As organizations continue to evaluate Hadoop for large scale data analysis, Hadoop software vendors are refining their products for enterprise use, addressing concerns around reliability and expanded use.
This week, Cloudera and MapR both announced that they are working on new features for their Hadoop distributions. And enterprise software vendors VMware and Splunk have announced products that should help the Apache Hadoop data processing platform work more easily with other IT systems.
The companies all announced their new work at the O'Reilly Strata Conference and Hadoop World 2012, going on this week in New York.
"We find that most of the Hadoop workloads are things you couldn't previously do, like combining disparate data types," said Kirk Dunn, Cloudera chief operating officer. As a result, Hadoop vendors are rushing to meet the new demands required by these new workloads.
Cloudera is working on a database engine, code-named Impala, that can query datasets stored on the HBase database through SQL (Structured Query Language). Until now, organizations tended to use Hive to execute SQL querying against HBase. This approach, however, can be slow, since Hive uses the Map Reduce framework, which requires the results of each query be written to disk. This process can be especially tedious when multiple sub-queries need to be made to form a single query.
The Impala database engine uses the Hive metadata directory, though it bypasses MapReduce, while still offering SQL as an interface, said Charles Zedlewski, Cloudera vice president of products. As a result, it runs queries much faster than Hive.
Eventually, Impala will be the basis of a Cloudera commercial offering, called the Cloudera Enterprise RTQ (Real-Time Query), though the company has not specified a release date. The company has released the Impala source code under an Apache Foundation license. A number of business intelligence software providers have already tested their own products against Impala, including Karmasphere, MicroStrategy, Pentaho, and Tableau.
MapR is adding new features to make its own HBase database distribution more reliable. The database can now be replicated and mirrored, so that if one copy goes down the system can switch to the backup copy.
While a generic version of HBase does offering mirroring capabilities, it relies on HDFS (Hadoop File System), which is a "write-once" file system, said Jack Norris, vice president of marketing for MapR Technologies. As a result, it can take up to 30 minutes to switch to a back-up copy of HBase. MapR uses its own file system, which has been extended to handle tables as well.
"The files and tables are side-by-side in the volumes and directories. HBase reads directly from those tables, so now you have instant recovery. You can read directly from the snapshots," Norris said.
The database comes with a number of other new features as well. The new database does not compact, or compress data, which should allow the database to perform more consistently. The company claims that inserts and updates occur much more quickly. The database now supports in-memory columns. The row and cell sizes have been increased to accommodate larger objects as well, up to 1GB in size. And in this version of HBase, users can create more than a trillion tables.
MapR's M7 version of HBase is fully binary compatible with Apache HBase, and can run Apache HBase within the cluster with M7's own version of the data store software.
The beta version of M7 is now available to select users. It will eventually replace the company's M5 distribution.
A number of other companies have also released new products in conjunction with the Hadoop conference. Machine search software vendor Splunk has released Splunk Hadoop Connect, which facilitates the passing of data between Splunk and Hadoop. The company also released a Splunk monitoring module for Hadoop, called The Splunk App for HadoopOps. VMware announced an updated version of its Project Serengeti, which is software for running Hadoop in virtualized environments.