Ardent Tool of Capitalism

IBM Designer Sheds Light on Micro Channel Architecture

Source: InfoWorld, Volume 9, Issue 38, 21 Sep 1987
Author: Michael J. Miller and Alice LaPlante

In an exclusive interview, Chet Heath, the chief designer of IBM's Micro Channel Architecture gave InfoWorld a close look at the new bus design, detailing what it means for PC users. Heath was interviewed by InfoWorld following a technical seminar he gave on the MCA at PC Expo.

IBM's Micro Channel Architecture is a versatile and complex design whose capabilities will become clear only through a long-term education process, according to Heath.

"The complexity of the Micro Channel is such that it makes educating the industry a difficult job," said Heath, who has been traveling around the world over the past months, giving lectures and seminars on the capabilities of the new architecture. "It's like future shock - it's going to be a gradual education process."

Some benefits of the Micro Channel - such as requiring smaller add-in boards as well as allowing switchless installation of those boards - are obvious, said Heath. Other features are more subtle, such as the lower electromagnetic interference generated by the bus and the fact that boards designed for the MCA will be more reliable and problems easier to diagnose and repair.

Some capabilities of the new design have yet to be exploited, said Heath. For example, the MCA can support up to eight "master" microprocessor chips - chips that directly control access to the machine's memory. This could ultimately lead to fault-tolerant systems or computers that run multiple processors.

Finally, the bus was intentionally designed with plenty of room for growth: a significant number of features have been "reserved" for this purpose, and Heath says that even IBM cannot predict how those resources will be used. "The MCA architecture merely shapes the boundaries of what is possible," said Heath. "It's the PC industry that is going to determine how it is used."

Why Chance The Design?

Contrary to industry speculation. IBM did not deliberately change the bus structure to ensure a proprietary PC design, Heath said. Instead, work on the MCA began in 1983, prompted by problems IBM engineers were having with electromagnetic compatibility (EMC) on its original PC line. IBM also wanted to respond to customer requests for a "switchless" setup of add-in boards and the capability to use more advanced processors.

"We could have shoe-horned a 386 chip into a 286 box without any trouble," said Heath. "This would have provided users with increased performance and would have been a generally OK solution."

At first, the main impetus to change the bus design was an FTC requirement that manufacturers meet certain EMC standards to prevent computers from distorting radio and television reception. IBM's original PC XT had difficulty meeting those standards.

"With the PC XT, we had to copperplate the chassis to make sure that electrical currents returned to the power supply," said Heath. "Then, it still wasn't good enough, so we had to nickel plate it - a very expensive process. We finally got it working, but we spent a lot of money on it and didn't get any increased functionality for our customers."

So meeting FCC requirements was a top priority, said Heath.

IBM also began working on a new bus design because users complained about the difficulties of installing add-in boards due to the plethora of DIP switches on those boards. "No one liked the switches - it took close to an hour just to set a system up," said Heath, 'in addition, switches are the No. 1 cause of 'no problem found' diagnoses on service calls - and someone had to cut those costs, either the user or the dealer. So we decided that the switches had to go."

In addition. IBM needed changes in the bus to accommodate advanced processors - such as the 32-bit 80386 CPU - and also wanted to be able to perform true multitasking.

According to Heath, one of the more difficult decisions was to change the size of the bus - the decision that would render existing add-in boards obsolete.

"What may seem obvious now is the result of a lot of soul searching," said Heath. "But once we made the decision to change the physical format, the door was wide open to improve everything."

Priorities

The next stage for IBM was to compile a list of priorities that would determine what features actually got implemented in the new design:

1. Safety
2. Data Integrity
3. System Integrity
4. Reliability
5. Compatibility
6. Functionality
7. Performance
8. Cost

Heath acknowledged that to the PC industry, the surprising aspect of this list is the relatively low priority assigned to functionality and performance when compared to such things as data and system integrity and compatibility.

"Performance is important, but not as important as other things," said Heath, who said that IBM was searching for balanced performance that would best suit its customers. "For example, we put a very high priority on data integrity - even listing it higher than system integrity," he said.

"The reason? For most people, the value of the data stored in a computer far exceeds the value of the hardware itself."

Heath said that the decision to switch to 3 1/2-inch floppy disk drive media from 5 1/4 -inch formal aptly illustrates the use of this prioritized list.

"First, the 3 1/2-inch media is much more reliable, which fits item No. 2 on the list," said Heath. "Since data integrity was more important than compatibility, which was more important than cost, you can see how we made that decision."

Compatibility also came before performance and functionality, said Heath, who said that if IBM hadn't had to worry about compatibility with the existing software base, it could have been much more innovative and thus increased performance in "creative" ways.

"The compatibility issue meant that we didn't have the option of being truly creative," said Heath. "Instead, we had to get away with being clever."

Meeting FCC Requirements a Top Priority in Micro Channel's Design

Of utmost priority when designing the Micro Channel Architecture (MCA) for IBM's new PS/2 line of microcomputers was meeting the Federal Communications Commission (FCC) regulations for electromagnetic compatibility (EMC) - the radioactive emissions generated by a computer that can distort radio and television reception.

"Meeting FCC requirements was a top priority," said Chet Heath. IBM's senior designer of the Micro Channel, who said that many of the most important design changes from the "classic" PC bus were a direct result of EMC issues.

For purposes of controlling EMC, the FCC has created two classifications of computer equipment: Class A for "commercial" computers and Class B for "home" computers.

Class B regulations are much more stringent than Class A regulations because there is a greater likelihood that emissions from home computers would interfere with domestic radio or television reception, according to the FCC.

Computers used commercially are less likely to do so, simply for the reason that there is less television and radio use in industrial parks, office complexes, and other business environments.

"Class B requirements have to be stricter because there are so many opportunities for creating interference in a residential environment," said Bob Cutts, chief of the authorization and evaluation division at the FCC's Office of Science and Technology Laboratory, in Columbia, Maryland.

However, although theoretically the stricter Class B designation is intended to apply only to computers used at home, that is not the way the guidelines are effectively enforced by the FCC.

Instead, all microcomputers - even the PCs bought for and used in a business environment - are automatically required to meet the tougher Class B standards. The reason for that, according to Cutts, is that PCs are small enough and light enough to be easily transferred between homes and offices.

"Unless there is a solid reason that a computer can only be used - and I mean only - in a manufacturing plant or other industrial setting, it is considered Class B," said Cutts.

For example, large mainframe and minicomputers kept in air-conditioned rooms obviously couldn't be moved from a commercial to a home setting, said Cutts. However, desktop PCs could - thus making it necessary to ensure that PC emissions don't exceed a certain limit.

At IBM, work on the MCA began in 1983 precisely because of problems IBM was having with EMC on its original PC line, according to Heath. IBM engineers often were required to perform a lot of last-minute work on PCs simply because they failed to meet FCC standards.

"We were becoming increasingly aware that EMC issues were not being effectively managed," said Heath. "Although we never shipped anything that didn't meet FCC regulations, a lot of dollars were invested in last-minute solutions to EMC problems."

But this last-minute approach to EMC emissions eventually hurt end-users of IBM products, said Heath, "EMC issues always seemed to come up at the end of a design cycle, and we would end up throwing a lot of money into solving something that didn't bring any enhanced functionality to end-users," said Heath. "It was sort of a sales tax from Mother Nature."

According to the FCC's Cutts, PC vendors are required to send their machines to the FCC to be tested before bringing any products to market. The FCC then determines whether each PC model passes the Class B requirements.

"If a machine meets our technical standards, and if the vendor's application is in good order, then they get an authorization to go ahead and manufacture and market that computer," said Cutts.

Yet at every Comdex show, the FCC makes a point of swooping onto the exhibit floor, seizing any PCs being exhibited without proper FCC authorization. Usually, a significant majority of the machines - estimated between 60 and 80 percent - have violated FCC regulations. (See "FCC Cracks Down on Vendors at Show." June 8.)

"We're aware of a large number of noncompliant manufacturers," said Cutts. "It's really not too surprising - this is a very competitive industry, and prospective vendors want to get their products out the door as soon as possible."

Trade shows are great places to exhibit new PC products, said Cutts - but if a vendor wants to show a machine that hasn't yet received FCC approval, they are required by law to give notice of that fact.

"And sometimes they don't do that," said Cutts.

Companies marketing PCs that haven't been approved by the FCC can be fined or prohibited from selling product, said Cutts.

— Alice LaPlante

Physical Layout

The physical layout of the MCA as released in the PS/2 Models 50, 60, and 80 includes three different types of physical connections to add-in cards. These include:

1. A 16-bit connector
2. A 16-bit connector with a special video connector
3. A 32-bit connector

All connectors are considerably more sophisticated than those on the "classic" PC bus, said Heath.

One example is that boards designed specifically for the MCA will haw faster input and output for tasks such as controlling ports or disk drives (see sidebar below).

The Physical Layout of the Micro Channel Architecture

The final Micro Channel Architecture as released in IBM's PS/2 Models 50, 60, and 80 includes three different physical connections to add-in boards: a 16-bit connector, a 16-bit connector with a special video connector, and a 32-bit connector.

All of these connectors are considerably more complex than the bus used in the original IBM PC XT and PC AT, according to IBM's Chet Heath. With the Micro Channel, there are now multiple signal lines (the means of exchanging information between the system board and an add-in board) as well as more power and ground lines.

The 16-bit channel was designed to accommodate 77 signal lines, 29 power and ground lines, a separate audio line, and five reserved lines in a 58-position, 50-mil pitch (pin-to-pin spacing) card. In fact, it's divided into a 45-position 8-bit section plus an 11-position 16-bit extension, separated by a space two positions wide to ensure that the board is correctly installed. You can get two signals for each position on either side of the plug-in board.

The 32-bit channel extends this design even further to allow 32-bit memory addressing and data transfer capabilities. It does this by adding 31 signal lines, 15 power and ground lines, and 16 reserved lines to the 16-bit extension.

It also adds a "matched memory section" used to accelerate memory transfers faster than 10 MHz with three signal lines, three power and ground lines, and three reserved lines to the original 8-bit section to accommodate the greater memory addressing, according to Heath.

All told, the 32-bit boards have 93 pins including the matched memory extension.

These boards reflect several major advances over those used in AT-type machines - including 386 machines based on the PC AT architecture. Boards designed specifically for the MCA will be able to use more signals for input and output, for such functions as controlling ports or disk drives. When building PC AT - and 386/AT - machines, board designers used the 16- and 32-bit capabilities only for addressing memory, Heath said. Such designers continued to build 8-bit boards, rather than 16-bit boards, for input and output, so that their products would work in both PC XT- and AT-class machines.

But in the PS/2 family, the 80286-based Models 50 and 60 have only 16-bit connectors, while the 386-based Model 80 has both 16- and 32-bit connectors, making more sophisticated input and output likely, said Heath.

The MCA also provides a broader distribution of power and ground lines. Every fourth pin is either a ground or power pin, offset on either side of the board by two pins, so that no signal is more than one pin, or .1 inch away from a ground. This reduces the size of the loop made by the electronic signal to one-twentieth the size used in a PC AT. This has several advantages, including the fact that it dramatically cuts down on the electromagnetic interference, and it creates a cleaner, more reliable electronic signal, said Heath. These advantages would be lost if the architecture accommodated even one old slot, he said.

Each machine also has one 16-bit slot with an extra 10-pin video connector, this lets users avoid paying for the circuitry when extending the function to maintain compatibility.

With the video connector on the Micro Channel bus, a new card with a higher graphics standard - such as IBM's 8514 card - no longer has to include the circuitry for backwards compatibility.

Instead, it can pass the signal back to the VGA chip on the main system board.

Through this connection, video signals can be merged and synchronized. In addition, images are refreshed at a slightly higher speed in most modes for reduced flicker.

In addition, the architecture was designed to accommodate surface-mount technology and very large scale integration (VLSI). Besides being cheaper, this means that add-in boards designed for the MCA are considerably smaller than PC AT boards.

The smaller size also helps IBM meet ergonomic requirements in European countries, which require the center of the monitor be less than approximately 10 1/2 inches above the desktop.

Easy Installation

One of the most obvious differences is a feature called "Programmable Option Select," which means that users won't have to worry about setting DIP switches when installing add-in hoards. Instead, the Micro Channel essentially replaces switches on add-in boards and on the main system board with a set of memory registers on each card that contains setup information.

The switchless installation has many advantages. It saves installation time and makes it easier to install add-in products.

In addition, boards can be easily and automatically reinitialized to prevent conflicts, making it easier to put in multiple identical boards. All told, with an optional extension protocol, the special registers allow users to get more than 128,000 switch possibilities.

A side benefit is that now the machine "knows" which cards are inside of it, making diagnosing problems simpler.

Each add-in board designed for the MCA card has its own unique identification number and comes with a disk containing an .Adapter Definition File (.ADF), a text file that describes which resources each plug-in board requires in order to function.

Whenever a user installs a new board, the information on the ADF is stored in nonvolatile RAM on the main system board: this set-up information is then sent to the appropriate add-in board when the machine is turned on. From that time on, the system knows exactly which boards are installed, which greatly simplifies reconfiguring the machine and diagnosing trouble because of the identification number.

The bus also allows users to create a "restore" file on disk, which duplicates the information in the setup RAM. This way you could set up your system easily in case you remove the system battery or it fails.

Alternatively, users could completely configure and set up one microcomputer, then insert the boards and use the restore file to ensure that all machines have exactly the same setup - something that's difficult to do on current AT-type machines, considering the multiple DIP switches on most add-in boards.

One crucial element is that each type of board should have its own identification number. In the basic plan, 32.000 such numbers were reserved for IBM, and 32,000 were reserved for independent developers. IBM has acknowledged that developers have had trouble getting through to register their numbers, but said that such problems have been solved.

Basic Workings

The MCA was designed to be completely processor independent, relying instead on a default timing cycle of 200 nanoseconds, with extendable cycles for synchronous or asynchronous operations.

Heath emphasized that any processor could be used in the MCA - even non-Intel chips such as Motorola's 68000 family used in the Macintosh and Sun and Apollo workstations.

"Just about any processor will run on this bus - including non-Intel family processors," said Heath. "It is completely processor-independent."

The MCA also includes a special protocol for fast system memory, which allows the use of faster 80-nanosecond, zero-wail-state memory chips in the Model 80.

One major distinction of the Micro Channel is in the way it recognizes and handles interrupts - the signals sent by add-in boards to the central processor.

Both the IBM PC and PC AT use an "edge-triggered" interrupt scheme, meaning that any peripheral that wants to send an interrupt merely has to change the signal level from low to high at the beginning of the interrupt, Heath said.

In contrast, the Micro Channel uses "level-sensitive" interrupts, meaning that boards will hold the line active throughout the interrupt process. In addition, some interrupts have higher priority than others. This means that multiple interrupts can be active at the same time, with the system prioritizing and deciding which to operate on.

This makes it easier for add-in cards to share logic with the main system board, and it reduces the possibility of a signal getting lost or of a spurious signal (caused by a faulty board or by outside electronic interference) accidentally causing a problem.

Another, more immediate advantage of having interrupts with different priorities is that the number of communications ports has multiplied. The PC AT was designed to accommodate at most two such ports; the PS/2 can accommodate up to eight.

Multiple Device Arbitration

The prioritized bus arbitration mechanism on the Micro Channel allows multiple Masters, devices like processors that control their own memory independent of the main system memory. These could be input/output subsystems, graphics coprocessors, or even other central processing units.

The current implementation of the Micro Channel allows for up to eight Masters, in addition to the Direct Memory Address (DMA) "slaves," or devices that do not control their own access to memory, but instead rely on a DMA controller chip.

Printers and communications devices are usually interrupt driven, but would be more effectively used as Masters, Heath said. This could lead to concurrent processing - or, more likely, intelligent subsystems, such as an intelligent disk controller or communications board, perhaps with built-in caching or file encryption.

Assigning these peripherals Master status means that the main system board can act as an "executive," controlling the activities of other processors being used on the system.

Since the MCA allows a number of different devices attached to the system, each working independently, it also needed a way of prioritizing the interrupts from all of these devices - in other words, deciding which requests for bus access are the most important and deserve to be processed first. In MCA jargon this is called arbitration.

Through arbitration, the Micro Channel looks at which DMA devices have requested interrupts and gives the go-ahead to the device with the highest priority.

As part of this scheme, interrupts only go into effect when they are confirmed by both the MCA's hardware and controlling software.

The Micro Channel currently allows for 16 levels of arbitration between devices on the bus. Eight of these levels are assigned to various DMA devices: seven are "reserved" for future use; and the system board processor accounts for the lowest level. In addition, the system board controls two higher levels of arbitration for error conditions and memory refresh, which are not available to devices on the bus. Two of the currently assigned DMA channels provide what is called "virtual DMA," meaning they can be reassigned among various devices, thus leaving open the possibility of extending the Master concept through many different levels.

For example, such a hierarchical system might permit users to build an "expansion box" containing up to 16 processors or linking to even more expansion boxes. However, Heath said this would require very sophisticated control throughout the system.

Although most devices can accomplish a "transaction" in the single memory cycle granted to them when their turn comes up in arbitration, some devices require multiple cycles to transfer blocks of data. For these, the MCA includes a feature called "burst mode," which allows a device to use multiple cycles.

For example, burst mode allows a disk controller enough time to access multiple disk sectors in one pass, allowing for a 1:1 interface on the disk drives. Combined with the built-in caching scheme, this allowed IBM to use slower hard disk drives without losing speed. In addition, IBM claims that the slower drives are ultimately more reliable and longer-lived than the faster drives.

To manage all this, and to ensure that all devices can get access in a timely fashion, the Micro Channel uses a "fairness" algorithm. The allows devices to use the burst mode, but makes them wait after they get their turn until all other devices have had a shot, regardless of priority.

Fairness and arbitration will be particularly important in moving toward a multitasking, multiprocessing environment in the future, said Heath. The current PC AT architecture allows for an alternate Master but has no burst mode and no fairness algorithm, according to Heath.

Tomorrow's systems could allow up to 16 processors with the ability to arbitrate, hut to do this you will need an operating system that supports "multiple threads" - in other words, multiple operations continuing at the same time. OS/2 will be one of the first operating systems for microcomputers that allows this.

Reliability

Heath said that several Micro Channel features should make systems based on it more reliable. The number of signals in and out of chips is reduced, thus aiding LSI design.

For example, one issue that often crops up in the IBM PC or PC AT architecture involves how systems react if they receive extra interrupts from a device, due either to a bad board or extraneous electromagnetic interference. Heath said that the MCA has resolved that issue. Unlike the old bus, the Micro Channel can check the state of a board at any time and get a positive acknowledgment of which board sent a given signal. If a board is producing bad signals, the Micro Channel can detect this and report the error.

Similarly, the Micro Channel should be able to detect and then map out a bad sector of memory, allowing you to continue computing even if you have a bad memory chip.

All these features should make it easier to run diagnostics on machines based on the Micro Channel, according to Heath, who cited this as a major reason IBM was able to drastically reduce its maintenance contracts for the PS/2 as compared to similar contracts for the PC AT.

"Obviously, the best of all possible worlds is that problems don't occur," said Heath. "But the next best thing to that is being able to pinpoint exactly what went wrong."

Possible Implementations

The Micro Channel Architecture makes possible innumerable variations on existing PS/2 hardware, according to Heath.

"Users are going to build things on their computers that you wouldn't have dreamed could be done on a PC," said Heath.

For example, by using a "sleep" signal, users could design a system that contains two identical add-in boards, each working independently. Then, if one board failed for some reason, the system could put it to "sleep" and let the other board continue working. This would allow for fault-tolerant computer operations.

Another option would be to put into a computer multiple concurrent processors, each working simultaneously. For instance, you might be able to put a number of plug-in boards, each containing 186 chips plus their own memory, into a Micro Channel. Even paying as high a price as $2,000 per plug-in board, users could get a nearly a 1 MIPS/$1,000 ratio, a vast improvement over today's average .25 MIPS/$1,000, Heath said.

Room for Growth

Above all. Heath emphasized, the Micro Channel gives the PC industry room to grow in.

"Just in case IBM didn't think of every thing - and we know we didn't - we kept a lot of the potential in reserve," said Heath. "We're waiting to see what ideas the industry can give us with the 16-bit and 32-bit implementation of the Micro Channel."

In 1983, when work on the MCA bus began, Heath said. IBM didn't have the answers, but only knew the problems and limitations of the old architecture - an architecture that technically left IBM no room to grow in.

"Having the knowledge that a solution is required and having the solution itself are two very different things," said Heath. "But never again will we make the same mistake of not allowing ample room for technical growth."

With the Micro Channel. IBM hopes to have an architecture that is applicable not only to the PS/2, but to future generations of machines as well.

Mix-Up of PS/2 Board ID Numbers Resolved

Earlier this summer, there was a furor in the PC industry following reports that IBM was dragging its feet on giving out identification numbers for the Micro Channel. Some board makers even charged that IBM was refusing to give out ID numbers so that only IBM-made boards could be used in the PS/2 family (see "Micro Channel IDs Could Delay Add-Ons," June 8).

However, Chet Heath, the senior design engineer of the Micro Channel Architecture (MCA) said that was nothing more than a misunderstanding.

According to Heath, what happened was this: Third-party board makers were told to call an 800 number to get an ID number assigned to the PS/2 boards they were building; when they called the number, they got what sounded like an answering machine message followed by a beep. At that point, most callers then left messages detailing their names, phone numbers, and requests for a PS/2 ID number. Unfortunately, said Heath, it was not an answering machine but a recorded voice asking them to hold the line until an operator was available,

"And of course they never got a return call from IBM because IBM never got any message," said Heath, who added that he himself had called the number and made the same mistake.

IBM has already taken steps to make getting an ID number easier for third-party board makers, said Heath, who emphasized that IBM never meant to "assign" the numbers, but only intended to assist in making sure that the ID numbers were not in conflict with one another.

Of the 64,000 possible ID numbers, IBM has reserved 32,000 for itself, saying that the remaining 32,000 belong to the industry.

"We could have been greedy and kept those numbers ourselves, but we reserved 32,000 for non-IBM developers," said Heath.

Also contrary to other industry rumors. Heath emphasized that the reason for the ID numbers was not to allow PS/2s to discriminate between IBM and non-IBM cards for proprietary reasons.

But because of the initial problem with readily getting an ID number from IBM, many board makers just went ahead and either made up a number themselves or looked at IBM-made PS/2 boards and used the ID number assigned to them.

This is not a good idea, according to Heath, "I can't predict what might happen if board makers do this," said Heath. "This is not good design practice."

Heath said that if board makers simply keep trying, there will be no problems getting an ID number assigned.

"It's been a busy phone number, but if board makers persevere, we'll assist them in finding an ID number that no one else has used," Heath said.

— Alice LaPlante