Ingres challenges Microsoft's DataAllegro warehouse steal
Stealth project tethered down
Memo to Microsoft: A year after you blocked its foray into fast and affordable data warehousing on mass-market Intel hardware, Ingres is back.
Today, Ingres announced the Ingres VectorWise Project to put a high-performance storage engine into its open-source database for computing-intensive data warehouses running on Intel iron.
The goal is to deliver the kind of rapid read and access query results that have been possible on data warehouses built using expensive proprietary hardware and customized software.
It's Ingres' second crack at data warehousing for the masses, coming a year after Microsoft bought  DataAllegro, who'd been working with Ingres. DataAllegro had devised a test version of a storage engine at the time Microsoft snatched away DataAllegro last July.
DataAllegro has now been put to work in Microsoft's own planned data-warehouse-on-Intel project, called Madison. Due next year, Madison uses Microsoft's proprietary SQL Server 2008 R2, and it's being built in tandem with Bull, Dell, EMC, Hewlett-Packard, and Unisys.
To realize it's own dream this time, Ingres has tapped VectorWise , at database spin-out from mathematics and computer scientific research hothouse Centrum Wiskunde and Informatica (CWI ) currently in stealth mode. Some of that team worked on column-oriented MonetDB  database during the early 1990s.
And, it seems, Ingres has taken steps to avoid a repeat of the Microsoft incident. Ingres has signed an exclusive option to buy VectorWise once integration with its database has been successfully completed. Chief financial officer Tom Berquist told The Reg that Ingres hopes to buy VectorWise, which only spun out from CWI in summer 2008.
Berquist said he expected the new Ingress database would come into competition with Microsoft, in addition to Oracle and potentially IBM and Teradata, in the new area. Ingres already competes against SQL Server in the regular database market.
The VectorWise Project is designed to take Ingres into large reservation, bank, and freight management systems and into business intelligence systems that are built on hundreds of billions of data fields and need to be read in real time.
"We are hoping to be an engine inside for those kinds of applications as opposed to have them to write custom code - convince them to use off-the-shelf hardware and software," Berquist said.
The challenge for applications like Ingres and emerging multi-core and mult-threaded chips from companies like Intel trying to serve this market is how to apply techniques and ideas used in the proprietary and customized worlds.
Systems such as those running Intel processors tend to pass off queries to a server disk. This is relatively slow and will suck in more servers as an application and its number of users grow, pushing up costs and complexity.
But increasingly, multi-core processors like Intel's Nehalem Xeon combine both the speed and threads needed to process massively parallel queries. The trick is how to make the software and the processor work together to load and run complex requests on the processor.
The planned VectorWise storage-engine architecture will process jobs on the CPU with the main memory only used to handle buffer for I/O and large intermediate data structures.
A series of compression schemes, meanwhile, have been devised that the companies suggested can compress single or multiple gigabytes per second, compared to compression the GPL-licensed LZOP that can only decompress a few hundred megabytes per second on a 3GHz processor. The VectorWise architecture further boosts speed by parsing entire vectors instead of single tuples. VectorWise would not reveal more about its architecture at this stage.
The companies said the architecture would help extract the maximum performance from "modern" hardware such as Nehalem. Intel's chip is available with two cores, but will go up to eight and 16 later this year. Intel is targeting the volume market with Nehalem.
The duo are not ready to presented any public benchmark but claimed a 10 fold performance improvement compared to existing systems running the engine on a server with two Xeon X5560 processors at 2.80GHz with Fedora Core 10.2.6.2.
Berquist noted, too, that Ingres has yet to decided whether the storage engine will be in the company's flagship database or a brand new, separate SKU that specifically targets high-performance read-only deployments. He noted a separate SKU could be a possibility.
"My guess is our existing customers will remain on the Ingress database codeline and if they want to get into query intensive applications or business intelligence that's where they would purchase the VectorWise SKU of the database," Berquist said. ®