People to Watch 2019

Gregory Kurtzer
CEO & Founder
Sylabs Inc.

Gregory M. Kurtzer is the CEO and founder of Sylabs Inc., the company behind the open source container project Singularity. Sylabs caters to the needs of various compute based workflows like traditional simulation, data science, real time analytics, and AI use-cases. Previously, Greg has spent most of his career enabling massive scale compute focused use cases where he created and led various open source projects along that mission, including the Warewulf cluster management toolkit, CentOS Linux, and most recently, the container system Singularity.

HPCwire: Hi Greg, congratulations on your selection as a 2019 HPCwire Person to Watch. After a couple years of fast adoption, one has the sense container technologies are now broadly embraced by HPC. What’s your assessment of the HPC community adoption of container technology and what changes/advances/features in the technology are still needed to serve the traditional HPC/scientific community?

Greg Kurtzer: What an amazing honor it is to be selected as a person to watch in 2019 by HPCwire; thank you! I’ve been very lucky to have been a part of some really amazing open source projects over my career thus far, such as CentOS Linux, Warewulf, and most recently Singularity, as well as the creation of the company Sylabs Inc.

Sylabs, the company behind the container system Singularity, has solved some major pain points for many people within HPC. For example, Singularity provides a secure means to take an entire application stack (or OCI/Docker based container), and encapsulate it into a container that can be cryptographically signed (encryption coming soon) image format, that is a single, simple file owned by a user. It is just data. Where and how you use, copy, archive, or share that file is up to you, and it’s always exactly the same no matter where you run it! This completely changes the packaging, reproducibility, and mobility paradigms for applications across multiple systems on-prem, off-prem, multi-prem, as well as simplifying the onramp to the cloud!

Moving forward, one of the biggest needs we are hearing about is a better solution for building containers. HPC sites want to give their users the ability to build containers on their systems without the need for root access, bleeding edge unsupported kernels, as well as support building containers for multiple architectures, like ARM and Power. For this reason we created the Remote Builder: a cloud and on-prem based service that allows users to build containers in a trusted way.

Singularity, coupled with the Remote Builder, the Container Library, and the KeyStore, provide the guaranteed reproducibility, immutability, and accountability that is becoming a requirement in many industries (e.g. pharma/FDA, EDA, aerospace, oil and gas, and financial services).

And to make it even easier to use Singularity and facilitate support of container development, testing, and general usage, we will shortly be releasing a desktop version of Singularity that runs on Mac OS X natively (Windows coming). This means that you can build, design, test, and sign containers from your personal

laptop, and then use them anywhere, from an HPC resource to the cloud, making the transition and onramp anywhere seamless.

One major area that is really needing more work is where ABI compatibility is needed from kernel to user space. OFED is a good example of where we need to invest. The good news is that the right people are in discussions now, and we would love to hear from others that would like to be part of that solution.

HPCwire: Turning away from academia and the national labs for a moment, it seems like Sylabs is now focused on the enterprise. Why is that and how are enterprise needs different? What the current breakdown Singularity user base vis a vis traditional HPC and enterprise?

We are facilitating all computational and application workflows at Sylabs. Simulation, research, science, AI, data science, analytics, etc., are all in our wheelhouse. Even though I sometimes compare HPC to “enterprise” in general, it is worth mentioning that there are many enterprises that rely heavily on their large, traditional HPC deployments like pharmaceuticals, oil and gas, aerospace and transportation, financial services, etc.

But there is a new compute focused paradigm shift which is occurring. Many enterprises that have historically only required on non-computational type workloads are rapidly discovering the need to run more computationally intensive workloads, such as machine learning and data analytics. But many of these enterprises, particularly ones which are not ubiquitous to the HPC industry, lack the expertise, tools, and funding necessary to achieve quick and meaningful results in this new data-driven landscape.

This provides Sylabs with a very interesting opportunity: to facilitate the cross pollination between HPC and non-HPC communities. For example, from the HPC side, we’ve been doing parallel, high IO based compute models for decades, which are very similar in nature and requirements to AI workloads. From the microservices side, there are massive strides forward in orchestration of resources, applications, and services which can be used for the streaming of data for real time compute. This is why we have built Singularity integrations for Kubernetes, Kubeflow, and Hashicorp Nomad (with Apache Mesos coming soon).

To summarize, this is a tremendous bi-directional gain for both traditional HPC and historically non-computational based enterprises.

HPCwire: Generally speaking, what trends and/or technologies in high-performance computing do you see as particularly relevant for the next five years? Also, what’s your take on near-term prospects for quantum computing and neuromorphic technologies?

One trend that we are seeing over and over across all forms of compute is the move toward a new kind of real-time streaming workflow as I mentioned above. Users in many different contexts want to run a series of containers as services and want to stream massive amounts of data to these services in real time. Then, rather than saving this data, users want to orchestrate compute focused containers to analyze and then discard this data in near real time. This basic pattern is of data analysis is becoming a ubiquitous requirement.

And of course, to be further buzzword compliant, is the continuing proliferation of artificial intelligence, machine learning, and deep learning within the traditional HPC ecosystem. For example, Tensorflow is becoming a standard installation (usually through Singularity) on many HPC resources, as is the various Nvidia containers distributed via NGC.

As for quantum and neuromorphic computing, it’s still in its early days; though, to state the obvious, there is no doubt that these technologies will have an incredible impact on the HPC landscape for certain types of workflows.

HPCwire: Outside of the professional sphere, what can you tell us about yourself – personal life, family, background, hobbies, etc.? Is there anything about you your colleagues might be surprised to learn?

My background has been centered within government and academia, spending almost 20 years at Lawrence Berkeley National Laboratory as a senior computing architect. During that time, I would love to spend my free time solving problems and building communities around the open source projects that I’ve been associated with (the aforementioned CentOS and Warewulf for example).

Now with Sylabs and Singularity, I’ve turned my love for open source, collaboration, and solving complicated problems into my full time job! This is the most fun I’ve had in a long time, and every day is super exciting with new possibilities, collaborations, architecting solutions, and meeting amazing people from around the world along the way.


Lori Diachin ECP	Talia Gershon IBM	Gopal Hegde Cavium/Marvell

Steve Oberlin Nvidia	Jim Keller Intel	Ken King IBM

Gregory Kurtzer Sylabs/Singularity	Forrest Norrod AMD	Thierry Pellegrino Dell EMC

Michela Taufer SC19 Chair	Steve Scott Cray	Jack Wells OLCF