Netezza to bake analytics into appliances
Big math chews big data
Data warehousing appliance–maker Netezza wants to set its TwinFins loose on data analytics, thereby doubling the usefulness of the boxes and positioning them to better compete with alternatives from IBM, Oracle, and Teradata.
The TwinFin appliances marry X64-based blade servers and home-grown accelerator chips to a heavily customized PostgreSQL database. Like other data warehouses, the Netezza appliances were designed to run ad hoc SQL queries against giant data sets. The special sauce in the Netezza gear is a field programmable gate array (FPGA) that handles data compression and decompression as it goes on and off the disk, thereby speeding up throughput, as well as providing sophisticated data filtering so that only the relevant portions of a giant data set are passed on up to the X64 engines to run the SQL query.
Netezza launched the TwinFin appliances last fall, using IBM's BladeCenter chassis and two-socket HS22 Xeon blade servers and plunking the FPGAs (one per Xeon core) onto a companion blade; the pair together are known as a Snippet or S-Blade in the Netezza lingo. The TwinFin appliances launched last year had from 3 to 12 S-Blades in BladeCenter-H data center chassis and spanned from 8TB to 320TB of data warehouse capacity.
In January, Netezza shot low for smaller customers looking for data warehousing appliances with its Skimmer appliance, which puts an S-Blade into the office-style BladeCenter-S chassis and gives the user 10TB of disk space to play in.
A few weeks ago, Netezza announced it was hooking up with NEC to build appliances based on NEC server iron that would be sold by NEC as well as Netezza. (IBM doesn't resell the Netezza appliances based on its BladeCenter hardware.)
According to Phil Francisco, vice president of product management and marketing at Netezza, in the second quarter the company will open up the Netezza appliance so it can do data analytics, not just run the PostgreSQL database and process SQL transactions against the data warehousing.
"Now you don't have to do large data extracts to run analytics - you do it right in the data warehouse," says Francisco.
This is, of course, exactly what IBM is pitching with its Power-AIX Smart Analytics Systems, which debuted last July running a mix of Cognos and now SPSS analytics alongside Big Blue's InfoSphere Warehouse add-on for its DB2 database. Oracle is pitching its Exadata V2 machines as clusters that are capable of running data warehousing or online transaction processing workloads equally well, and it won't be long before analytics are tossed into the mix. Teradata has a mix of appliances for supporting data warehousing and analytics, too.
Netezza is not providing all the feeds and speeds of its analytics offering, which will be launched in the second quarter and which will be called TwinFin(i). But Francisco says that the Linux operating system and PostgreSQL database engine at the heart of the appliance is being opened up with a slew of APIs that will allow new kinds of programs to execute on the X64 portions of the appliance, and process more than SQL transactions against data stored and managed on the TwinFins by the FPGAs.
Specifically, a set of APIs will be opened up in the Netezza stack that will allow Java, C++, Fortran, or Python applications to be created that can use the FPGAs to extract data from the warehouse and do complex analytics on it. If you like using SAS tools or the open source R statistical programming language, the Netezza stack will have APIs that let these hook in, too.
And if you want to go crazy and use MapReduce or Hadoop to chew on large data sets like Google, Yahoo, and other Web 2.0 companies do, there's a set of APIs that let these tools hook into the PostgreSQL data warehouse and chew on that data as well. There's another set of math libraries that will allow it to do matrix math on your data, if that is your thing. The TwinFin(i) will come with a development environment that can snap into Eclipse-based IDEs.
This is not the first time that Netezza has allowed languages other than SQL to hit its data warehouse. In 2007, with a feature called onstream processing, the Netezza appliance was able to allow C and C++ applications to extract data from the warehouse, just like they might any file system or database, and make full use of the FPGA's data compression, tabling joining, and filtering capabilities.
The TwinFin(i) will be sold as a standalone appliance, but customers using the existing TwinFin appliances will be able to do a software upgrade to get the same functionality. Netezza did not divulge what the incremental cost would be to add the analytics ports to the system, but presumably this is something Netezza will try to charge for.
At the moment, Netezza has lined up a number of partners who want their analytics software to run on the appliances, including SAS, Tibco, MicroStrategy, Pursway, DemandTec, and QuantiSense. ®