Evaluating columnar databases
The market for columnar databases was once a narrow niche that played a distant second to traditional row-based relational databases, as espoused by Oracle, IBM, Sybase, Microsoft, and others. However, organizations seeking to speed up data query and access are now looking more closely at this technology to meet the increased pressures put on business analytics – namely, performance and speed to insight. The exponential growth of data warehouses – from gigabytes to terabytes and even petabytes – is starting to take its toll on query performance. That is hindering companies from deeper analysis of their business data for fear of “runaway queries from hell” that take all night to process. To address this problem, columnar databases optimized for data analytics are now in development, and organizations are starting to implement them as an alternative or adjunct to data warehouses traditionally built around relational data structure constraints.
This report aims to introduce the basic concepts and benefits of columnar databases in a straightforward manner. Its value proposition should be balanced against limitations of the technology, and organizations should guard against swallowing vendor rhetoric. Hence, this report also presents a multidimensional evaluation model that highlights the key functional considerations that organizations should look out for when choosing a columnar database for analytics. It is important to remember that while columnar databases are applicable to several data problems, there are situations to which they are not suited. The report ends by providing practical advice for both IT end-user organizations thinking of buying and implementing a columnar database, and IT vendors developing and selling these systems.
This report aims to introduce the basic concepts and benefits of columnar databases in a straightforward manner. Its value proposition should be balanced against limitations of the technology, and organizations should guard against swallowing vendor rhetoric. Hence, this report also presents a multidimensional evaluation model that highlights the key functional considerations that organizations should look out for when choosing a columnar database for analytics. It is important to remember that while columnar databases are applicable to several data problems, there are situations to which they are not suited. The report ends by providing practical advice for both IT end-user organizations thinking of buying and implementing a columnar database, and IT vendors developing and selling these systems.
SUMMARY
Impact
Ovum view
Key messages
THE ANALYTIC CASE FOR COLUMNAR DATABASES
Optimized for business questions
The business benefits
Faster performance
Storage efficiency
Lowered DBA costs
Increased analytic productivity
CHOOSING THE RIGHT COLUMNAR DATABASE
A once-sleepy market is getting noisy
A practical evaluation model
Analytic performance
Data and end-user scalability
Data load times
Range of analytic functions
Data schema flexibility
Data access techniques
Data type support
Database administration overhead
System availability
ALTERNATIVE VIEWS
The counter-argument
Other analytic options to consider
In-memory databases
Vector databases
Database appliances
OLAP accelerators
Database clustering
Data-matching systems
Adaptive indexing
Associative databases
Others
RECOMMENDATIONS
Recommendations for enterprises
Consider performance trade-offs between queries and data loads
Target query-intensive applications
Be realistic about size
Rows are here to stay
Make the columnar system a good BI citizen
Expect some resistance from IT
APPENDIX
Further reading
Methodology
Impact
Ovum view
Key messages
THE ANALYTIC CASE FOR COLUMNAR DATABASES
Optimized for business questions
The business benefits
Faster performance
Storage efficiency
Lowered DBA costs
Increased analytic productivity
CHOOSING THE RIGHT COLUMNAR DATABASE
A once-sleepy market is getting noisy
A practical evaluation model
Analytic performance
Data and end-user scalability
Data load times
Range of analytic functions
Data schema flexibility
Data access techniques
Data type support
Database administration overhead
System availability
ALTERNATIVE VIEWS
The counter-argument
Other analytic options to consider
In-memory databases
Vector databases
Database appliances
OLAP accelerators
Database clustering
Data-matching systems
Adaptive indexing
Associative databases
Others
RECOMMENDATIONS
Recommendations for enterprises
Consider performance trade-offs between queries and data loads
Target query-intensive applications
Be realistic about size
Rows are here to stay
Make the columnar system a good BI citizen
Expect some resistance from IT
APPENDIX
Further reading
Methodology
LIST OF TABLES
Table 1: Ovum taxonomy of columnar database providers
Table 2: Benefits of columnar databases
Table 1: Ovum taxonomy of columnar database providers
Table 2: Benefits of columnar databases
LIST OF FIGURES
Figure 1: Structural differences between row- and column-based data
Figure 1: Structural differences between row- and column-based data
