Here we're entering into the realm of incremental announcements of the kind we're seeing from lots of big data platform vendors. The theme is improved reliability, manageability, compliance control and so on, with upstart big data vendors more or less matching functionality already available on more mature platforms. Case in point, Cloudera announced Tuesday that it's introducing Cloudera Navigator, Cloudera Enterprise BDR and version 4.5 of Cloudera Manager software.
Cloudera Navigator is an all-new data-management tool for Hadoop that provides data access control, provisioning and auditing capabilities. This is essential stuff for security- and privacy-sensitive healthcare, financial services and government organizations, so it's more like checking an overdue requirement than bringing a breakthrough to Hadoop. The same goes for BDR, which offers enhanced backup and disaster recovery capabilities. Cloudera Manager 4.5 supports rolling updates, so you can now stage upgrades of Hadoop software without bringing down an entire cluster. The update also enhances monitoring capabilities.
[ Want more on Intel's big data news? Read Intel Unveils New Distribution For Apache Hadoop. ]
Again, this is all stuff that big organizations will demand from Hadoop as they move applications out of pilot and into production. Cloudera is executing on obvious requirements here rather than pioneering breakthrough capabilities.
Hortonworks and Microsoft deliver as promised.
Hortonworks and Microsoft announced in 2011 that they would co-develop a Hadoop Distribution capable of running on Windows. In October they released the beta version of that software. On Monday the partners formally introduced the Hortonworks Data Platform for Hadoop.
The key appeal for Microsoft shops, beyond running on Windows, is that the distribution is integrated with Microsoft System Center for administrative control and with Active Directory for access control and security. It also works Microsoft's virtualization platform, including Hyper-V and the System Center virtualization fabric, making it the first distribution to run Hadoop on virtualized infrastructure. Hortonworks would underscore that it's 100% open source, but there's not much more to say than "it's finally here."
MapR and Google rev their engines.
Performance-oriented Hadoop software supplier MapR and Google announced Tuesday that they have set a new world record for the MinuteSort benchmark, sorting 1.5 terabytes in 60 seconds using Google Compute Engine and the MapR Distribution of Apache Hadoop. The previous record, held by Yahoo with open source Apache software, was about 0.5 terabytes, and that was done with 3,452 physical servers, versus the 2,103 virtual/cloud servers employed by the Google Compute Engine.
Does that say more about Google's cloud service or MapR's software? "This shows off the ability to use Google Compute Engine to get very high-performance computing very quickly and in a very consistent manner, and you're seeing incredible IO performance," Google Developer Programs Engineer Marc Cohen told InformationWeek.
MapR's Jack Norris, VP of marketing, hastened to add that MapR's architecture -- which uses the Network File System in place of HDFS and implements other proprietary performance tweaks -- takes better advantage of Google Compute Engine power than would rival Hadoop architectures. For now, MapR has an exclusive partnership with Google. Unfortunately, the Google Compute Engine and the MapR Hadoop service running on that engine are still in limited preview release, and Cohen declined to comment on when it might become generally available.
Does a benchmark test have anything to do with real-world use of Hadoop? "A sort is a special case of a large MapReduce job, but any MapReduce work involves a lot of parallelism," said Cohen. "It requires that you bring up and coordinate thousands of machines, so we're proud of the fact that we can do it reliably and very fast."
So there you have it. Six announcements from Strata ranging from potentially huge -- if they live up to their billing -- to nice-to-have, expected and, in the last example, a gee-whiz lab stat. We're still at a stage in big data where everybody sees huge potential, but we always seem months, if not years, away from seeing real-world proof that the breakthroughs will be as tantalizing as the announcements suggest.