasheets were critical in that in some cases, the equivalent component from one vendor varied enough from another vendor that the network adapter would not work.
The design engineer took every precaution in the design, trying to be as accommodating as possible to the subtle differences between like components. The layout engineer used the datasheets to define components in the CAD software used to design the adapter’s PCB and to start designing the layout of the board.
Glitches along the way
The first glitch was the design of the PCB itself. Previous designs had used fewer components, on average 10 integrated circuits plus a few discrete components; however, the ARCNet MCA adapter posed new challenges. The engineer knew that he needed to design the board so that radio interference would not be a problem. The components were much closer together and he had 4 layers to work with. Two of these layers were used for power and ground while the remaining two were used for control signals and data.
The ARCNet HIT (High Impedance Transceiver) had to be placed as close to the media as possible to minimize crosstalk from the other control and data signals. The PCB layout engineer found that component placing was also critical. The HIT had to be as close as possible to the rear bracket while components that were connected directly to the MCA slot had to be as close as possible to the slot.
The HIT was a modular component responsible for placing the data signals directly on the cable, or in other words, the implementation of the Physical layer of the OSI Model. The theoretical concept of the HIT was its support for different media types, depending on the vendor and model of this component. The pin out was standardized so that the designer could use any of the available HITs available using the same PCB layout.
One critical aspect of the design was the timing on different I/O lines that came directly from the MCA slot. The electronics engineer discovered that the timings were slightly different on the varying IBM PS/2 models. He had to have just the right timing that would fall within an acceptable margin for the three models.
The first prototypes: A Nail in the Coffin
Once the PCB layout engineer finished his work he generated the Gerber files and other support files and sent them off to the facility that was going to produce the prototype PCBs. While this was happening, the development team started to talk to purchasing so that it could procure the necessary parts to assemble the prototypes. By convention, 10 prototypes were always ordered. One sample was destined to be the “phototype”, a name that the development team gave to that one sample that should be assembled as carefully as possible and then passed on to marketing so that they could begin elaboration of the promotional and packaging materials. The technical writer was busy elaborating first drafts of the User’s Guide, any technical notes needed by the assembly plant and a prototype of the quick configuration guide.
Once the boards arrived they were checked to confirm that they did not have any obvious errors, such as having a short between power and ground or not having the MCA slot contact pads misaligned. Once they passed inspection the prototypes were assembled.
This is where the greatest flaw was discovered. For some reason the pin out of the HIT was reversed! The executives were not happy. We had limited time to have the prototypes corrected and the PCBs redone. We started an emergency investigation to determine what went wrong. This did not take long. The pin out information on the datasheet was wrong. Although the development team explained the findings to the executives (CEO and CTO), they did not find this flaw excusable. They mentioned that it was not possible for such a basic error to have happened when the company was already producing a fully functional ISA ARCNet adapter; however, there was one aspect that had not been taken into consideration.
While the company had been producing an ARCNet adapter, the original design and PCB trace routing had been done by a third party. When the company decided to do it in-house, the layout engineer received the necessary file to import into the CAD software. He had basically redesigned the PCB, but using previously existing “net” information that contained all the connections for the adapter. On the other hand, the ARCNet adapter for MCA had been designed from scratch. The most basic information that the development team had for the design was the datasheets.
While the executives were not happy, there was a simple workaround to the problem. The HIT could simply be installed on the solder side of the adapter. While this was an ugly solution, it allowed the project to continue and not be bogged down another 2 to 4 weeks waiting for another batch of PCBs to be made. No one was happy with this solution because the company wanted the prototypes to be as presentable as possible for the NetWorld ’87 event that was coming up in a few weeks.
Unfortunately, due to lack of time and money the development team had no choice but to use the workaround.
The Other Glitch
Once the boards were assembled testing began. The design engineer found the usual set of culprits and gremlins that prevented the board from working correctly. The software architect was basically done with the drivers needed and was just about ready to test. The final glitches were ironed out and finally the time had come to test the ARCNet board with its own software. When it was installed in the PS/2 computer, along with the software, the PC was turned ON. The good news was that neither the PC nor the adapter exploded (this had happened in the past with one of the adapter’s chips). The bad news was that once the network software was started the adapter did not connect.
At this time it was less than a month away from the Networld ’87 event. The executives had been busy planning the logistics of the trip and were just about ready. It had been designated that the 2 executives, the electronics engineer and the technical writer would represent Micron at the event.
After some investigation with a digital analyzer, the engineer found that for some reason the interrupt was not being reset after it had been triggered. This was very important. The board was interrupt driven. Every time that any data was received by the board the SMC chipset would generate an IRQ, that is, a hardware interrupt. One important component of the driver was the interrupt handler, the piece of code that got executed whenever the IRQ was asserted. Both the hardware and software team members worked together to resolve the issue.
Finally, just about a week before the event, the MCA ARCNet board worked! There was jubilee. The executives we ecstatic, since Micron was but one of a handful of companies that had announced an MCA ARCNet adapter and even fewer of the ones that announced that it would be ready for the NetWorld ’87 event. The only downside was that ugly HIT that had to be attached to the solder side of the board.
Many more tests were made to ensure the functionality of the board. One stress test that Novell recommended was playing Snipes, a game that was included with Netware. According to Novell, this game taxied the driver to its limits, as there was a huge amount (for 1987) of data being received and sent. It was especially interesting and rewarding seeing the co-workers somewhat distraught when they saw the entire development team playing Snipes for extended periods of time.
NetWorld ’87: Everything happens for a reason
The four Micron representatives arrived at the Loews Anatole hotel in Dallas. They were ready to demonstrate the MCA adapter. When they arrived at the show and started to setup the computer with the Micron ARCNet MCA adapter they received some unexpected news: The only hookup available at the show was through fiber optic. The HIT that the board used was compatible with RG-62 coax cable. After so much work Micron would be unable to demonstrate a fully functional MCA ARCNet adapter. This also came as a surprise to the other 2 or 3 venders that also had an MCA ARCNet adapter. The media types were simply incompatible.
At that time another company called CODENOLL was heavily marketing an ARCNet fiber optic solution. I believe that at that time we all saw that Ethernet was gaining ground and with a speed of a whooping 10 Mbps it was definitely faster than ARCNet’s 2.5 Mbps. Although the token passing protocol was better optimized for heavy traffic loads, Ethernet’s sheer speed made it the most viable choice. Another advantage that Ethernet had over ARCNet was the addressing scheme. While Ethernet used a MAC address, basically allowing unlimited stations to be part of a LAN, ARCNet used an 8-bit addressing scheme, meaning that for any ARCNet network, there could only be 256 stations. Since the zero node address could not be used, in effect it was 255 unique addresses. Broadcasts were sent to the zero address, so it was imperative that no node have this address.
CODENOLL’s product at least addressed the speed limitation. They were at the show promoting and lending out samples of their fiber optic HIT module; however, it was not possible for the other vendors to swap out the HIT already in place. But, with the presence of the electronics design engineer at the show, he said that he could indeed swap out the coax HIT and replace it with CODENOLL’s product. As he installed the HIT he noticed that it was at least double the height of the coax HIT. As he popped the adapter back into the computer everyone noticed something incredible.
The additional height of the CODENOLL HIT did not allow the adapter to be installed back in the PC if the pin out had not been reversed. The only way that the adapter could be installed was if the CODENOLL product was on the solder side of the board, which it was!
Because of this imperfection in the PCB, Micron was the only company at the show that had a fully functional MCA ARCNet adapter actually connected to the show’s network. It goes without saying that the executives were happy.
After the show, Micron was ready to ship the final product within the next 2 months. Interestingly enough, Micron did not see sales for this model skyrocket. The PS/2 was relatively new and businesses were not flocking to it. Hardware vendors were not comfortable with the idea of possibly paying a licensing fee to IBM for just commercializing MCA compatible adapters. As time has told us, MCA never fully took off and was mostly forgotten.
One good technology that came from the PS/2 line of computers was the video standard, at that time simply known as VGA. This standard is still with us today, albeit much changed and improved.
What lessons did the development team learn?
- Everything has a solution
- Check and double check the information that is being used for a given design.
When putting together a complex project it is always a good idea to have different check points that can be quickly tested and analyzed. Modularity is a good thing and any oversight can not only be corrected, but in some cases, it might also turn out to be beneficial.
As we suspected back then, Ethernet overtook ARCNet. The CODENOLL product never fully rescued the ARCNet market as there was a prohibitively high cost to use fiber optic to “wire” up a LAN. Today ARCNet is limited to extremely niche markets and fiber optics has found a comfy home as backbones.
Want to know more about ARCNet? Here is a basic tutorial:
Update May 30, 2019: The above link is broken. Gone forever. Try this one instead: