Generally speaking, server makers of all stripes and persuasions are a pretty secretive bunch. They do not provide a lot of internal statistics that in any meaningful way describe their businesses because they think that such information can only be used against them by their competitors. But, every now and then, a vendor gets to bragging, or in other cases feels the need to show that its particular server line is strong. And then out comes the data.
Such was the case at the recent System z Summit, which IBM held at the St Regis Hotel in New York for a gaggle of industry analysts who may not be as familiar with mainframes and how they are used by corporations around the world as analysts were in decades gone by, when the mainframe was the unquestionably dominant machine in use for data processing among the world's largest corporations.
Because IBM wants to get the word out about the strength of the mainframe market, a few puzzled journalists were also invited to the System z Summit and were given some hard data that would normally have never seen the light of day outside IBM's walls.
The data that IBM's top mainframe brass shared in their presentations and in private interviews sheds some light on the mainframe market and why IBM remains staunchly committed to this venerable platform.
Bob Hoey, vice president of worldwide sales for the System z line at IBM, spends a lot of time on the road visiting mainframe shops.
And while the mainframe base is not small enough to know every account personally, there are by no means as many mainframe shops as there are companies using Windows and Linux on X86 and X64 servers or those using RISC/Unix machines.
In fact, Hoey levelled with me - as few IBMers have done in the past - and said flat out that there were about 10 000 mainframe footprints in the world. "I know where the machines are by serial number and zip code," Hoey joked. About a third of these machines are in the financial services sector, and when you add in insurance and related businesses, these businesses account for about half of the footprints.
These 10 000 footprints drive many billions of dollars in mainframe hardware sales, and then an incredible amount of storage and other peripherals as well as very profitable software sales. (Exactly how much, IBM will not say.) But what Hoey did say is that IBM spends $1,2bn a year investing in the System z stack - including hardware, software, and services.
About 65% of the revenue that comes from the mainframe base comes from software, according to Hoey, and not surprisingly, about 65% of that $1,2bn in annual investment in research and development for the mainframe line goes into software. (This investment includes, among many other things, the five-year, $100m project IBM started last year to create wizards to make managing a mainframe not only easier, but more like Windows, Linux, and Unix servers so it is easier for mainframe shops to train people to manage their boxes.)
Even though mainframe sales are only projected to grow at around 5% annually, IBM feels it is getting an excellent return on its mainframe investments. The reason? Gross profits on the mainframe stack are very high - probably on the order of 85% to 90%, but again, IBM will never say.
Jim Stallings, general manager of the System z line, took the helm of the mainframe in January 2006, just as the mainframe was hitting what looks now like a new stride. He is obviously quite pleased with the way the mainframe keeps confounding its critics and surviving as it adapts to new computing paradigms while supporting legacy applications that are older than many of the people administering the machines today.
In his presentation at the System z Summit, Stallings said that mainframe hardware sales in the fourth quarter of 2006 were the largest that IBM has seen since the fourth quarter of 1998--which is when the Y2K problem was in full swing and lots of mainframe shops were upgrading while others were defecting to other platforms.
(A decade ago, there were an estimated 20 000 mainframe footprints, according to some estimates.) Last year was a record year for the specialty mainframe engines, and IBM broke through 11 million installed MIPS of aggregate processing power.
The growth in the installed base of MIPS capacity is probably something of a surprise to IBM's competitors in the high-end server market. But the growth in capacity is a testament not only to the platform, which has obvious benefits in terms of throughput, robustness, and familiarity, but also to the pricing elasticity for mainframe capacity even when there are much cheaper alternatives - cheaper meaning the price for raw hardware and software unadjusted for utilisation rates. (When you adjust for utilisation rates, the economics can change dramatically.)
Back in the first quarter of 2000, IBM had a little under 3,5 million MIPS of mainframe capacity installed. As IBM moved through successive generations of CMOS processors, it was able to grow that base more or less linearly through the fourth quarter of 2001, when growth slowed. Still, by the end of 2002, the mainframe base had almost doubled to 6 million aggregate MIPS.
When the global economy recovered in 2003 in the wake of the dot-com bubble bursting and the 9/11 terrorist attacks, IBM was ready with a revamped zSeries 990 server and followed in quick succession in the spring of 2004 with the midrange z890 servers.
By the fall of 2005, IBM had boosted the base through 9 million aggregate MIPS, and while the growth has not been linear every quarter, the base stood at 11,1 million aggregate MIPS in the first quarter of this year. That is nearly a factor of four growth in the installed base in seven years - which is a far cry better than the massive contraction in mainframe footprints and capacity that IBM suffered through in the 1990s.
One of the key drivers of the mainframe base has been the advent of the so-called specialty mainframe engines. These are the Integrated Facility for Linux (IFL) engines for running mainframe-native Linux, the System z Application Assist Processor (zAAP) for running Java, and the System z Integrated Information Processor (zIIP) for accelerating DB2 database functions.
These specialty engines cost as little as a quarter of a regular mainframe engine, and they have made the mainframe an attractive option for customers who like centralised computing to both keep workloads on the mainframe and to bring new workloads onto the box.
According to the data that Stallings provided, the installed base of specialty mainframe engines is growing faster than the base as a whole. In the first quarter of 2007, IBM had around 1,7 million MIPS of mainframe processing capacity from all types of specialty engines.
About 1,2 million MIPS was for IFLs running Linux, with a little more than 100 000 MIPS coming from zAAPs and just under 100 000 MIPS coming from zIIPs. Stallings said that about 25% of MIPS sold today are for Linux, and that about 60% of the MIPS IBM peddles now are for specialty engines. Imagine if IBM had not created these specialty engines a few years ago. The business would be sinking, not growing.
But, as it turns out, in the market for servers that cost more than $250 000, the System z has been outgrowing the competition since the end of 2000 and is now above the market share in this sector of the server business that IBM held with mainframes back in 1996 - in excess of 35% revenue share for big iron.
The IBM mainframe has done better than the high-end server lines from Hewlett-Packard, which have under 20% revenue share in aggregate in this server price band, and big boxes from Sun Microsystems and Fujitsu, which are fighting for a 10% share of the revenue pie.
It is interesting to note that Stallings did not put IBM's own System p Unix line in this comparison using IDC's quarterly sales data. That is because IBM's high-end Unix server business has gone from third place to a dead heat for first place in the same time.
IBM did not want to talk about System p at the System z Summit.
They can host their own event.