Overview
Your organization has spent millions of dollars implementing a data warehouse or
data mart using the corporate standard database, such as Oracle or DB2. After
all, in-house expertise is available for these databases and volume purchase
discounts and site license agreements with the vendors in place. But despite the
considerable expertise, service breaches are occurring frequently, users are
dissatisfied with the response they are getting, a backlog of enhancements is
mounting and any big ideas your organization has about extending Business
Intelligence (BI) has dissipated. Faced with these problems, you have employed
large teams of database administrators to tune the database to make it respond
better to complex queries and you have upgraded your system as far as your
budget will allow. But all of this is in vain: every step you take just seems
to take you right back to the beginning.
Like most first time visitors to www.netezza.com, it is exactly this frustration
that has led you to where you are today.
At Netezza, we believe that data warehousing shouldn’t be this complicated.
Instead of spending time making data warehouses run efficiently, Netezza users
deploy purpose-built data warehouse appliances that solve
business problems. Netezza data warehouse appliances are installed quickly and
easily, integrate with your preferred ETL and BI tools and can be largely left
alone to get on with the job.
Netezza appliances provide a database-server-storage configuration in a
purpose-built system designed to perform complex queries against large volumes
of stored data. Netezza data warehouse appliances are designed for blisteringly
fast analysis against terabytes of data 10-100 times faster than traditional
solutions, with a lower TCO and greater ease of use. Netezza uses massively
parallel processing and an architecture that puts processing right inside
storage to provide a brute force solution that can deal with complex analytics
against large data volumes.
But what’s wrong with conventional databases?
Because they were designed for a completely different application profile, OLTP
databases provide a poor platform for data warehousing. They are designed to
read small indexed records from disk, moving them to memory where they are
updated and written back to disk as the persistent record. Analytical systems do
the reverse: they trawl through vast amounts of data looking for exceptions. At
a level of one terabyte and above, databases designed for transaction processing
struggle to deliver acceptable response times and their owners begin to spend
inordinate amounts of time and money on performance tuning. However, the root
problem is not in the database software per se. It is the latency in moving vast
data volumes from disk across a network and into the memory of the computer so
the database can then start to do its job. The problem is outside the control of
the database. In analytical systems, database tuning is a self-defeating task:
to tune, the administrator must be able to predict which queries will access
what data. Analytical processing has at its core the "problem→question→better
question cycle"; the analyst, let alone the database administrator, cannot
predict which complex query will be used next.
Traditional database architecture is unable to process data until it has been
transferred from disk across a network and into memory of the CPU(s) running the
DBMS. This movement of data represents a technology bottleneck caused by several
factors:
- Disk transfer I/O rates cannot read terabytes of data quickly enough
- Network transfer rates cannot move terabytes of data quickly enough from disk to memory
- Memory density growth cannot keep up with data growth making traditional caching less effective over time
And increasing CPU performance becomes irrelevant since the gating factor in
traditional implementations is moving the data off disk to the CPU.
Next >