Just recently, people have been talking a lot about Project ECLipz, and, I have to admit, I wasn’t really sure what they were talking about. Yet, it appears that this IBM project has been in existence since 2001 or thereabouts. And the reason behind the recent gossip about it has been the POWER 6 announcements earlier this year.
Basically, it seems, IBM has a goal to converge all its non-Intel server lines onto a single platform. Those non-Intel servers, of course, are what we know as zSeries, pSeries, and iSeries – and if you reverse the order you get the last three letters of the project (i, p, and z). In fact the whole acronym is meant to stand for Enhanced Core Logic for iSeries, pSeries, and z Series. Cynics among you will probably say that as Sun is a major competitor of IBM, they came up with the word “eclipse” and then made it into an acronym – you know,” total eclipse of the sun”.
Why would IBM want to converge their server technology like this – the answer is very simple, it saves money. Obviously, the development of new hardware has a cost, so if you can develop hardware that works for two types of server, that will cut your development costs in half. If you can split it three ways, well then it’s only a third of the cost!
If you look at the iSeries, you find that OS/400 runs on more-or-less a pSeries POWER5 system. This is probably the easiest convergence for IBM because, I’m told, the OS/400 underlying instruction set is quite similar to intermediate code, which meant that only very few parts of the operating system needed porting to PPC. In terms of costs, the hardware development is split in half, with the OS/400 people needing to do a little software development. Now that’s got to look good on the budget sheet.
The big problem for IBM and the ECLipz project is that zSeries hardware is quite a bit different from the other two. For a start there are all those wonderful coprocessors, there are System Assist Processors (SAPs) that are used for I/O, and there’s support for hexadecimal floating point and support for decimal numbers in packed and region formats. Not forgetting those go-faster stripes down the side!
POWER6 itself was announced in February and should be available next year. Some sources, outside IBM, are suggesting that POWER 6 includes some type of z/OS emulation through on-chip microcode that would create a CISC mainframe instruction from RISC instructions. What this means is that the POWER6 processor would be the first hardware able to support the three server lines.
Converging the server lines does make economic sense for IBM. For users of the servers, anything that helps reduce the cost has got to be a good thing. IBM needs to bear in mind that zSeries users, while paying a high price premium, do expect something that is very fast and extremely reliable. Those users won’t be happy if compatibility issues halt the inexorable increase in processor performance they have come to expect.
It just goes to show (as the old Chinese curse would have it) that we live in interesting times.