Home  /   News  /   Latest News
Latest News

Data Analytics Accelerator (DAX) technology – Faster with SPARC

Oracle Provides Details on Software in Silicon for Faster Data Analytics

M7

It’s hardly a secret that Oracle’s SPARC processor technology is built for speed. “It is the fastest commercial processor on the planet—nothing comes close,” said Rick Hetherington, Oracle vice president of hardware development, during an Oracle OpenWorld 2016 session.

But that’s just the start. By adding Software in Silicon to Oracle’s latest SPARC processors, the M7 and S7, the company has created technology custom-tailored for today’s data-fueled economy.

“Our focus is on developing technology that supports data acceleration,” said Hetherington.

Hetherington noted several factors driving the need for Software in Silicon, which builds software functions directly into the processor. On the hardware side, trends such as plateauing thread speed and the near-peaking of total throughput will impose a limit to the power a system can deliver and dissipate, giving specialized, power-efficient hardware an advantage over general-purpose cores. On the industry side, the need for faster analytics has shifted the focus from general transactional processing to managing, mining, and securing data.

“A hardware-software codesign offers unique opportunities to accelerate performance, given the limitations on the chip-trend side,” Hetherington said, before diving into some technical details, including the following:

  • Data Analytics Accelerator (DAX) technology. Oracle’s DAX technology boosts performance on a variety of analytic workloads by offloading tasks from the core thread to DAX. “Offloading tasks to DAX helps reduce the time it takes for the thread, upon completion, to read results and continue on with the query,” Hetherington said. The result is higher query performance along with as much as an eightfold reduction in core utilization.
  • Analytics and machine learning on DAX. Although DAX is in its early stages, Hetherington cited some use cases and performance gains. For example, in machine learning, using Java Stream JDAX speeds a process called allMatch by as much as 21 times, and increases the speed of data Filtering much as 10 times. Streaming data analytics such as Top-N perform as much as 4 times faster, while in-memory analytical queries perform as much as 8 times faster.
  • DAX placement on the chip. Located in the high-bandwidth memory path of the chip, DAX’s 32 in-silicon accelerator engines can scan as many as 170 billion rows per second. Their placement prevents the engines from fetching an enormous amount of unwanted data by avoiding lower-level memory caches, according to Hetherington. “We can save on power and maximize memory bandwidth by putting the engines where they are,” he said.
  • DAX programming model. New APIs for SPARC M7/T7/S7 features are now available on the Software in Silicon developer program portal to help developers test-drive Silicon in Software technology. “The Libdax API hides communications overhead in a simple function call that is easily called in Java, Scale, Python, C, and so on,” Hetherington said. “It provides building blocks for analytics acceleration.”

Hetherington said there’s more to come with DAX. “We were conservative with the initial DAX 1.0 with what we can address,” he said. “Going forward, software overhead is one thing that needs to be trimmed back, so that’s a candidate for future improvement.”

Comments are closed.

ACARDIA LIMITED 12th Floor, Ocean House, The Ring, Bracknell, Berkshire RG12 1AX. United Kingdom.