Tornado was not named after Intel’s Cyclone. The name, Tornado, signifies tremendous power (the universe of all tools) focusing on a single point (an embedded target with restricted, production level resources).
RE I2O, there follows my promised assessment of the background and potential for I2O.
Intel and WIND jointly announced an I2O initiative last winter, in which Intel will be shipping an I2O chip, the i960 RP processor, to become an industry standard for relieving Central Processing Units (CPUs) of many I/O duties. Each i960 RP will come with a copy of Wind River’s IxWorks. WIND has indicated that Intel will pay it a run-time license fee for every unit shipped. There is no doubt this is a big deal; the attempt here is to try to estimate just how big it will be for WIND. Intel was expecting to begin shipping samples about now, with production shipments beginning early this fall. WIND should begin seeing noticeable license fees during its fourth quarter.
1. I2O vs. peripheral input/output processing
Peripheral input/output processing has been around for a long time, particularly with mainframes, in which a peripheral processor aids the CPU in processing input and output. Network companies already make add-in cards to facilitate input/output in this way, including IBM with its TURBOWAYS ATM Adapter cards for the ISA, MCA, Sun’s S Bus, and for the RISC System/6000. Mylex makes the PNA960 PCI Network Accelerator, and Intel makes the EtherExpress PRO/100 Adapter Family for either Ethernet or Fast Ethernet. The EtherExpress PRO/100 uses the PCI bus and the PC’s CPU; while the EtherExpress PRO/100 Smart Adapter contains an i960 processor (not the new RP variant) on the card.
None of these products uses the new i960 RP processor, which has yet to ship in production quantities, and none implements I2O. All of the products save the (dumb) EtherExpress PRO/100 Adapter uses an Intel i960 flavored processor for off-loading I/O processing. IBM’s peripheral input/output products that run on the ISA and MCA buses cannot perform I/O at the same speed as those running on faster buses.
So what is I2O? There are four industry I/O problems addressed by I2O, of which only the first, off-load I/O processing from the CPU, is solidly addressed by existing products. Existing products may relieve the CPU, but they do not simplify the job of creating software drivers to connect other I/O peripherals to a computer, the second objective of I2O. Exactly how any I/O devices can be added to a computer system depends on many factors including the data bus, the CPU, the operating system, and all the other I/O devices that are, or might, be connected. A major thrust of I2O is to relieve the peripheral manufacturer from these problems, by abstracting I/O interfaces and eliminating some interactions that might cause conflicts.
Third, I2O will become an enabling technology permitting the seamless use of higher-level standards for data handling, particularly peer-to-peer communications. An Intel EtherExpress Pro/100 Smart Adapter might be able to talk to another computer using a Mylex PNA960 PCI Network Accelerator, but only using existing network protocols. With I2O, it will be possible to send high-level messages to compliant peers (of any manufacture) requesting more sophisticated I/O handling. For example, future multi-media clients might request content data streams in on-the-fly MPEG-2 or MPEG-1 format, depending on the client’s own capability to decode in real-time. Because there are so many possibilities for how data might be communicated, only through rigorous control of I/O protocol standards and protocols about protocols will the industry be able to realize the benefits of seamless connectivity. I2O will make that possible. I2O compliant devices will not (by agreement) implement proprietary I/O protocols as a means of product differentiation, meaning that all I2O devices will interoperate seamlessly (at least at the version level).
Fourth, I2O can implement significant portions of I/O consistently in software, e.g. protocol details subject to upgrade and revision. This allows organizations to deploy new technology earlier than they might otherwise, knowing that final protocol revisions can be broadcast throughout their network of devices as a straightforward and inexpensive software upgrade.
I2O is an open standard for implementing these properties. While other smart ways to off-load I/O have preceded I2O, one expects products implementing non-standard techniques to convert to I2O quickly, or fail in the market place for lack of interoperability.
In principle, and by design, it is possible to use up much of any computer’s CPU to implement I2O in software, thereby enabling even a PC to become a fully compliant I2O client without additional hardware. (However, if any high-level protocols have a hard real-time requirement, it would not be possible to implement fully compliant I2O on most desk top and server computers.) But just for the price of an i960 RP chip on the motherboard or an add-in card, I2O can be implemented efficiently, greatly enhancing the data throughput on any computer using the PCI bus. Look for other implementations in the future using other buses, but for now Intel has a giant lead by implementing I2O on top of its popular embedded processor for I/O, the i960, using the PCI bus.
The PCI bus has importance to at least the Intel implementation of I2O, because of its many advanced features, providing both adequate speed and bandwidth for the foreseeable future, and a robust way to interoperate with the CPU and other I/O occurrences simultaneously. These features of the bus enabled I2O to be implemented with minimal interaction with the native OS. Thus, Intel’s I2O chip can be installed on your x486 or Pentium computer, not to mention a Pentium Pro, or Apple, or whatever, as long as a PCI bus is available. This is why MSFT is on the periphery of I2O; they ultimately benefit from reduced driver work and reduced support costs, but they do not need to make changes to Windows or DOS for I2O to work. I2O might also work at this same level of abstraction with another bus; it is just that to date it has been implemented only on the PCI bus where the concept works perfectly.
2. How and when will I2O be deployed?
At a large analyst meeting in New York on May 24, Intel made it very clear they are now gunning for the $52 billion non-Intel Architecture (non-IA) server segment of the computer market. They already totally dominate the $161 billion PC segment, and can only grow that segment by growing the market, not market share. Even with a robust Window NT 4x OS, Intel will need serious data handling on Pentium Pro platforms to take over the server market. The speed factor and rich data type handling of I2O is now a critical component in Intel’s server strategy.
I2O will appear on motherboards and add-in cards starting in the 4th quarter 1996. The add-in cards will take the form of upgrades of what has been developed already. These will mainly consist of network adapter cards and disk controllers, especially RAID controllers. I2O will be on motherboards (e.g. the Tiva Microcomputer Corp) probably usually sporting multiple Pentium Pro processors, implementing RAID or other disk controllers using I2O. Intel also will be adding I2O to motherboards starting 1st quarter 1997. This means that in the beginning, multiple instances of I2O will be found on servers, being used mostly as network adapters and disk controllers. Later, some OEMs will figure out how to use a generic I2O on the motherboard, or an add-on I2O board, for specialized purposes, providing cheaper versions of their products for computers already blessed with an I2O capability. Shared I2O functionality will be particularly popular for client computers.
Even if we assume multiple instances of I2O processors will be common for Wintel servers, the total number of I2O processors that would be needed annually would be only in the low millions, since Intel’s forecast of NT Advanced Server units is small compared to desk tops.
Just as Intel initially targets servers for their latest processors, later to become the mainstay of PCs, so will I2O. Most of the benefits I2O brings to servers are just as important to client PCs, including high-level protocol APIs, peer-to-peer connectivity with rich (multi-media) data types and flexible implementations of leading edge protocols, not to mention the added speed. Even off-loaded I/O will be viewed from the start as a must-have for power users, since I/O bogs down desk tops, too. Intel expects that over 10 million NT Advanced Workstations will be shipped in 1997, each a candidate for an I2O processor on the motherboard or adapter cards.
While I2O adapter cards will work with Pentium and other non-Pentium Pro processors, it is instructive to project Pentium Pro numbers alone, for those processors are associated more with newer designs (at least with the PCI bus, if not an I2O chip), and their owners probably would be most desirous of high-speed I/O. Intel projects that the world-wide PC shipments for 1997 will exceed 80 million units. Pentium processors are peaking right now, and will begin to fall this year, being all but out of production by the 2nd half of 1998. Pentium Pro processor production will equal Pentium production by the end of the 1st quarter 1997. Thereafter, Pentium Pro production will continue to ramp up, at a faster rate than Pentiums, and not peaking until 1999. About one-half of the 1997 80 million PC units will be Pentium Pros. By 1998, by far the dominant portion of the 95 million PCs that will be shipped will be Pentium Pros.
Intel claims about 90% of US and European corporate PCs (clients like workstations) will be connected to networks. About 40% of Rest-of-World corporate PCs will be networked in the near-term.
Other possibilities for I2O abound. Most networked office equipment could benefit from I2O, as would virtually all communications equipment forming the intranet/internet. In many of these cases, the primary benefit would be the out-of-the-box standardized protocols, but these devices also have CPUs, and might benefit also from more efficient I/O processing. Commodity PC/TV’s or settop boxes could benefit from the rich data typing, ease of deployment, and standard, upgradeable protocols and turn to I2O for these reasons, if not for speed. When implemented in commodity devices, I would guess that I2O processing would be integrated with other processing requirements on a single affordable chip. Exactly what IxWorks role would be in this integrated environment is unknown, but it would at least open the door for WIND to be a major player for these devices.
3. IxWorks I2O run time license fees
A first hint of the acceptance by the industry of I2O, and a peek at the breath of coming applications, will be at the fall COMDEX. Thus, it is premature to attempt projecting I2O units with accuracy. Nonetheless, as investors it is useful to begin to estimate both the amount and timing of future I2O run-time license fees, expecting to refine our estimates as more information becomes available.
I have updated my model of WIND’s run-time license fees by (roughly) projecting by year every category of Intel-Architecture computers, including: World Wide PC Shipments w/o Windows NT, Percent Pentium Pros, Percent Pentium Pro non-NT using I2O, Windows NT Workstations, Percent using I2O, Windows NT Servers, Percent using I2O (assuming an average of 2 I2O Chips per server), and the Number of Network Communication Devices using I2O.
My best guess is that the I2O revenue stream will start this fiscal year, but be very small, about $500K. By the following year, the I2O revenue stream could jump to about $7 million, and accelerate from there. By FY 1999, and for a number of following years, I2O run time license fees could equal run time license fees from all other sources, with both more than doubling each year.
Allen |
