Katana Graph optimizes analytics engine on 3rd gen Intel Xeon - Embedded.com

Katana Graph optimizes analytics engine on 3rd gen Intel Xeon

Katana Graph, a high-speed graph analytics startup focused on processing large unstructured data sets, announced it has optimized its graph engine for the new 3rd generation Intel Xeon scalable processor and memory systems.

The Katana Graph engine (KGE) is a platform for high-speed graph analytics, pattern mining and querying on heterogeneous clusters of CPUs and GPUs, providing what the company said is unmatched compute capability for processing even the largest graphs such as web-crawl graphs with billions of vertices and trillions of edges. It extracts actionable insights from massive unstructured data sets, using high-performance graph algorithms. Scale-out provides more aggregate computational power and DRAM for analyzing properties of larger graphs than is possible on a single machine.

The KGE with accompanying partitioner, communication, virtualization and storage technology modules are the culmination of more than a decade of advanced research in graph technology and high-performance computing by co-founders Chris Rossbach and Keshav Pingali. The KGE is based on research performed in Pingali’s intelligent software systems group at the University of Texas in Austin, which works on parallel programming models, compilers and runtime systems for applications with large-scale, unstructured data.

On CPUs, the Katana runtime system optimizes program execution to exploit NUMA (non-uniform memory access) locality; for example, it performs NUMA-aware dynamic load balancing to ensure that computational load is spread out evenly between the cores of the CPU. On GPUs, the Katana graph engine incorporates performance optimizations for reducing the overhead of kernel launches and atomic operations.

For execution on clusters, the KGE partitions graphs between the machines in the cluster using a rich variety of graph partitioning policies including edge-cuts and vertex-cuts. Application-specific graph partitioning policies can be implemented in the Katana graph partitioner.

Communication is often the performance bottleneck in the distributed-memory execution of graph analytics programs. To avoid this, the Katana graph engine has a communication runtime that has been optimized for graph computing.

The Katana core graph libraries provide highly scalable concurrent data structures such as concurrent graph representations and concurrent worklists to implement work-efficient algorithms. These libraries can be used by data scientists to write new applications in Python, leaving it to the underlying system to optimize the program for distributed, heterogeneous execution.

High-performance analytics on Intel Xeon

The platform can now take advantage of the latest generation Intel Xeon scalable processors and Intel Optane persistent memory technology to process massive graphs on much smaller clusters. This will better support organizations with huge unstructured datasets and graphs, including online retailers, financial institutions, and identity management companies, in understanding their customers, operations and opportunities.

Katana Graph’s platform is deployable out of the box on both 2nd gen and 3rd gen Intel Xeon scalable processors with Intel deep learning boost, with no additional modifications required. On 3rd gen Xeon scalable processors, Katana Graph’s solution offers a marked improvement in speed combined with the ability to process much larger datasets while using fewer machines.

“We are proud that our Katana Graph Engine runs up to twice as fast on Intel’s 3rd gen Intel Xeon scalable processors than on the previous generation right out of the box, and delivers even more performance with optimizations,” said Keshav Pingali, Katana Graph co-founder and CEO. “Our customers need the high-performance analytics capabilities that we are unleashing by working closely with Intel.”

On behalf of Intel, Wei Li said, “Katana Graph has leveraged the innovative architecture of the 3rd gen Intel Xeon scalable processor in its state-of-the-art graph analytics engine, which is exciting news.” Li, who is vice president and general manager of machine learning performance in the architecture, graphics, and software group at Intel, added, “Our mutual customers will directly benefit from our close collaboration as we address the challenges associated with processing massive datasets.”

In addition to the increased speed and the ability to handle massive amounts of data, end users will also see notable savings in terms of cluster size and management cost. Katana Graph said this is part of its mission to provide the best graph software stack to allow organizations to glean deeper insights faster from their unstructured data, and to use that actionable intelligence to expand their business and accelerate growth.

Intel Capital leads $28.5 funding round

Katana Graph was founded in 2020 by Pingali and Rossbach who were professors at the University of Texas at Austin. They recognized the synergies made possible by research breakthroughs in graph algorithms, storage, and hardware acceleration that can enable developers to work with irregular and unstructured data at massively increased scale and efficiency.

Earlier this year, the company completed a $28.5 million series A financing round led by Intel Capital with participation from WRVI Capital, Nepenthe Capital, Dell Technologies Capital, and Redline Capital.

Intel Capital managing partner, Anthony Lim, said, “Katana Graph’s platform helps large enterprises make sense of their large unstructured data sets, and we’re seeing this demand across a variety of industries from social networks to biomedical and pharmaceutical research. We are excited to lead Katana Graph’s new funding round based on our collaboration last year where Katana Graph’s technology was optimized for Intel Xeon processors and Xeon-based clusters.”

Scott Darling, president of Dell Technologies Capital, said the Katana platform was a breakthrough solution integrating data ingestion, querying, and analytics with unprecedented scale and performance to address the data deluge problem for unstructured graph data. He added, “We are also excited to back the company given Dell Technologies’ and Katana Graph’s shared roots in the University of Texas at Austin.”

Meanwhile, industry veteran and investor, Lip-Bu Tan, the founding managing partner of WRVI Capital, said, “I am excited and honored to provide initial capital and partner with professor Keshav Pingali and professor Chris Rossbach to provide the most comprehensive platform for large-scale graph data mining, query and analytics.”

Katana Graph last year added a growing number of enterprise clients in the pharmaceutical, fintech, identity, security, and EDA markets, as well as strong momentum in the big data analytics market. While founded by researchers, the company has also added a commercial-focused leadership team with track records of running successful technology businesses or divisions, including Farshid Sabet, chief business officer, who was GM of edge AI at Intel. In addition, N. R. Narayana Murthy, founder of Infosys and Amy Chang, board member at Procter & Gamble and Cisco, joined as board advisors for Katana.

Katana Graph’s key value proposition is the ability to handle massive unstructured data and integrate complex pattern mining workloads on large enterprise data sets and knowledge graphs, with superior performance for complex graph AI, graph pattern mining and graph analytics algorithms. It supports heterogeneous clusters of computing resources including x86 CPUs, Arm CPUs, GPUs and other accelerators. Its platform enables development of high-performance graph AI, graph pattern mining and graph analytics applications, and seamless integration with graph querying.

Related Contents:

For more Embedded, subscribe to Embedded’s weekly email newsletter.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.