Short version

Here's a few vignettes I've had to write for different speaking engagements and proposals. I put them here so I don't have to write them from scratch every time I'm asked, and hopefully they're up to date.


Here's a biosketch as of April 2024:

Glenn K. Lockwood is a Principal Engineer at Microsoft, where he is responsible for supporting Microsoft’s largest AI supercomputers through workload-driven systems design. His work has focused on applied research and development in extreme-scale and parallel computing systems for high-performance computing, and he has specific expertise in scalable architectures, performance modeling, and emerging technologies for I/O and storage. Prior to joining Microsoft, Glenn led the design of several large-scale storage systems, including the world’s first 30+ PB all-NVMe Lustre file system for the Perlmutter supercomputer at NERSC. He holds a Ph.D. in Materials Science.

Research statement

The SC conference asked for my research statement (ostensibly) to match me to paper submissions as a technical program committee referee. I don't actually do research, so it's mostly a restatement of my biosketch.

Glenn K. Lockwood's research interests lie in the intersection of artificial intelligence and supercomputing infrastructure, with a specific focus on workload-driven system architecture design. His previous work focused on applied research and development in extreme-scale and parallel computing systems for high-performance computing, and he has specific expertise in scalable architectures, performance modeling, and emerging technologies for I/O and storage. His current work has pivoted towards holistic system design specifically for AI training and inferencing workloads and understanding how to map the needs of specific AI applications to different compute, network, storage, and infrastructure technologies.

Long version

Here's the story of my professional life. I try to keep it up to date.

Early Life and Education

I was born in Hawaii and moved to the mainland when I was quite young. I grew up in the suburbs of central New Jersey, just off of 8A on the turnpike. After completing high school, I attended the School of Engineering at Rutgers University where I majored in ceramic engineering and spent a year and change doing research using molecular simulation at the Interfacial Molecular Science Lab. I graduated in the final class of ceramic engineers at Rutgers; the program was replaced by one in materials science and engineering in the following year.

Ceramic science is a fascinating one, but a field trip to a local glass factory in my final year as an undergraduate made me realize that I really did not want to be an engineer when I grew up. To defer having to go out into the real world to make a living, I opted to go to graduate school instead and enrolled in the materials engineering program at Lehigh University in Bethlehem, Pennsylvania. My hope was to become an electron microscopist, as Lehigh had some of the world's most sophisticated aberration-corrected scanning transmission electron microscopes at the time (supercomputers of the microscope world, in a sense).

Those plans did not pan out, and after a semester in Pennsylvania, I returned to Rutgers to continue doing research with molecular dynamics simulations.

Getting into Supercomputing

In the four years I was doing my Ph.D. work at Rutgers, I started collecting old UNIX workstations. In the course of playing with them all, I began benchmarking my molecular dynamics codes on them, and this gave me an appreciation for the nuances and performance features of various old RISC architectures. I had also started getting involved in the HPC community online, and I credit my friends (now colleagues) on IRC and Twitter with providing the inspiration and knowledge to consider HPC as a career rather than a mere hobby.

During graduate school I also married my wife (and spent two months doing research out of her father's tractor shop on the Canadian prairie--an interesting experience!). When I finally completed my dissertation, I made the decision to change careers and pursue my interest in HPC full-time. And, since I had also forced my wife to live with me in suburban New Jersey for four years, I looked for jobs in places that were most un-like suburban New Jersey. Much to my great fortune, the San Diego Supercomputer Center had an open position in user services, and I was able to trick them into thinking I knew enough about HPC to be employable.

SDSC provided me with access to the resources, expertise, and support to develop a strong foundation in high-performance computing, and my role in user services put me in a position to solve challenging issues in a variety of scientific domains. Most notably, the next-generation sequencing industry (largely centered in San Diego) began growing into the realm of HPC at that time, and my understanding of the intricacies of SDSC's exotic data-intensive supercomputer led me to a number of consulting projects in sequencing.

My analyses of the computational requirements demanded by DNA sequencing caught the attention of a few companies who were technologically ahead of the curve. I was eventually extended a unique opportunity to join an early-stage sequencing startup that was developing a revolutionary new product, and my desire to experience the startup life and be a part of such a rapidly growing industry lured me away from SDSC.

Foray into Genomics

After eighteen months in San Diego, I moved up to Oakland and joined 10x Genomics as a DevOps Engineer (a job which I had no idea how to do). My job involved running a small cluster, maintaining a couple hundred terabytes of storage, and integrating 10x's software product with HPC infrastructure and cutting-edge DNA sequencers.

My heart never left supercomputing though, and shortly after my one-year anniversary at 10x, I returned to the world of supercomputing at the National Energy Research Scientific Computing Center (NERSC) at the Lawrence Berkeley National Laboratory (LBNL). Although I only worked there thirteen months, working at 10x was the experience of a lifetime. I had the honor of working with some of the most exceptional people I've ever encountered in my professional life, and I learned skills and perspective that have benefitted me tremendously in my career since.

Storage and I/O at NERSC

At NERSC, I joined the Advanced Technologies Group where I helped with systems design and procurement to ensure that the next generation of supercomputers will be able to keep pace with the demands of future scientific research. I was one of the resident experts in parallel I/O and new storage technologies, and the emphasis of my work was on parallel I/O system architectures. This included understanding the low-level hardware details of non-volatile storage media, new software technologies for high-performance I/O transport, optimization points for various parallel storage systems including Lustre, Spectrum Scale, and object stores, and where all of these technologies were going.

In late 2019, I was asked to lead NERSC's Storage Systems Group which is responsible for managing most of the center's production storage systems totalling over 200 PB of tape and 120 PB of disk. I was responsible for managing eight storage engineers and a significant capital budget, and during my tenure we brought in two new staff and procuring and deploying a 60 PB Spectrum Scale file system. However, the challenges of being a first-time manager amidst a global pandemic and the loss of the technical responsibilities led me to step down from this acting position a year later and return to the Advanced Technologies Group at NERSC.

After returning to Advanced Technologies, my role was still predominantly storage-focused and composed of three major responsibilities:

  1. Staying abreast of new technologies, coordinating relationships with storage vendors, and evaluating new technologies as they may pertain to future HPC systems.
  2. Advocating for NERSC externally (whether it be to other HPC facilities, the vendor community, or the public at large) to ensure that the great work we do, the great systems we deploy, and the unique workload requirements we face are well represented across the HPC community.
  3. Shepherding cross-team projects and providing technical accountability for the purposes of procurement, contracting, external oversight, and project management.

My role had me oversee increasingly larger projects. Once the 35 PB all-flash Lustre file system I oversaw was delivered and accepted as part of the Perlmutter system, I moved on to overseeing the overall technology integration strategy and program surrounding NERSC's follow-on system, NERSC-10. I struggled to settle into this role and ultimately left NERSC in May 2022.

Moving to the cloud

In June 2022, I joined Microsoft as a Principal Product Manager in the Azure Storage organization. I was designated Azure's go-to expert in HPC storage, and I was tasked with developing an overall product strategy for HPC storage across Azure's first-party and marketplace services. In practice, this meant helping Azure's own storage product teams (like the Blob and Azure Managed Lustre) define the performance and features required by HPC workloads. I also worked with external HPC storage companies to figure out ways their products could complement Azure's own storage offerings.

Over the course of eighteen months as an Azure Storage product manager though, I learned that there is a huge difference between being an HPC person in a storage organization and being a storage person in an HPC organization. I also learned that HPC, not storage, is what gets me out of the bed in the morning, and I don't make a great product manager. These epiphanies led me to realize that being a product manager in a storage organization was not what I wanted to do for the rest of my life.

So, in January 2024, I changed roles within Microsoft. I moved laterally along two dimensions:

  1. I went from the product management discipline into the engineering discipline. This brought me closer to the familiar world of being hands-on with HPC technologies.
  2. I left the Azure Storage organization and joined the Azure Specialized Workloads organization. Azure Specialized is the umbrella under which all the compute infrastructure that isn't general-purpose (like GPUs and InfiniBand) falls.

I am now a Principal Software Engineer supporting the HPC infrastructure used by Microsoft's largest AI customers. My responsibility is to understand what massive-scale AI workloads will need to accomplish in the coming years, how those workloads plan to approach those problems from a software standpoint, and how that workload-centric view should steer the overall system architecture of next-generation supercomputers.