Ingres' VectorWise rises to answer Microsoft
Open-source stealth project slips out
Ingres is inviting data munchers to take a bite out of its planned software architecture targeting high-performance analytics.
The open-source database company has released alpha code for its VectorWise project, a high-performance storage engine computing-intensive data warehouses running on standard Intel boxes. A first beta of VectorWise is planned for the current quarter.
Ingres claimed VectorWise is between 10 and 15 times faster than rival computational, client products that run on a single, multi-core machine.
VectorWise achieves this by allowing standard SQL applications to access the compute power on multi-core chips while serving up huge blocks of data, Ingres said
The project is designed to bring data warehouse-levels of processing and performance to standard clients. It started after Microsoft purchased specialist DataAllegro, which had been working with Ingres. Features from DataAllergro are now due to appear in the Datacenter Edition of Microsoft's SQL Server 2008 R2, planned for release in May.
That product is being tuned to hardware running up to 256 logical processors and an unlimited number of virtual machines per license.
Ingres has worked on its project with VectorWise, which spun-out from mathematics and computer scientific research hothouse Centrum Wiskunde and Informatica (CWI) and whose brains worked on the MonetDB.
Bill Maimone, Ingres senior vice president of worldwide engineering, told The Reg VectorWise is for machines with hundreds of gigabytes or terabytes of data where the user wants to do complex queries and that involve lots of computation.
The system is designed to simplify the construction of parallel data warehouse systems using SQL, where you've got to design roll up objects or define objects in a cube, where you have to guess what queries will be made. In some cases, people are writing hand-coded programs against flat files.
Not only are such architectures complex to build, servers are bolted on to provide additional performance, further complicating the architecture itself and meaning additional expense and systems management overhead.
To speed the performance in the VectorWise project, Ingres has done a number of things to let the software work with changes in the underlying x86 hardware.
Instead of managing data buffers in random access memory, VectorWise uses the cache on the underlying chip, while all instructions are loaded from cache onto a multi-core chip. Maimone reckoned most business applications only use 10 per cent of the available capacity in the underlying multi-core chip for processing.
VectorWise also transfers a large amount of data to get around the fact that while processors are speeding up, discs can only go so fast: VectorWise will read up to 500kilobytes of data compared to the usual rate of between four and eight kilobytes. ®
Sponsored: Becoming a Pragmatic Security Leader