GCSF does MWJ (Mac Weekly Journal). MWJ is a weekly eZine that does in depth analysis of what is going on in the Mac industry. They go into far more detail on subjects than magazines, and always offer lots of supporting links and details. I subscribed to MDJ and MWJ for the last couple years, and I really look forward to it, and think it is easily worth the $7.50/month.
Intel Corporation jarred the worlds of both PC makers and peripheral manufacturers this past week. At the semi-annual Intel Developer Forum, this time held in Palm Springs, the company announced that seven new core companies would cooperate on the USB 2.0 specification aimed at creating a faster version of the Universal Serial Bus.
Hewlett-Packard, Lucent, and Philips are joining the original USB Four (Compaq, Intel, Microsoft, and NEC) to create a new bus, intended to reach speeds between 120 Megabits per second (Mbps) to 240Mbps. USB 2.0 is to be both forward and backward compatible with USB 1.1 peripherals. USB 2.0 aims to use the same connectors and cables as USB 1.1, touted by Intel as "ubiquitous" in the computer industry today, but provide new peripherals with bandwidth of up to 240Mbps--a speed faster than wide SCSI-2 or narrow ultra SCSI-3.
Even with this news, Intel affirmed the future of the IEEE-1394 standard, marketed by Sony as "i.LINK" and by Apple, its inventor, as "FireWire." Intel notes that 1394 typically targets consumer electronics devices like camcorders and televisions, but USB 2.0 will reach the "full range" of PC peripherals, including hard disks, video conferencing cameras, high-resolution scanners, and more.
It sounds like a perfect solution--until you start looking beneath the surface. If you do, you'll find Intel's announcements about USB this week to be completely at odds with everything they've previously said about both USB and 1394. You'll find a set of technical claims that are, at best, questionable. You'll also find a plan for a technology that, even if delivered as promised, doesn't match what 1394 can do today. And you'll find plenty of reasons why USB 2.0, when it arrives in eighteen months or so, may not meet even these modest-by-1394-standards goals.
USB Before This Week
Those interested in an in-depth explanation of the Universal Serial Bus may wish to refer to the recent MacCyclopedia article on the technology (MWJ 1999.01.31). Most of us have heard the marketing slogans about USB since at least last May, when Apple unveiled USB as the sole supported peripheral expansion bus on the iMac (MWJ 1998.05.11): up to 127 peripherals per bus, adding and detaching peripherals while the computer is on, simple cables that only connect in supported ways--and, most importantly to PC makers, a simple way to attach peripherals without requiring slots or port configuration, known in the Windows world as "tech support job security."
But as an engineer we know is fond of saying, everything involves trade-offs. Intel and its partners got an inexpensive bus to implement by limiting what it can do. Specifically, USB protocols are intentionally simple to reduce the logic required in USB support chips. USB's choice of inexpensive cables and connectors also limits how fast the bus can run, as higher-speed signals eventually require heavier and more expensive wires, along with shielding to reduce interference from other wires in the same cable.
Given the choice between "inexpensive" and "all-encompassing," USB's designers chose "inexpensive." This is made clear throughout USB material available from the technology's creators as well as in the USB 1.1 Specification itself. USB 1.1's top speed of 12Mbps per second is more than adequate for the design goals as outlined in the specification, including a primary goal of connecting the PC to the telephone. Regarding port expansion, the specification says, "The lack of a bidirectional, low-cost, low-to-mid speed peripheral bus has held back the creative proliferation of peripherals [... USB is] consistent with the requirements of the PC platform of today and tomorrow." (Section 1.1, page 1.) The "Application space taxonomy" in Chapter 3 of the specification describes "high-speed" devices as those that need high bandwidth, guaranteed latency, and ease of use, citing "video" and "disk" as examples. Those devices are specifically and deliberately excluded from USB's design goals (section 3.2, page 12).
Did USB's designers think high-speed peripherals weren't important? Not at all, but there was already a technology to handle those: IEEE-1394, better known as FireWire. A note in Apple's USB Developer Documentation summarizes the concept: "High-speed in the case of USB is not comparable to high-speed devices on a FireWire bus. USB is a complementary technology to FireWire, not a competing technology. USB enables the use of affordable higher-speed consumer grade peripherals on Macintosh computers."
Since Apple invented FireWire and was not part of the USB consortium, you might be disinclined to take the company's word on USB design goals. In that case, try taking Intel's. Intel was the chief mover behind USB--the "usb.org" domain is actually registered to Intel. As it turns out, Intel and Microsoft have been pushing the line adopted in Apple's documentation--USB for low-impact peripherals, 1394 for more demanding devices--for several years now. 1394's additional features make it less-suited for very low-speed devices like keyboards and mice, but USB's catering to those standard computer elements limits its effectiveness at higher speeds.
This is not some theory that Intel and Microsoft kept hidden, either. In a white paper from two years ago, Intel's "evolution of the platform" to include "visual computing" depended heavily on 1394 for the bandwidth USB was never intended to provide: "USB and IEEE 1394 are complementary technologies that will provide low-cost and high-performance connectivity options to makers of peripherals." Microsoft, on its 1394 Web page, is equally clear: "The IEEE 1394 high-speed serial bus complements USB by providing enhanced PC connectivity for a wide range of devices, including consumer audio/video (A/V) components, storage peripherals, other PCs, and portable devices."
Also, many industry observers clearly understood the Intel-Microsoft plan to use the strengths of both USB and 1394 where appropriate. A NEWS.COM story from November 1997 talks about Intel's commitment to 1394 at Comdex, since it is so much faster than USB. Even before that, NEWS.COM reported that Intel was bringing developers up to speed on 1394 at the September 1997 Intel Developer Forum.
Most tellingly, when Intel introduced the "Basic PC" last year to define what should be in a sub-US$1000 computer, the specification included both a USB and a 1394 port, even though features like modems, audio playback, and DVD decoding were left as software tasks to run on the "Basic PC's" microprocessor (a Pentium II, of course).
Don't convince yourself that we've beat this horse to death until you see what happened this week. Many Macintosh users haven't closely followed the proposed future of the Wintel platform, so it's important that you understand how clearly this two-tiered serial bus specification is accepted in the industry. It's not a theoretical, pie-in-the-sky concept. USB for lower-impact devices and FireWire for the most high-performance devices has been the plan for at least two years, and was openly discussed long before that. USB's 12Mbps top speed is less than a third of the original SCSI speed of five megabytes per second (MBps), and its design was never intended for it to fill that need.
Until this past week.
Now Intel wants you to forget most of that two-tiered serial bus stuff. As it turns out, with the advent of USB 2.0, the Intel-designed bus can apparently do it all.
Intel vice president and Desktop Products Group general manager Pat Gelsinger introduced USB 2.0 in his keynote speech to the Intel Developer Forum on Tuesday. Many technologies defy an accurate one-line description, but USB 2.0 does not: it's USB 1.1 at ten to twenty times the speed with full compatibility. Intel's backgrounder on USB 2.0 actually claims "forward and backward compatibility with USB 1.1," but it's not clear what that actually means. Backwards compatibility with USB 1.1 means older devices work without hitches on the new USB 2.0 buses, and that's clearly a goal. "Forward compatibility" could imply that USB 2.0 devices should work on USB 1.1 buses but not at the faster speeds, something no one seems to be claiming.
Existing "low-speed" USB devices, the ones that use the 1.5Mbps speed with its even lower associated implementation costs (such as input devices), will continue to operate as USB 1.1 devices on a USB 2.0 system. Since everything works together, computers will still need only one USB port, not one for version 1.1 and one for 2.0. You should be able to mix and match peripherals as you like, as long as all of them at least adhere to the USB 1.1 specification. And while USB 1.1 is designed as a desktop standard, USB 2.0 is to provide power-management function "to allow aggressive management of I/O power consumption," according to the backgrounder.
At eight bits per byte, 240Mbps delivers data at 30MBps, faster than most SCSI implementations in Macintosh computers today, although not all of that space is likely to be available to a hard disk if multiple devices are active at once. You'll recall that 90% of USB 1.1 bandwidth is guaranteed for isochronous transfers--time-dependent data from sources such as video and audio controllers. 10% of the bandwidth is guaranteed for bulk transfers, like those from storage devices. Of course, guaranteed does not mean reserved; in both cases, the specification only insures that at least the given percentage is available for the stated purpose. With 1394, the percentages are 20% guaranteed for the rough equivalent of bulk transfers and up to 80% available for isochronous data.
Intel wants USB 2.0 out as quickly as possible. In his speech, Gelsinger issued a "challenge" to have the "1.0 specification" (meaning the final USB 2.0 specification--the preliminary version is, for some reason, called the "0.9 USB 2.0 specification," as if these folks don't understand that you don't need multiple version numbers for one item) ready for the next Intel Developer Forum in September of this year. That's widely regarded to be ambitious, and Gelsinger knows it. Right after issuing the challenge, he said, "Yeah, I see my Intel guys are cringing as I raise the bar on them." In the public material, Intel promised the "0.9" version of USB 2.0 by September. The seven companies leading the effort expect computers and devices with USB 2.0 ports to show up by the second half of 2000, or roughly a year and a half from now.
In a separate but not unrelated announcement, Intel also revealed that it's working on extending its ATA architecture, used on both Wintel and current Macintosh computers for hard drives and CD-ROM devices. The 33MHz version of ATA is today's standard, but a 66MHz version is supposed to become common in the second half of 1999. Intel now wants to make a 99MHz version, capable of delivering up to 1GB per second for internal storage devices.
Combined with USB 2.0 for external peripherals such as hard drives, Intel doesn't see much need for 1394 anymore, though they're certainly not declaring it dead yet--just pushing it into a niche. Here's what Gelsinger said about 1394 in his speech:
In the area of consumer electronics, we believe 1394 has a role as the digital convergence pipe to connect the PC to the consumer electronics world in the area of audio or visual high performance devices. We believe that the PC can add tremendous value to consumer electronics devices and that that's the role that we are seeking to have by enabling 1394 into the PC platform. Broad deployment requires reasonable licensing terms around this technology for it to be accepted and deployed by the industry. In that sense, we're very happy with the progress that's been made recently announced. And we're looking forward to seeing those licensing discussions complete. And we believe that the PC can play a very valuable role in enhancing the consumer electronics experience over 1394.
In other words, IEEE 1394 is still important for connecting computers to consumer electronics devices, but USB 2.0 makes it unnecessary for most other uses. The backgrounder has similar text: "USB 2.0 and 1394 primarily differ in terms of application focus. USB 2.0 will support the full range of popular PC peripherals while 1394 targets connection to audio visual consumer electronic devices such as digital camcorders, digital VCRs and digital televisions."
If this all sounds a little too good to be true, don't worry--it is.
Why USB 2.0 Doesn't Replace FireWire
There are several questionable assumptions and claims in the hoopla over USB 2.0, but for now, let's assume everything Intel is saying is true. Let's grant, for the sake of argument, that USB 2.0 will arrive on time with all the promised features. Even if it does, it doesn't match what IEEE 1394 can deliver to peripheral manufacturers today.
The main point of USB 2.0 is speed--but the developers can't even tell us how fast the new specification is supposed to be. Every announcement and discussion to date says "120Mbps to 240Mbps," or "ten to twenty times faster." If they only get it up to 120Mbps, USB 2.0 would top out at 15MBps--still faster than many SCSI implementations, but entirely unremarkable in today's high-speed peripheral world. Michael Teener, co-inventor of IEEE 1394, prepared a technical response to the USB 2.0 announcements that circulated to members of the 1394 Trade Association this week. In it, he said, "The idea of USB 2.0 for external storage is silly. [Storage device manufacturers] don't want to use 200Mbps 1394 because it's too slow, so how is 120Mbps/240Mbps USB 2.0 going to help?"
He has a point. On today's PCs, including 1999 Power Macintosh G3 models, 1394 runs at 400Mbps. Even if USB 2.0 delivers everything it promises, it's going to provide 60% of that speed in a year and a half. Most observers agree that 1394 will reach 800Mbps this year, and will probably be available at 1.6Gbps by the time USB 2.0 hits the streets. If both technologies adhere to schedule, USB 2.0 (when released) will run 15% as fast as the latest 1394 implementations. That's not exactly compelling.
As noted in MacCyclopedia, USB is designed as a low-cost peripheral bus. A Universal Serial Bus does not function apart from the computer that controls it. For example, any USB video cameras on a bus can't talk to any USB video recorders unless the computer is both powered on and directing all the traffic.
That's not true for 1394. FireWire supports direct peer-to-peer transfers, meaning devices can talk to each other without requiring a "master" device such as a computer. We'll explore the difference this makes throughout FireWire in an upcoming MacCyclopedia article, but Teener rightly points out, as does Intel, that consumer electronics devices aren't going to switch to USB 2.0 without peer-to-peer transfer. When Sony makes two 1394 video devices and markets them under the "i.LINK" brand name, they don't really want to tell consumers that they can't hook them together, and get digital video into a digital recorder without using a computer.
True Isochronous Transactions
Teener told Electronic Engineering Times that USB is not "truly isochronous" (we corrected EETimes's spelling) because it depends on a client-hub architecture. The loose definition of "isochronous," still an unfamiliar word in most circles, as "time-sensitive" leaves this concept a bit unclear. Data from devices like hard disks can transfer a little more slowly as long as it transfers perfectly--errors are not acceptable, and any transfer that has errors should be retried. Isochronous data, on the other hand, is time-dependent data from a CD player or video camera--if the data is bad, you don't want the last bit of music or video resent. You want the devices to just get on with it so there are no blips in the media stream. It's far more important for an audio CD to keep playing than to get every bit correct. A hard disk is exactly the other way around.
USB can, in fact, deliver isochronous information. What Teener is saying to EETimes is that, since USB delivers data only from peripherals to host or from host to peripherals, it's not possible to achieve true isochronous transmissions between devices even with a computer powered on. Your USB CD-ROM drive may handle the isochronous delivery of audio data to the iMac just fine (remember, the 1999 Power Macintosh G3 computers do not, as yet, support isochronous USB transactions, MWJ 1999.02.06), but the iMac may not then be able to turn the data around and deliver it to USB speakers without introducing skips or blips. 1394 buses don't have this problem, since peer-to-peer transfer allows true isochronous transfers between any two nodes.
Actually, all 1394 isochronous transmissions are broadcast--any node that wants to listen may do so. That's how Steve Jobs demonstrated one FireWire video camera broadcasting the same signal to two 1999 Power Macintosh G3 machines at once during his Macworld Exposition keynote speech. USB doesn't support this at all.
Timing Is Everything
Most importantly, 1394 does all these things today, starting at 100Mbps and going up to 400Mbps. These are not proposed capabilities for a year and a half from now; you can use them right now on a 1999 Power Macintosh G3 machine. There's no need for any manufacturer needing these capabilities to wait for eighteen months to find a bus that implements these features. Even the "low-power" modes for portable needs are already present in 1394--either fully approved, or very close to final.
Even accepting everything Intel and the USB 2.0 consortium say about the new bus as true, the only advantage USB 2.0 offers is a proposed lower cost structure--but as we'll see, that's far from given. And it makes little sense to wait 18 months to develop a peripheral when a stronger, faster, existing bus can handle it today.
Why USB 2.0 Claims May Be Optimistic
Even if everything works out as announced, USB 2.0 is a less than compelling solution for the "wide range" of peripheral Intel says it should host, especially when compared to today's 1394 buses. But the best laid plans often go astray, and there are significant technical issues standing between reality and Intel's promises.
Intel's director of I/O architecture, Richard Coulson, told EETimes that USB 2.0 implementations will add less than one thousand gates to existing USB chips. Client chips, that is--the controllers for devices that don't support other devices attached to them. Michael Teener, among others, is uncertain this can be accomplished. Once USB starts approaching the same speeds as 1394, the devices have to deal with the same high-speed design decisions--device logic, information buffers, and the internal device logic to handle it. Teener says these kinds of functions, common to any fast device no matter what its communication interface, are "the vast majority of the typical interface part." He claims the difference between the cost of building a USB 2.0 part and a 1394 part will be "just about zero."
However, that's just for clients. Devices like the Apple USB Keyboard, and the ports on any computer, are USB hubs. Those parts have far more requirements in USB 2.0 than they do in USB 1.1. Mixing and matching higher speed USB 2.0 devices with existing USB 1.1 devices means incompatibilities on the bus. For example, even though the connectors match, you won't be able to hook faster devices to a slower USB 1.1 hub--it won't know how to communicate with them. Newer hubs can talk to older devices, but that means each hub must contain duplicate logic--all the new USB 2.0 speeds plus the existing USB 1.1 speeds of 1.5Mbps and 12Mbps. As EETimes put it, "USB hubs and hosts will add significant complexity to deal with rate matching that will require FIFOs to compress or buffer data." (A first-in, first-out circuit, or FIFO, accepts information as it arrives and holds it until a destination is ready to accept it, making sure the first information it received is the first information retransmitted.)
Teener more specifically questions this philosophy: "This means [USB 2.0 hubs and hosts] have to have FIFOs and store-and-forward logic, sort of like an ATM switch, only with variable length packets. There is a bunch of special routing logic and special magic that hasn't been figured out yet that will make it even more complex. My understanding is that this will be MUCH more complex than a 1394 [physical layer]," even one that handles both the existing 400Mbps 1394a standard and the proposed 1.6Gbps 1394b standard. By contrast, 1394 transmits no high-speed messages between two nodes if all intermediate nodes don't support the specified speed. That requires more user knowledge to configure a 1394 bus optimally, but rearranging a few devices may be worth keeping the cost down. Even though USB is supposed to be cost-sensitive, USB 2.0 hub makers won't have these options.
Intel's Coulson is optimistic that advances in chip-making technology will absorb the extra costs associated with these more complex hubs, but there is no guarantee of that. Teener also believes the differences in the protocols will require the silicon to be so different between USB 1.1 and USB 2.0 that while one part may handle both, such chips will internally have two entirely separate data paths. That can't make things cheaper.
Even worse, Teener says that USB's existing low-cost cable design probably can't handle 120Mbps transmissions, much less 240Mbps. His sources say that radiation problems are already affecting USB cables at 12Mbps, and that the cheap USB connectors Intel promotes as an example of USB economy (US$0.13 each as sold in Taiwan) do not even meet the existing USB 1.1 specification. If Teener's sources are right, upgrading to USB 2.0 might require users to purchase new cables if they're connected to any high-speed devices.
Intel needs a lot of lucky breaks to make USB 2.0 as cost-effective as USB 1.1--even though the fully-featured 2.0 bus doesn't match today's 1394 implementations. That should be a huge question mark for any peripheral or computer manufacturer.
Weirder still is the strange situation created by Intel's continuing support of 1394 as a "niche" bus. Intel maintains that computers should have 1394 ports, and while the statements of this week partially imply that's only for high-end PCs, the company's actions tell a different story. Gelsinger himself said that 1394 will become "more ubiquitous" in both computers and consumer electronics devices through the year 2000. And not only did Intel's "Basic PC" of last year have a 1394 port, so did all of Intel's Pentium III-based "designer PCs" showed at the Developer Forum this week to great fanfare.
Intel and its allies are standing behind Gelsinger's position in his speech: 1394 is the way to connect consumer electronics devices to computers, but it's not going to be the best way to connect a broad variety of peripherals. Intel's backgrounder goes so far as to say that 1394 is targeted at consumer electronics, giving it a different focus than USB 2.0.
Unfortunately, this is a blatant lie. 1394 has always been targeted at a wide range of devices; Intel and Microsoft both know this and have embraced it until last week. One of Microsoft's 1998 Windows Hardware Engineering Conference ("Winhec 98") presentations included a significant section on 1394 hard disk storage, for example, including information on "migrating" hard disks to 1394 from existing ATA and SCSI standards. The 1394 Trade Association's specifications page includes preliminary device commands for music appliances as well as audio and video, the "consumer electronics" niche Intel now says 1394 is aimed at.
In fact, Microsoft spent quite a bit of time at Winhec 98 promoting 1394, as you can see from several presentations available online, including introductions, explanations, and one that discusses migrating ATA-based hard disks to IEEE-1394 in the near future.
Microsoft has previously pledged support for 1394 hard disks in Windows NT 5.0, in Windows 2000, and as an add-on to Windows 98; Windows 98 already has 1394 support for some other kinds of devices, including video cameras. Apple, of course, intends to support as broad a range of FireWire peripherals as they can manage--and probably hasn't been in any hurry to implement isochronous USB support on FireWire equipped Macs since 1394 handles that chore better.
Given support in both the mass-market operating systems for 1394 storage devices, and given that Intel still encourages 1394 ports on personal computers (even the brand-new "FlexATX" small motherboard design Intel introduced last week includes power lines specifically for 1394a devices), why would anyone need USB 2.0? They'll already have a port that does everything USB 2.0 does, and more, without having to wait. The only way this doesn't make sense is if Microsoft reneges on its 1394 support, a move that would certainly be strange for a company trying hard to make inroads into Apple's media-creation market.
Microsoft is, in fact, giving indications of an about-face. Windows hardware program director Carl Stork told EETimes that USB 2.0 "creates a challenge for 1394 proponents to make the interface compelling." This, again, in spite of the fact that USB 2.0 in 18 months offers nothing that 1394 doesn't offer today. Intel and Microsoft have been known to try to declare technologies as standards before. Sometimes it works (as with ATA), and sometimes it doesn't (as with Video for Windows). The problem is that it's impossible to predict which attempts will succeed.
Unless Microsoft backs out of its 1394 promises, however, it would seem that 1394 has more than one leg up on the proposed USB 2.0. Intel's proposal offers no new features, no performance benefits, and no time-to-market advantages over the existing IEEE standard. The only benefit that can't be dismissed out of hand is cost, and it's far from a certain win for USB proponents. Even stranger, Intel is advising that USB 2.0's main competition, 1394, be included on every computer.
It's truly bizarre, and we've only found one interpretation that makes sense.
What Is Intel Up To?
Go back and re-read the excerpt of Gelsinger's speech earlier in this article, and you'll notice something rather diplomatic. He raises the problem of 1394 royalties, but never mentions Apple Computer--the inventor of 1394 and the collector of the royalties. Although Apple has refused official comment on the issue, enough sources have told enough news outlets that the company is now seeking royalties of US$1 per 1394 port from chip and system makers incorporating 1394 into their products. According to EETimes, which has not been on Apple's side in this story (even running an editorial blasting the company for the alleged fee structure), Apple is now seeking US$3 in royalties on a 1394 chipset that sells for less than US$5.
We noted in MWJ 1999.01.23 that sourcing on these reports is anonymous, and that Apple denies anything has changed. It raised for us the possibility that some of Apple's competitors--perhaps like Intel--are being charged higher licensing fees than those cooperating with the Macintosh platform. We note now that Intel is a leading manufacturer of PC support chips, including the kinds of circuits that drive technologies like 1394 and USB. Intel has done well with USB technologies, but this week admitted that after years of advocating 1394, the company is not including 1394 in the chipsets it sells for personal computers.
That's not surprising if there's a licensing fee problem. In the USB 2.0 backgrounder, Intel makes the point none too subtly: "As with USB 1.1, members of the core promoters group do not intend to charge royalties for essential patents required to implement the USB 2.0 specification." Even if they did, of course, Intel wouldn't have to pay royalties for patents it owns on its own bus.
As long as 1394 royalties remain cheap, Intel has every reason to support a well-defined, widely-praised IEEE standard for high-speed serial access. But the 1394 devices beyond consumer electronics aren't here today, and the royalty structure may have gone up intolerably. If you're Intel management, what do you do? You try to get Apple to lower the royalties again--and that's not going to happen if Apple feels 1394 is your only reasonable alternative. You need leverage against Apple, to convince them that not only do you not need 1394, but that your support is critical to making 1394 succeed at all.
USB already copies some of 1394's best ideas, and Intel owns most of the USB patents. A new version of the bus wouldn't "beat" 1394, but it could certainly create FUD--fear, uncertainty, and doubt--for peripheral manufacturers and system makers. And FUD is all Intel really needs. Think about it. The USB 2.0 announcement, and the Intel-Microsoft spin that it makes 1394 "less compelling" or a "niche," sow confusion about which serial bus to choose. If the doubt kills 1394's chance as a mass-market technology, Intel's USB 2.0 is the only remaining alternative--and Intel sells lots of USB support chips. If both USB 2.0 and 1394 remain viable, Intel still has a royalty-free, medium-speed serial bus to push for all computers and all peripherals. If the market chooses 1394, Intel has continued to advocate it and has silicon ready-to-go if selling the chips becomes a wise move.
(We should note for the record that Dick Davies of the 1394 Trade Association disputes the "niche" spin. In an E-mail response to MWJ, Davies said, "Do not think Intel has said the convergence bus, 1394, is only a niche. That is an error--according to Intel itself. The Intel leadership sees 1394 as key to their plans despite lack of a decision to place it on the chipset in near future." However, Intel's Pat Gelsinger is quoted by EETimes, in the USB 2.0 article, saying exactly that: "We expect 1394 will play a role in connecting consumer electronics devices to the computer which can enhance those devices. But that's something of a niche, and it doesn't make sense to integrate that into our chip sets.")
What Apple Must Do
The real loser in any FUD war is Apple Computer. The company has committed to FireWire as the external expansion bus of choice on Power Macintosh G3 computers, and is expected to include FireWire ports on most computers going forward. For this strategy to work, there must be a wide variety of 1394 devices available for consumers to choose.
Apple was roundly criticized for restricting the iMac to only USB expansion, especially by critics such as Mike Langberg who said there wouldn't be enough peripherals available. Langberg had reason to worry--USB had been a standard feature in Wintel PCs for more than a year, but few USB peripherals were on the market. We suspect that's because it wasn't the best decision--unlike Apple, PC makers added USB, but left existing ports in place. Developers could make traditional peripherals and reach a huge audience, or USB peripherals aimed at a smaller audience. The iMac changed the landscape: with it on the market, developers could reach the hottest computer of the year as well as most new Wintel PCs with a single peripheral. Without the iMac, USB wouldn't even have been in the position to threaten 1394--a fact we're sure Steve Jobs will consider in the future before adopting any technology that benefits Intel.
(Those who work with Intel would also be wise to note that this kind of unilateral redefining of a technology's entire purpose wouldn't be allowed if it was a real standard, controlled by a technology group such as IEEE or ISO. Hypocritically, peripheral makers are so wary of Apple that they eschew the company's technologies if they're not "standard," but they tolerate competitive abuse of totally-owned "standards" by companies like Intel, Microsoft, and Sun Microsystems. When Apple tries to leverage its intellectual property, it's "greedy." When Intel does it, it's "competitive.")
The iMac decision worked because PCs already had USB, even if it was underappreciated. The choice to restrict Power Macintosh G3 machines to FireWire requires similar dynamics, and was undoubtedly based on Intel and Microsoft's strong support for 1394 on all PCs. But if 1394 becomes a "niche" for consumer electronics devices, Apple is in trouble. FireWire isn't suitable for professional users if there are no hard disks, scanners, RAID systems, and other high-end peripherals available for it. Without broad industry support for 1394 as a general-purpose bus, it becomes another ADB--a bus for Macintosh-only peripherals, aside from the consumer electronics devices Intel and Microsoft are still touting.
This is where Apple's vaunted system integration capabilities come back to haunt the company. We're very much in favor of companies like Apple profiting from inventions, particularly technologies like FireWire that took years to create and shepherd to acceptance. But Apple's need for FireWire to succeed is far more important to the future of the Macintosh line than any 1394 royalties are to the company's bottom line. One of the Mac's main problems continues to be a perception of isolation in the computing world, that standard computer hardware and software doesn't work on Apple's products. Even if fifty million 1394 ports shipped this year and Apple collected US$1 in royalties on each of those ports (and this estimate, by the way, is more than double the rosiest estimates we've seen), Apple would receive US$12.5 million in royalties per quarter. The company takes in at least US$1.6 billion in revenue per quarter, even during the leanest times. Each Macintosh sale lost due to fears of isolation or too few peripherals costs Apple far more revenue than lowering licensing fees would.
Because the Macintosh product line depends on FireWire, Apple gains more in the end by allowing more 1394 peripherals at lower revenues, even if that shrinks or eliminates income royalty. If dropping royalties to a quarter of current levels encouraged a fourfold increase in FireWire peripherals, Apple would win all around. If such a drop created a doubling of FireWire peripherals, Apple would lose royalty money but the Macintosh--and Apple as a whole--would still win.
Ideally, Apple would institute a new royalty structure that completely waives 1394 licensing fees for any device that works out of the box with Macintosh computers and say so on the box. Similarly, PC makers who include 1394 ports should pay only a minimal licensing fee, since each port encourages more peripherals, each of which could have a Macintosh advertisement on the box. Now that's leveraging intellectual property. Administering such a program might be difficult, but we believe the rewards would be well worth the effort.
As much as it pains us to say it, Apple needs to blink. Intel has enough market dominance to pose a real threat to 1394 just through FUD. Nothing about USB 2.0 is a compelling alternative to FireWire, but Apple is not popular among technology companies. The evidence strongly suggests that higher royalties and the possibility of cheaper, well-promoted alternatives is leaving developers skittish about 1394. Apple can't tolerate such a position. The company tried to profit from QuickTime 3 licensing fees in 1998 and almost created another disaster, as developers scrambled to investigate other multimedia technologies when Apple's new licenses required exorbitant fees or annoying on-screen advertisements. Fortunately, in that case, Apple came to its senses before permanent damage was one. Their window for similar disaster recovery with FireWire is closing rapidly, and there are alternatives with more benefits than a raw revenue stream.
Certainly the press and peripheral makers will say that Apple blinked, and that it bowed to pressure. Let them say it. A healthy and robust 1394 market is worth a few Apple-bashing articles that will probably get written about the company's greed if it takes no action. If every peripheral you'd put on a SCSI bus today is available for 1394 in three years, Apple will have the last laugh.
Reprinted from MWJ 1999.02.27 (February 27, 1999).
Copyright © 1999 GCSF, Incorporated. All rights reserved. All trademarks are the property of their respective holders and owners.