Comment It would be hard to find two technologies that would seem to be more diametrically opposed in the data center than the IBM mainframe and the open source Linux operating system. But the combination of the two, which then-IBM president and now IBM chairman and CEO Sam Palmisano championed (or more precisely, gets credit for championing within Big Blue regardless of whose idea it may have been), saved the System z mainframe business.
Saved it from what, you ask? Well, not from immediate economic oblivion. Given the entrenched nature of COBOL and CICS transaction processing applications, batch workloads, and DB2 and IMS databases underpinning heaven only knows what at some 6,000 to 7,000 mainframe shops worldwide, it would be very hard to replace mainframes. As the proprietary and Unix system vendors who have been trying for decades have learned. But they have crunched the mainframe hardware business, which accounted for $13.1bn in revenues in 1989, before the bottom dropped out thanks to a recession that in turn caused a mainframe consolidation and Unix frenzy.
Just to show you how different IBM was back then, the company had $41.6bn in hardware sales that year, and pushed PCs, printers, disks, minicomputers, telecom and networking gear, retail systems - you name it. And those mainframes had wickedly high profit margins and drove mainframe software sales where nearly all the money dropped to the bottom line - and helped pay the bill for all of those unprofitable hardware businesses, by the way.
By 1993, mainframe hardware revenues dropped to $6.5bn, and slid down to a low of $2.9bn in 2000, the year that Linux debuted on the big iron and breathed a little life on it. The combination of low-cost mainframe engines - which IBM called the Integrated Facility for Linux but which are really just plain-old mainframe engines with the z/VM hypervisor carving up virtual server slices and only sold to support Linux, not z/OS or other operating systems - and low-cost Linux operating systems was a hit with at least some mainframe shops, who were once again looking to cut costs in yet another recession. (The Linux engines cost about a quarter the price of engines intended to run z/OS, IBM's flagship operating system for its big iron; and even at $18,000 to $25,000 per core, Linux was a bargain compared to z/OS.) For those CIOs who build and maintain System z empires for a living, Linux on the box was a boon.
And perhaps equally importantly, the Linux/IFL combo fostered the delivery other low-priced "specialty engines" with lower prices and relegated to other workloads that might otherwise be shifted off the mainframe. First, in 2004, came the System z Application Assist Processor (zAAP) for offloading Java and XML workloads from the central processors (CPs) running z/OS and IBM's WebSphere middleware. Then, in 2006, came the System z Integrated Information Processor (zIIP), which accelerates DB2 databases by offloading certain functions from the CPs running z/OS and DB2 to these zAAPs.
Not all customers adopted IFLs, zIIPs, and zAAPs, but enough did that when the economy started to recover in the mid-2000s, the mainframe hardware business was back on its feet. If you average hardware sales between 2006 and 2008 (when IBM was doing the System z10 transition), IBM brought in around $3.5bn in mainframe processor sales - a little bit less than the $4bn in averaged for x64 iron and considerably less than the $5bn it has averaged for years with its combined RS/6000-AS/400 Power server lineup, now converged to a single Power Systems line.
In 2009, the economy went into the tank again just as IBM was readying its System z11 mainframes for market in 2010, and the combination of the two hammered the mainframe hardware biz down to the lowest levels it has ever seen, about $2.6bn.
Big Blue doesn't like to talk numbers, but Karl Freund, vice president of System z strategy and marketing at IBM, said that in 2009, which was an awful year for servers of all types and stripes, processing capacity on mainframes (as measured in Millions of Instructions Per Second, or MIPS) dedicated to Linux workloads grew in excess of 10 per cent. And the installed base of Linux-related MIPS is now in excess of 2.5 million MIPS, more than twice the base in 2007. (It is hard to say what percentage of overall mainframe processing capacity this is, since IBM did not provide the MIPS number for the whole base.)
Freund says that the average mainframe IBM sells these days has two or three IFLs, which are used for development, test, and production workloads. The number of IFLs is rising despite the poor economy - or maybe because of it. Linux accounts for well over one-quarter of mainframe MIPS sold, which is up from around a tenth of capacity sold a few years back, and zIIPs and zAAPs are growing rapidly, too. Which would seem to indicate that IBM is charging too much for generic mainframe capacity.
Here's a stunner: about half of new mainframe shops, says Freund, are Linux-only shops. No z/OS. No z/VSE. And only z/VM to run IFLs and Linux images. (That may say as much about how few new mainframe customers there are in the world as it does about the popularity of Linux on mainframes.)
The success of Linux on the mainframe has compelled software vendors to port their Linux apps to the System z, and after a decade of cajoling and encouraging, there are now 3,150 Linux applications certified on the System z platform - a little less than the application count on the flagship z/OS operating system, but pulling towards even. Of course, these numbers are for third party applications, databases, middleware, and other system programs, and ignore the vast installed base of homegrown z/OS code running on mainframes.
Whatever IBM has planned for the impending System z11 mainframes, if they are indeed called that, you can bet that Linux will be a big part of the sales push. All Freund would say is that there will be Linux-only bundles of hardware and software like the Linux Solution Edition and Enterprise Linux Server configurations that were created for the System z10 mainframes. ®