Covering Disruptive Technology Powering Business in The Digital Age

Written by: Martin Dale Bolima, Journalist, AOPG

In a virtual media briefing held this week, executives from NVIDIA, inventor of the Graphics Processing Unit (GPU), and Cloudera, a leader in enterprise cloud data, discussed fine details about their recent partnership. The two tech giants have collaborated to enhance data workflows in a time when data engineers are routinely bombarded with massive data sets and different business challenges. This collaboration will see NVIDIA’s pioneering RAPIDS Accelerator for Apache Spark 3.0 getting integrated into the Cloudera Data Platform (CDP) to ensure smart, fast and insightful analytics despite ever-increasing data volume and velocity.

Scott McClellan, Senior Director of NVIDIA’s Data Science Product Group, and Sushiil Thomas, Vice President of Machine Learning at Cloudera took point in the briefing. McClellan kicked things off by detailing NVIDIA’s investments in enterprise data centres and why it is collaborating with Cloudera in the first place.

“Cloudera has a longstanding track record of leadership and of bringing open-source technologies, especially technologies from the big data ecosystem to enterprise and making them consumable with their platforms and their excellent support of their products,” said McClellan in explaining NVIDIA’s vision in collaborating with Cloudera.

McClellan next highlighted how Artificial Intelligence (AI) and machine-learning (ML) have driven demand for accelerated data centre infrastructure and how this NVIDIA-Cloudera partnership can address the three most common data science problems that enterprises encounter:

• Growing, building, training and iterating models are very time-consuming.
• Large-Scale CPU infrastructure is extremely expensive.
• It can be incredibly difficult to perform comprehensive data processing operations.

In particular, this open-source GPU-boosted CDP helps organisations use AI and ML to accelerate data collection and processing and automate data analysis through AI and ML model training. These technologies make GPU-accelerated CDPs perform as much as 10 times faster than legacy and about 5 times faster than CPU-based servers.

Equally important, organisations looking to deploy this system need not worry about optimising their current data pipeline as Apache Spark integrates seamlessly with the CDP, which currently has over 2,000 customers, around 4,000 partners, approximately 400,000 data centre CPU servers and 5 exabytes of data.

“What a customer can do is add in a rack of GPU servers to their existing Cloudera cluster and we’ll take care of the rest,” explained Cloudera’s Thomas. “Any Spark sequel workloads and Spark data frame workloads that can now take advantage of GPUs can use them without any required application changes.”

The result, according to Thomas, is an optimised solution for modern enterprises that can improve model accuracy and run more iterations and ensure full-stack acceleration. It is also fully supported by both NVIDIA and Cloudera, which means “it can be used for production workload at massive scale.”

McClellan elaborated on this seamless integration in the Q&A, saying, “From the perspective of the end-user, there are no modifications. It’s a standard Spark application that’s exactly the same on CPU and GPU. We’ve worked a couple of years with upstream Apache Spark to ensure that Spark itself has GPU awareness scheduling and handles columnar versus row orientation of data and can transform back-and-forth.”

While customers do not need to modify their existing platform, McClellan noted that optimisation can nonetheless be performed and that doing so “opens up whole fields of possibility” in the future to solve business problems that may arise.

“Just by integrating the right bits into the CDP, whether you use Spark sequels or Spark data frames to access the data, with zero application changes, you see immediate performance gains,” added Thomas.

McClellan and Thomas also touched on the success story of the United States Internal Revenue Service (IRS) to further highlight the advantages of the NVIDIA-enhanced CDP. Thomas even showed a quote from Joe Ansaldi, Research Applied Analytics & Statistics Division Technical Branch Chief of the IRS, stating how the agency is already “seeing over 10x speed improvements at half the cost,” for their data engineering and data science workflows since deploying the NVIDIA-Cloudera integration in April.

The GPU-accelerated CDP, though, is set to be offered on the private cloud only. But Thomas assured the media that “public cloud deployment and support is coming very soon as well.”

(0)(0)