Don't listen to Snowden ... Intel: We've switched on CPU crypto for Hadoop
'Transparent encryption' for tables, columns among updates for analytics engine
Chip giant Intel is redoubling efforts to defend its valuable data center turf – by developing its own technologies for data management and analysis technologies, such as Hadoop.
As part of its mission to make sure that Xeon chips get top-billing when data center admins mull what to run a large Hadoop cluster on, Intel announced on Tuesday several new features and technologies added to the Intel Distribution for Apache Hadoop – its own, mostly open-source, variant of the software.
The new stuff includes the second release of the Intel Graph Builder for Apache Hadoop; version 3.0 of the Intel Distribution for Apache Hadoop; the Intel Analytics Toolkit for Apache Hadoop; and the Intel Expressway Tokenization Broker.
One of the main projects Intel has worked on for Hadoop include "Project Rhino", which provides a framework for using x86 AES processor instructions for hardware-accelerated encryption and decryption in Hadoop. This support was developed amid whistleblower Edward Snowden's allegations that the NSA has managed to compromise cryptographic capabilities of unnamed popular chipsets, leading to FreeBSD reducing its reliance on the x86 RdRand operation. OpenSSL is also urged to do similar.
For this Hadoop release, however, Intel has switched on "additional encryption capabilities in HBase," Ritu Kama, director of product management for Intel's Big Data group, told El Reg. The features enable "transparent encryption of HBase tables and column families and also extended for encryption at cell level in HBase."
This approach is twenty times faster than doing the same in software on an equivalent hardware stack, Kama said.
Other new features include the Intel Analytics Toolkit, which gives data wranglers access to a set of algorithms and machine-learning models.
"We are developing a set of building blocks or algorithms in the toolkit which customers can directly use to build an app whether recommendation, clustering," Kama says.
"You don't have to start from scratch every time. We will deliver a pipeline where customers will be directed to put data into an input directory. It could be any format - weblog files, structured or unstructured... and then we will help them build a pipeline where the data gets normalized into a format these algorithms pick up," she explained.
In the future, Intel "may also deliver a programming environment or IDE integration where developers can visually drag and drop things," she said.
Along with the toolkit, Intel is releasing a "Graph Builder" that will help admins slurp up Hadoop-stored data and build graphs – "retailers can create a graph based on information from their sales history data and their social media data to better understand the relationship between brand sentiment and purchasing habits of their customers," Intel explained in a canned release.
Intel is devoting so much effort to Hadoop because it feels the platform will become a prominent software system for the manipulation of data. And Intel wants to make sure its chips have an advantage against those fielded by rivals such as AMD. For this reason, much of the tech is open source apart from an overarching "Intel Hadoop Manager" layer.
"It is not really our intention to build a large set of exclusionary intellectual property blocks around this," explained Jason Fedder, general manager of channels, marketing, and business operations for Intel's Datacenter Software Division. "What we're trying to do is create an optimized set of building blocks to accelerate our core data center Xeon product lineup."
The analytics toolkit will be available in the first quarter of 2014, at which point Intel will release pricing. The Graph Builder toolkit will be available as an open-source download in January.
As for the Intel distribution (which includes the manager), it will cost between $1,500 and $3,300 per node, "depending on the number of nodes and support coverage option (24X7 or 9X5), an Intel spokesperson told us via email. ®