A Report on I2O and NGIO – Introduction and Summary of Findings
Introduction.
In celebration of the three-year anniversary of this thread, I undertook a detailed look at what has become of I2O, the exciting initiative jointly announced by Intel and WIND that prompted me to wonder then about its potential. My review focused not only on the status of the I2O roll-out, but how I2O dovetails with Intel's latest initiative, the Next Generation Input/Output. This would be difficult enough were it not for the recent formation of the Future I/O consortium. This group, led mainly by Compaq and IBM, is intent on derailing NGIO, first by promoting PCI-X as an upgrade to the PCI bus, and second by defining a seamless migration from PCI-X to Future I/O, the architecture that is directly competitive with NGIO. The resulting bus war has complicated the messages emanating both from Intel and the Future I/O group.
I don't think I could possibly sort everything out were it not for Intel's willingness to be open and frank about discussing these issues with me. After the surprise introduction of NGIO and the subsequent formation of the Future I/O consortium, nothing is exactly as it might appear on the surface. Any attempt to understand these dynamics must begin with an awareness of fundamental trends in computing that have reached the point of forcing a sea change in computing architecture. This coming sea change has let loose a torrent of proposals, claims and counterclaims, as titans in the industry attempt to reposition themselves and gain comparative advantage over rivals. While the war is between NGIO and Future I/O, I2O has been caught up in early skirmishes involving the X extension to the PCI bus, and eventual feasibility of NGIO.
For fear that important conclusions from my review will be lost in the detail of future posts, I decided to begin by distilling all the messages to facts, and summarizing them below. However, if you like intrigue, don't skip reading the subsequent posts.
Summary of Findings.
Three years ago, Intel proposed I2O as an open software standard for bringing intelligent input/output processing to servers. It took until the middle of last year for industry participants to complete specifications and developments necessary to begin producing I2O-compliant products. Last fall, Intel surprised the industry with the announcement of another related technology dubbed Next Generation Input/Output (NGIO). This announcement excited, confused, and in some cases, irritated the industry. I suspect that Intel's messages, prior to and immediately following the NGIO announcement, contributed to a feeling that I2O might be short-lived. Some mistakenly view I2O as a PCI-based I/O standard that will be displaced by the next bus generation, or if not by the next generation of bus-less computing.
But of course this is not true, and I was pleased to discover that no one at the I2O SIG or Intel is confused about the assigned role and importance of I2O. What I found out is that NGIO and I2O are more than complementary; they are synergistic. Neither absolutely requires the other, but each is enhanced by the other to the point that the whole is greater than the sum of the parts. In practical terms, this means that I2O's production roll-out is likely to receive a significant boost because of the introduction of NGIO. If the truth were known, I suspect that a successful roll-out of Future I/O also would depend as much on I2O – but don't expect this claim to be confirmed publicly by the Future I/O consortium anytime soon.
As a marketing manager from Intel's I/O Processor group indicated, “Successful initiatives take about four or five years to really take off, and they always seem to hit a lull at about three years. The PCI bus roll-out had this characteristic, and I2O seems to be behaving similarly.”
The typical lull he refers to is not a coincidence. Successful standards must always displace existing, often proprietary methods, which provide comparative advantage to entrenched vendors. Only when enough pressure builds up behind a new standard to blow away resistance by entrenched vendors will the replacement become established. Many companies already offer proprietary intelligent solutions, providing them added-value over lesser solutions. Further, large, established device makers have the capability to supply drivers for most operating systems, a capability they view as a valuable barrier to entry for less established suppliers.
Nevertheless, it appears that recognition and participation in I2O by OS vendors has achieved critical mass. OSM's are available for storage products for Netware, SCO Unixware, Windows NT and OS/2. Solaris for Sparc, as well as Intel-Architecture, either is available now or will be soon. After relaxing I2O licensing rules last fall, the Linux community has jumped all over I2O, as have hundreds of embedded developers, forming an activist sub-culture within the I2O SIG – about 350 joined the SIG within the last three months. At this point, few OS vendors can afford to ignore I2O by not supplying efficient Operating System Modules (OSM) for all device classes. (Microsoft has yet to supply one for LAN connections, mainly because Microsoft integrated the TCP/IP protocol stack too tightly in the OS, thereby negating much that can be gained from intelligent I/O LAN processing. Expect that to change soon to keep Netware from taking share with effective intelligent LAN adapters.) Of course, there are other OS vendors fully compliant with I2O, including WIND and, from prior press releases, INTS. As indicated above, the high-end embedded community is learning to appreciate I2O, if only because it rids them of the tyranny of the driver monopoly enjoyed by mainstream operating systems.
Another indicator of acceptance of I2O is i960Rx production. The I/O Processor marketing manager referred to above, stated that i960Rx chips are ramping nicely, and even the latest and greatest i960RM version will be in full production by next quarter. Intelligent RAID is a clear winner and seems to be doing well. High-end embedded network devices also seem to gaining momentum. He believes cost-savings from sharing I/O processors, once another application or two takes hold, should secure I2O in the mainstream of server computing. Further, he believes that I2O is certain to extend from the server and other network devices to the high-end Workstation.
The i960Rx production is only an indicator of I2O acceptance because some i960Rx processors use IxWorks with I2O messaging disabled or altered (usually for some legacy reason), and some specialized I2O implementations do not even require an I/O processor. And, of course, other I2O implementations use non-Intel processors.
An unfortunate consequence of the bus war started last fall is that I2O will be caught on the battlefield, and will take a few hits. For fighting purposes, I2O is being made lean and mean. Dubbed fixed or specialized I2O, this reduced version of I2O is designed to blunt criticism that NGIO requires I/O processors at both ends of the communication link. Actually, NGIO does not even require I2O, much less an IOP, but everyone I talked to accepts that the preferred route to NGIO is through I2O.
We need to know about the bus wars, and the mixed messages they produce. But the future will continue to follow the immutable trends in enterprise computing, so it is equally important to rise above the skirmishes and observe the sea change underway. Full-featured I2O is part and parcel of future network computing, the final destiny of enterprise computing.
Allen
I2ONGIOX – Search key to find all postings in this series |
