Netezza Needs IBM, But Why Does IBM Need Netezza?
September 27, 2010 Timothy Prickett Morgan
Sometimes it is easier to buy off a competitive threat and figure out how to make best use of it later than to worry about making complex justifications about an acquisition at the front end. That’s what IBM did when it shelled out $810 million to buy NUMA-based server maker Sequent Computer Systems in the summer of 1999, and that is what is going on now with Big Blue’s $1.7 billion acquisition of Netezza last week.
The Sequent deal was every bit as much about blunting a future attack of Windows-based NUMA servers that offered the kind of scalability that IBM was offering in its AS/400, RS/6000, and mainframe lines at the time as it was about IBM getting its hands on really good NUMA clustering technology. IBM was designing its own x86 and Power NUMA chipsets and bought Sequent to get its engineering brains and intellectual property, certainly not just to scale up its Windows machines to compete against Unisys, NEC, and Dell. Remember that the Sequent NUMA chips were the best of their kind, and they have been the backbone of the 570-class machines in the iSeries, pSeries, and Power Systems lines.
So what does IBM need with Netezza? It has its InfoSphere Warehouse, Cognos, and SPSS data warehousing and data analytics tools, and it has built clusters of Power, X64, and mainframe machines that bring plenty of data dicing and slicing performance to its customers in the Smart Analytic Systems. The company builds very fast processors and has flash-based storage in its servers, one of the key components of Oracle‘s Exadata online transaction processing and data warehousing appliances.
What Netezza has done is clever, and that is why its sales are projected to grow at 30 percent or so this year to $250 million and maybe keep on growing to $325 million next year. The company has taken a field programmable gate array (FPGA) co-processor and married it to a heavily customized version of the open source PostgreSQL relational database. The FPGA is the key, in that it prefilters data coming off storage subsystems and compresses it before that data is passed on to the database nodes (which are X64-based blades) that actually chew on the data. This is akin to what Oracle’s Exadata storage software is doing inside of its Exadata appliances.
Netezza’s first implementation of its appliances, which debuted in 2003, put the FPGAs inside of customized Power-based servers OEMed from someone other than IBM–probably Bull but maybe not. The follow-on TwinFin appliances as based on IBM’s BladeCenter HS22 two-socket Xeon blade servers, put eight FPGAs (one for each core in the X64 blade) on a companion blade that snaps onto it. Netezza is just getting ready to upgrade these appliances, and it stands to reason that it will put a two-socket blade out using Intel’s latest six-core Xeon 5600 blades and hence put a dozen FPGAs on the companion blade. That would yield a 50 percent boost in performance right there.
The company is also promising some major performance enhancements in its software stack with the next TwinFins, and has ramped up a partnership with Japanese server maker NEC that has NEC pairing the Netezza software and FPGAs with its own Express5800 server line and selling the combined product, called the InfoFrame DWH Appliance. If NEC was sniffing around to acquire Netezza, IBM had to react fast. And with Oracle taking a keen interest in hardware, Hewlett-Packard and Dell on the prowl for some data analytics action, IBM wan forced to move quickly and pony up what I think is an absurdly high premium for Netezza at $1.7 billion.
Netezza has well-regarded data warehousing technology and has recently been providing hooks to so-called “big data” products such as Hadoop. But what it doesn’t have is a big sales force or a huge customer base, which numbers 470 and 350 at the moment, respectively. Steve Baum, Netezza’s president and chief executive officer, said that the biggest problem the company had was putting feet on the street to chase the opportunities, and now, with IBM having some 6,000 consultants dedicated to data analytics, we’ll see how much wind can be swept across Netezza’s sails to get it ripping through the waters.
I think there is something else IBM wants from Netezza, and that is intellectual property that it can use to add accelerators to its System x, Power Systems, and System z servers and help significantly boost the performance of its DB2 database. Netezza could not do this by itself, of course, without IBM’s help, with the three different DB2s being closed source. If Oracle can close on that $1.5 billion pipeline for Exadata appliances that CEO Larry Ellison keeps bragging about when he talked to Wall Street and IT analysts, then IBM is going to need some performance advantages to compete better in data warehousing. And letting Netezza slip into Oracle’s hands, as IBM did with Sun Microsystems, could prove harmful to IBM’s plans and schemes.
While I think it would be interesting to see Netezza FPGA accelerating all kinds of database-driven workloads, I cannot say if the technology can be easily ported to the DB2 for i variant of IBM’s database, or if Big Blue would be inclined to do it even if it was technically possible. I certainly think anything that would improve the price/performance of the IBM i platform is a good thing to do. It is better to preserve the OS/400, i5/OS, and IBM i base than it is to let it slip to Windows/SQL Server (which is where most customers would go) or to alternatives like Oracle’s Solaris or Linux stacks.
I would not be surprised if someone doesn’t try to swoop in and outbid IBM for Netezza, and even if no one does, it will be interesting to see what IBM does with the technology.