MapR boosts HBase application performance
MapR has released a new edition of its Hadoop distribution aimed explicitly at enterprises using NoSQL databases. M7 offers significant new features over and above those included in its existing M3 and M5 distributions. M3 is a free community version, while the paid-for M5 has support options and includes additional features such as mirroring and snapshot support.
The new M7 distribution goes a step further in that it gives a significant speed boost to the Hadoop-based key-value store Apache HBase. HBase combines storage and real-time analysis with Hadoop's MapReduce techniques. According to MapR, however, HBase has so far failed to achieve its full potential due to repeated operational problems, such as uneven performance, data loss, and complex administrative processes, partly caused by the restrictions imposed by Hadoop's write-once filesystem. Consequently, MapR has reworked the architecture for M7.
M7 does away with HBase's layered architecture. HBase applications now access data directly with just a single network hop; this eliminates delays caused by additional communication layers. The M7 architecture integrates file and table data storage, which should simplify administration and development, improve reliability, and allow improved performance and scalability of HBase applications.
Integration with commercial search platform LucidWorks, which is based on Apache technologies Lucene and Solr, is also available as a beta. This enables users to hunt through data stored in the Hadoop File System or on other filesystems. LucidWorks Search also simplifies installation of Lucene or Solr. The technology also has native support for multiple data sources and includes both a graphical user interface and a security framework. MapR is planning to incorporate the search options into all three of its distributions.