Businesses are replacing departmental databases - known
as data marts - with enterprise data warehouses designed to pool
information across the company.
"Data marts are very expensive," said Gartner analyst Donald
Feinberg. "If you have six departments each with its own data mart,
you end up with six hardware systems and six database licences. You
also need people who can maintain each data mart."
Businesses often find they end up with the same information
replicated in each data mart, he said. They no longer have a single
master copy of the data and have to spend more on storage.
"In an enterprise data warehouse, data quality is higher," said
Feinberg. "A
data warehouse project does not have to take a 'big bang'
approach. You can start on a small project and design it around a
data warehouse."
Over time, as more projects are added, the data warehouse grows
into an enterprise data warehouse, holding all business data.
Auction website eBay runs one of the world's largest data
warehouses, gathering 40 Tbytes of data a day, and has 5,000
business users. Speaking at the Teradata European user conference
in Lisbon this week, Oliver Ratzesberger, eBay's senior director,
architecture and operations, said that like many companies, eBay
had too many data marts. "Data marts will kill your business
strategy," he said.
Two years ago, Ratzesberger began a project to encourage
business units to abandon their individual data marts and use
eBay's central
Teradata enterprise data warehouse.
"We made this service free so business users could no longer
argue they could do data analysis cheaper in a MySQL or
SQL Server database," he said. "We told the business: we can
provide you with analytics faster than if you bought a server from
Dell and implemented the data mart yourself."
Ratzesberger's strategy is to provide data warehouse software
within eBay as a central service, run by the IT department, that
allows business users to experiment. "We want the business to fail
as fast as possible, try out an idea in days, and if that doesn't
work, try a new idea and keep trying," he said.
Ebay allows business units to test theories by running data
analysis using their own data within Teradata, said Ratzesberger.
The idea is that the data warehouse service is not designed for a
production system.
"There are no SLAs and users have a six-month limit on the data
they upload into Teradata," he added.
If the prototype analysis within Teradata proves fruitful,
Ratzesberger said business users would be able to use the prototype
as a starting point to develop a production system to run
permanently in Teradata.
"Usually our developers would spend 90% of their time fixing
data quality issues rather than building the data warehouse. With
this new service, they benefit because the prototype identifies
data quality issues."
Such issues can be fixed before developers embark on building a
production system, he added. "This speeds up development of the
software."
Although the data warehouse installation is extremely large,
Ratzesberger's approach illustrates how IT directors can reduce the
proliferation of data marts in their businesses by giving users a
low-cost alternative - an enterprise data warehouse.