OpenPOWER enables datacenters to rethink technology approach

Dipankar Sarma of IBM explains how the company is accelerating performance by adopting an open source approach in processor technology.

Sejuti Das Apr 05th 2017

The advances in artificial intelligence (AI) have led to significant innovations in hardware technology. A robust hardware has now become the next great frontier as the speed of decision making matters more than ever before.

Businesses are aggressively focussing on machine learning, deep learning, and natural language-based sensory data at a massive scale. Meanwhile, as AI and deep learning are becoming the absolute imperatives, hardware is no longer being spoken of widely. 

Also read: Can AI knock the human fallacies out of HR?

On this note, Dipankar Sarma, Distinguished Engineer, India Systems Development Lab at IBM, highlights how the company is changing the game when it comes to advancing in processor technology with a new open source approach. This is accelerating applications while improving performance for greater datacenter efficiency.

IBM has taken traditional capabilities of POWER processors to use them to solve problems like providing cognitive assistance and enterprise decision-making.

Now that robust hardware has become the next great frontier, speed of decision making is of utmost importance. What are your thoughts?

For enterprises, decision making goes far beyond a common set of instructions and extends to the way of using data, workloads, and systems. In order to meet customer demands, businesses need open system design that provides robustness, greater flexibility, speed, and scale-up capabilities at lower costs.

Robustness in hardware can happen in multiple ways—resilience to failure, good quality of service under high demand, and ability to detect and mitigate problems. These capabilities enable timely decision making in organizations.

We, at IBM, have taken the traditional capabilities of our POWER processors and use them to solve problems, like providing cognitive assistance and enterprise decision-making. 

Also read: Intel bets big on its datacenter business and AI in 2017

How is the industry innovation capturing growth against vertical segments like retail, telecom, banking, etc.?

With quintillion bytes of data created each day, there is a definite requirement for advanced systems that are able to crunch structured and unstructured data to deliver actionable insights within seconds. However, virtual explosion of data and growing customer demands are accelerating change in our industry.

Be it retail, telecom, banking or even e-commerce, the ability to manage, train, sort and analyze the data through the system is critical to gain insight quickly from the plethora of sources and traditional architectures.

Also read: E-commerce: The industry is sold on AI and machine learning

Until now, these high-performance architectures were closed and proprietary with no collaborative development. But IBM has adopted an innovative, open business model through its OpenPOWER initiative.

OpenPOWER initiative has created an ecosystem of collaboration that is driving real innovations. An open server architecture allows tighter integration at all levels of the hardware and software stack, which is essential for an effective data-centric design. It allows any OpenPOWER foundation member to bring their own innovation to the table.

OpenCAPI standards evolving from this enable truly heterogeneous computing with processing elements from IBM and its partners, all coherently sharing memory. These innovations enable incredible speed-ups in application performance with accelerators, deploy a processor architecture designed for HPC performance, and benefit from ecosystem innovations.

What are the innovations that India Systems Development Lab (ISDL) is bringing to the table on a global scale?

ISDL is a part of the IBM Systems worldwide technology development lab. Developers at ISDL deliver technology innovations across the entire systems portfolio. The team works across all areas of computer systems design and development—starting from processor design, firmware, operating system, to software-defined storage.

We also have a considerably strong presence in the open source technologies like Linux, OpenStack, Docker containers, and kubernetes. These are the key parts of infrastructure offerings running on our systems. Our engineers have also worked on key storage technologies that integrate public cloud with on-premise storage and next-gen storage transportation.

OpenPOWER initiative has created an ecosystem of collaboration that is driving real innovations.

What are IBM’s contributions to the OpenPOWER ecosystem?

The OpenPOWER foundation was established as an open technical membership organization that will enable data centers to rethink their approach to technology.

IBM and its OpenPOWER consortium partners have successfully allied with multiple nations to define their national agenda on large-scale supercomputing projects. These projects include CORAL Supercomputer for US Department of Energy, UK's Science & Technology Facilities Council, Barcelona SuperComputing Center, Human Brain project with Juelich Super Computing Center in Germany, and Exascale agenda with GENCI France.

We have also recently announced the OpenCAPI consortium, which essentially establishes OpenCAPI as the next-gen cache-coherent interface that can also have accelerators like FPGA.

What are the challenges IBM is facing while participating in an OpenPOWER ecosystem?  

One of the main challenges for IBM is to remove perceptions about applications for OpenPOWER. It provides a tremendous platform to run open source software and has resulted into industry leading performance numbers for open source databases and runtimes. All of the popular Linux distributions, both enterprise and open source versions, are available on the platform, along with many other open source software packages.

This platform demonstrates 10,000 docker containers running on a two-socket system, and has also broken the barrier of one million queries per second.

What is the hyper-scale opportunity for IBM OpenPOWER?

Last year, we announced 50 new innovations in OpenPOWER to help companies solve problems around big data. Several of these are technologies for hyper-scale data centers. These include new Open Compute Specification compliant servers developed by Rackspace, and Google using POWER8 and POWER9 chips.

IBM also has a product line called Linux Cluster (LC), which addresses this market. The LC line infuses OpenPOWER technology into IBM’s scale-out server lineup.

Can OpenPOWER and collaborative framework of the foundation enable India's national computing agenda and home-grown innovations?

The massive expansion of India’s computing needs in cloud cognitive systems involving OpenPOWER technology can play a big role. To efficiently run workloads, traditional analytics is transforming itself to use more cognitive machine learning capabilities and OpenPOWER ecosystem.

IBM has announced the opening of the first OpenPOWER Research Facility (OPRF) at IIT Bombay to help drive the country’s National Knowledge Network initiative. This aims to interconnect all institutions of higher learning with a high-speed data communication network, facilitating knowledge sharing, collaborative research, and innovation.

OPRF will provide technical assistance and infrastructure to further indigenous research. It aligns well with the recently-launched National Supercomputing Mission of the Indian government.

What does the future look like?

The data deluge that many had predicted before, is now a reality. We have already seen businesses struggling to manage volumes of data generated by today's interconnected world.

In the future, we will witness a shift in computing, which will bring processing closer to data in order to derive the best insight in a timely manner. We will also see a stronger role of special purpose computing using accelerator technology, allowing a lot of data to be pre-processed before it reaches the main compute elements.

Additionally, we will witness real heterogeneous computing with accelerators working side-by-side to provide the most efficient data processing capabilities needed for high-performance cognitive applications.