Picture the scene: you design and manufacture a very complicated line of computers. All the time your experts are beavering away to find how they can make things run faster and how they can make everything smaller. From time to time you build, and start to sell, new and better versions of existing components. This whole process costs you a lot of money and affects how much you have to charge customers for your products (and the size of the dividend paid to shareholders).

Now let's suppose you produce two quite different computer lines. That means you will have two teams of people doing exactly the same job, but the results of their work will be used on different computers. Then, one day, you wake up in the morning with a really bright idea. Why not see how many components for these two quite different computer lines are really very similar? And then - and this is the really clever part (as far as the finance department is concerned) - employ only one team of people to design those parts and use only one manufacturing plant to make those components. Think of the amount of money you could save!

Now take this idea one stage further. Imagine that you actually have three completely different computer lines. Think how much money your company would save by having a single component running on all three lines that was designed by a single team and built by a single manufacturing plant. Think how large your pay rise is going to be when you suggest savings of this magnitude to your boss!

Well, maybe this isn't quite how IBM's Project ECLipz came about, but, whoever thought of it, you can easily see the compelling financial logic behind such an idea. And IBM does have three distinct product lines, and has had for some time now. What was originally AS/400s, RS/6000s, and mainframes, is now System i, System p, and System z. (And I'm not even mentioning the x86 line of Blade servers etc.)

IBM's Project ECLipz has been around, at least, since 2001. The acronym is said to stand for Enhanced Core Logic for System i, System p, and System z. (The source of this expansion of the acronym appears to have come originally from UK-based consultancy Isham Research rather than IBM.) There are two reasons why I use the phrase "meant to stand for." Firstly, IBMers have always been a little coy about discussing its existence, while at the same time there have been suggestions that IBM is itself providing the leaked material. It's almost like the old cold war days. You can picture yourself in a John le Carré novel - perhaps "Tinker, Taylor, Soldier, Mainframer." Using "Moscow rules", the man from Rochester drops a parcel into a litter bin, and ten minutes later a reporter takes the package from the bin, slips it under his coat, and walks to his car! The second reason for saying what it's "meant to stand for" is almost equally as frivolous, but I'll tell you anyway. At one stage, the biggest competitor to the RS/6000 (System p) was Sun Microsystems. What was needed - from IBM's point of view - was some way to get ahead of Sun in terms of image and sales. What IBM wanted was a total eclipse of the sun! Some people are speculating that the idea of an eclipse came before the acronym. And, the story goes, the acronym was designed to fit with the desired thinking. (Again, various sources on the Web are attributing this idea originally to Isham Research.)

There has always been some discussion among people sharing rumours about Project ECLipz that IBM is really hoping to consolidate its different families of servers on to Power architecture. So, interest in Project ECLipz was re-ignited following presentations by IBM at the International Solid-State Circuits Conference in San Francisco at the beginning of February this year. A lot of the talk was about IBM's new Power6 chip. IBM has changed from using 90 nanometer wires to 65 nanometer wires, and, as a consequence is able to fit more transistors on a chip - 276 million of them, to be precise. Smaller, closer transistors mean faster performance. IBM is nodding and winking to suggest that system functions will be incorporated into the Power6 chip. Power6 is a dual core processor or dual-core module (DCM).

Although it would be quite possible to share other components between IBM's three families of servers, perhaps the Holy Grail would be to share processors. All this extra stuff that's squeezed into the Power6 chip might very well lead one to suspect that perhaps the chip could run z/OS, albeit in some sort of hardware-assisted emulation mode. Virtualization might be a way to run the z/OS operating system on the chip. CISC mainframe instructions and RISC hardware instructions would then be able to work together.

Even without being able to run z/OS, Power6 provides System i and System p users with almost double the performance they can expect from Power5 chips.

In terms of convergence, there is a growing similarity between System i and System p servers. It would appear that OS/400's underlying instruction set is similar in many ways to the intermediate code of the Power chip. Therefore, only part of the operating system needed porting for it to work on the chip. So once you get the operating systems running on the same chip and you get them to use the same hardware for everything else, then the actual cost of production of two different lines of servers is almost halved. In fact, the OS/400 instruction set is called TIMI (Technology Independent Machine Interface). TIMI is a virtual instruction set, which means it is not the instruction set of the underlying CPU. All user-mode programs are stored as TIMI instructions, so they can't use the instruction set of the underlying CPU. As a consequence, this ensures hardware independence. This is the reason that System i can share so easily with System p.

No one knows the actual figures here, but the cost of developing a new processor is quoted as being around $40 million. On top of that, IBM's X3 servers are reputed to have cost $100 million to develop. So any savings (sharing System i and System p components and development), particularly almost halving the cost, is going to look good on the balance sheet and would be an easy sell in the boardroom.

In fact, by 2001, the differences between the System i machines (at the time they were called iSeries) and the System p machines (likewise they were known as pSeries) were I/O processors and hardware multithreading. The iSeries had them and the pSeries didn't. The code-named ‘Squadron' launched as the i5 and p5 and was able to run multiple operating system partitions under the control of a hypervisor (which now had the functionality of the SLIC (System Licensed Internal Code).

Combining Systems i and p would appear to be very simple compared with what would be required if IBM was planning to create one chip to fit all its servers. There are a number of obstacles it would have to overcome because System z hardware is quite a bit different from the other two. It has SAPs (Systems Assist Processors) for I/O, it can have specialized coprocessors such as zIIP and zAAP, there's support for decimal numbers in packed and regional format, and support for hexadecimal floating point.

But even if IBM's plans aren't as ambitious as having a single chip that would work for System i, System p, and System z servers, even if the plan is simply to share the other parts, that would still show huge savings for the company. IBM could build almost the same box for all server platforms and just plug in the appropriate chip to the appropriate socket. I have heard it suggested that sharing memory, the chassis, ASICs, interconnects, etc could lead to savings of around 75%. And, hopefully, a large part of that saving would be passed on to consumers. It wouldn't make mainframes cheap by any means, but it could make them more affordable - particularly at the lower end (I'm thinking of sites that have not too many MIPS and are probably running VSE) where IBM must be losing customers to other server vendors.

Perhaps IBM should come clean and acknowledge the existence of Project ECLipz and give users a roadmap of how they see the project developing over the next five years. Maybe it's ambitions (if it exists!) don't include mainframe processors, just every other component - or perhaps they do. And if it doesn't exist, IBM should make that clear. But for a non-existent project, Google managed to throw up 136 entries when I searched on "Project Eclipz," and 1420 when I simply searched on Project Eclipz (no quotes). What's that phrase? No smoke without fire!

Trevor Eddolls is CEO of iTech-Ed, a UK-based company that specializes in consultancy and writing about mainframe-related topics. E-mail: trevor [at] itech-ed [dot] com. Web site: www.itech-ed.com. Read Trevor's blog at www.mainframeweekly.com.