The OpenPower Foundation – great fleas, little fleas and industry standard servers

BiPolar to CMOS, CISC to RISC, RISC to Itanium, RISC to x86 – the big changes in server processors always promised cheaper machines through disruptive technology. It reminds me of the nursery rhyme:

“Great fleas have little fleas upon their backs to bite ’em,
And little fleas have lesser fleas, and so ad infinitum,
And the great fleas themselves, in turn, have greater fleas to go on,
While these again have greater still, and greater still, and so on.”

However the speed of adoption is controlled by the availability of old workloads on the new platforms. In fact sometimes specific workloads stay put – IBM’s z Systems (CMOS processors) still run the vast majority of the world’s transaction processing for instance. Another barrier is the high cost of becoming a server chip manufacturer – these products ship a few million each year rather than the hundreds of million in the phone and PC markets.

20 years ago analysts were weighing up the potential shift from Unix to x86 servers. Up until that point server manufacturers carried the expense of designing and often fabricating their own chip designs and building an operating system to go on top – something akin to all theatre companies being run by actor/managers perhaps. Mips Computer in the Unix market was probably the first independent server chip manufacturer before Intel and AMD started the ‘PC server’ market along with Microsoft, whose Windows operating systems became de rigour.

For many years it’s been common to refer to x86 servers as ‘industry standard’, because you can buy very similar servers from a number of different OEMs and because the competition between alternative manufacturers of the chips, disk drives, network cards, graphics cards, operating systems, etc. keeps prices down. In practice these systems have increasingly become defined as a combination of Intel Xeon and Microsoft Windows, both of which are proprietary products – even if the instructions running on the processors are in someway common. I have particularly negative views about SAP HANA systems, which harness a very narrow set of specific technologies, making all systems identical in all but brand… and also about ‘software defined’ approaches which only are often restricted to Xeon and Windows.

Today I’m very interested in any movement that can extend the concept of ‘industry standard’ server – why I’ve written a lot about the chances of ARM chips in the server market and Linux.

IBM clearly didn’t like a narrow definition either, which is why in offloading its System x range it’s also promoting its Power8 chips in a different way by designing a ‘little endian’* version to run Linux and launching the OpenPower Foundation – an international organisation of 120 different vendors, loosely designed on the concepts of ARM’s ecosystem. Yesterday IBM announced the current state of play, which includes 10 new announcements of specific systems, servers, motherboards and cards shown in the Table.

Table – OpenPower Foundation offerings – March 2015

In addition it reported on the creation by the Chinese government at the end of last year of the China POWER Technology Alliance (CPTA) – a public-private partnership with the aim of creating ‘the world’s top technology solutions that leverage the latest Big Data and cloud computing capabilities and apply these outcomes in bank, telecommunications, energy, transportation, internet and Smarter City technology initiatives in China.’

So what are the chances of OpenPower servers becoming a new ‘industry standard’? Possibly ‘more and sooner than ARM’ is the answer, especially as there is an easy migration path onto these chips from the x86 world. Nonetheless there are a number of challenges. For instance:

• Even if there are 1,200 software companies working to bring Linux applications across from x86 to OpenPower, the majority of servers run Windows; what a coup it would be to persuade Microsoft to play a two horse race!
• Without Windows OpenPower servers are less likely to find success with small and medium businesses, which will thereby restrict the number sold
• In a global market the Chinese government’s decision to back IBM over Intel for server chips is counter-intuitive – do we really want a new nationalism?
• It will be important to balance IBM’s own systems against those from the  OpenPower Foundation
• The group will have to demonstrate the comparative advantages of OpenPower over x86 in real world examples; up until now they have been largely theoretical

My definition of an ‘industry standard’ chip is one that is available from multiple manufacturers, can support multiple operating systems and has an installed base of over 1 million. OpenPower ticks the first 2 boxes, while the thrust of the Chinese government and suppliers suggests it could handle the last as long as the Foundation continues to expand. The success of the Foundation represents a very positive step in IBM’s transformation from the old to the new world. IBM is taking more risks and moving faster than its competitors and stands to gain most from the coming growth in enterprise IT spending.

Notes: *’endianism’ describes the byte order of how instructions travel through chips – x86 is ‘little’, while Power has always been ‘big’ until the latest version; making things the same now makes it easier to move applications from one to the other