BUS WARS

by Wayne M. Krakau - Chicago Computer Guide, April 1995 - NewsWare, August 1996

ISA, MCA, EISA, VL, PCI - it's alphabet soup time again! First let's get to the definitions. A bus (it's supposed to be "buss", but I'll bow to current common usage) in a PC is the slot into which you plug circuit boards to add functions.

ISA means Industry Standard Architecture. In English, that means a specification taken from reverse-engineering the original IBM AT bus. (Since IBM never published a thorough description, reverse-engineering was required to derive this specification.) It is a limited bus, but it served the industry well as a standard for growth. It is now only used for limited speed applications such as simple serial and parallel ports.

MCA originally meant Music Corporation of America, but, after some legal wrangling (after the fact) by IBM lawyers, it also became the legitimate abbreviation for Micro Channel Architecture. This bus was IBM's attempt to create another proprietary standard, and, basically, screw up the PC industry for their own benefit. (Remember FUD - Fear, Uncertainty, and Doubt, IBM's unofficial strategy for larger computers?) The only thing it screwed up was IBM's PC division (which is taking massive financial losses). Even IBM has given up on this architectural dead-end.

EISA means Extended ISA. EISA was invented by a committee of manufacturers to combat IBM’s MCA. It has done very well in file servers and other specialty machines, but has not made many inroads in the regular PC (single-user and network workstation) market. If you need a machine full of high-speed slots, such as a file server, EISA is currently the only choice.

The VL bus is another committee design. It was meant to be THE next generation of bus, but has proven to be only an interim step. While its base performance is adequate (it often can outgun its main competitor, PCI), its lack of expandability has doomed it. Each additional VL slot puts more stress on the overall computer system, so your won’t see computers with more than 3 VL bus slots. This makes it acceptable for now, but not for the future.

The PCI bus is THE bus for the future. Even Apple has chosen it! The problem is that "future" is the operative word. Though it is theoretically possible to stuff a machine full of PCI slots, current Intel-based motherboards usually support only two. This limitation is built into the existing Intel chipsets. Some manufacturers have decided not to wait for Intel’s new chipset which will allow three PCI slots (still insufficient for servers) by tweaking their current Intel-based systems to support three. This means that anyone buying these computers is gambling on a kludgy modification to the Intel’s design. ("Kludgy", as in "kludge", an ad hoc change or patch - a quick, inelegant fix.)

Some motherboard manufacturers have rejected Intel’s chipsets completely and have designed their own. The viability of the resulting motherboards is dependent upon their accurate interpretation of the PCI specification - essentially an Intel invention. These manufacturers offer multiple PCI slots and even combinations of EISA and PCI in a single machine.

PCI has more problems. It has been said (with admitted exaggeration) that when adding a PCI board to a system that already contains one, you have about a 50% chance of having the combination work. Supposedly, this is not due to the PCI specification being inherently bad, but to board designers misinterpreting the definition! Great! That’s just what I want to hear when I’m on the phone to a technical support person while an anxious client looks over my shoulder, expecting me to perform miracles. All I need is to have one side say that the definition is ambiguous and the other side say the definition is precise, but that the interpretation is faulty. (Less filling - Tastes great!) I DON’T CARE! I just want these people to get their act together and make products that will work together every time.

Another PCI problem is with communications between the PCI interface circuitry and the Pentium interface circuitry on certain motherboards based on older Intel chipsets. Transparent data changes have been reported in these motherboards in the national press. "Transparent" in this case means data is being changed without any warning! That’s more scary than an outright system failure! I’m sure that the IRS will be thrilled to hear that your company is reporting a year-end total of $1.25 worth of withholding for your 100 employees! (Maybe this will scare people into using the integrity checking routines built into almost all modern accounting software.)

Note that this problem is separate and distinct from the more widely reported errors inside Pentiums. Unfortunately, I haven’t found a specific reference as to the exact model numbers or precise vintage of these "older" chipsets and motherboards.

An additional, though temporary, problem is the lack of availability of specialized server-oriented boards for PCI. These devices are currently made only for EISA systems. Luckily, that will change over time.

My current bus selections revolve around the limitations and real-world conditions that I have listed here. For servers, I usually specify EISA. I am in the process of switching to combined EISA/PCI systems, and will eventually use pure PCI.

The reason I am willing to use the "combo" systems, is that I will only work with PC manufacturers who design and build their own motherboards, regardless of whose chipsets they use. This gives me a way out when any question of compatibility arises. For example, my research into problems with the original implementation of Novell’s Netware SFT Level III (System Fault Tolerant, Level 3, using dual servers), was extremely shortened because I was able to have a discussion of EISA bus timing issues with one of the people who designed the motherboards that I was using. That conversation provided enough information to immediately eliminate motherboard compatibility from my list of possible errors and get on with researching other possibilities.

I am still selling mostly VL-based workstations, but am gradually increasing the use of PCI. I have never had (or even heard of) problems with the VL bus. You just put a board in, and it works - fast. If future expansion is vitally important and the client understands the issues mentioned above, then I am willing to sell PCI.

There are some additional motherboard issues that need to be brought forward. The first concern is that there are no standards for "sideways" slots of any type. Sideways slots use some type of riser card mounted in the motherboard using either a proprietary slot or a tweaked standard slot. The riser itself contains slots. Circuit boards are placed in these slots parallel to the motherboard. Machines with sideways slots have a long history of compatibility problems and are usually difficult and expensive to work on, so I discourage people from using them.

The second motherboard issue is the number of slots. The trend in computing is to add multiple complicated hardware-based features. These can include sound boards, accelerated video cards, speech cards, telephony-related cards, and many more. Because of this, I prefer to use computers with the old standard of eight slots. This leaves room for expansion and provides a way out if you need to replace the motherboard due to malfunction, obsolesence, or compatibility problems. This standard size motherboard can be replaced by almost any other standard-sized motherboard. The lack of a built-in video controller, disk controller, or even input/output ports means that in case of failure or obsolescence, you don’t have to purchase an overpriced proprietary motherboard when all you need is a single inexpensive component.

And you thought the only bus you had to worry about had a large picture of a skinny dog on its sides!

1995, Wayne M. Krakau