April 30 – May 2, 2024 | Stanford University in Palo Alto, CA

Cloud Forum 2024 Presentation Abstracts

— Back to the Cloud Forum Schedule

April 30 Presentations

Telling Our Story: The Somewhat Painful, Probably Never-Ending Search for Cloud Metrics

Rick Rhoades, Penn State University
1:45 pm | Slides

As Penn State continues to mature their cloud offerings, the need to measure success and tell our story has become increasingly critical. In the ever-changing world of cloud, all good adventurers know that these journeys should not be taken alone. Join us as we embark to discover the metrics needed to navigate this unfamiliar terrain with confidence.

GreenEx: Put the Planet into your TCO

Bob Flynn, Internet2
2:15 pm | Slides
The decision about where to run compute workloads typically involves factors like cost, security, and available skills. Yet rarely do you hear sustainability cited as a factor or a motivation to use resources like the public cloud.

Meanwhile, although the public cloud is mostly transparent on service charges, is cheap actually the same as least expensive? What about the cost to the planet? Much is made by the hyper-scalers about the power efficiency of their data centers, the clean energy used, and the low emissions. Is this all greenwashing or is there substance to their claims? Which metrics do they tout and which do they gloss over? Are adequate metrics released to allow us to make a quantifiable comparison between providers and even individual regions or availability zones?

Finally, if we had solid metrics, what would it take to get individual researchers, departments, and institutions to factor sustainability into their TCO calculation and decision-making when it comes to computing? Most institutions obscure the cost of power and cooling at an institutional level, rather than factoring it into the total cost of computing. What kinds of carrots could be developed to incent a shift in thinking to always choose the more efficient and sustainable option?

In this presentation, we will outline the status of efforts to fully scope and quantify the sustainability of the public cloud providers, walk attendees through efforts (and challenges) to do the same with an on-premises computing environment, and propose ideas for how to develop the right incentives to move higher ed institutions to put their computing money where their sustainability mouth is.

Lightning Talks

3:15 pm | Slides

  1. Cloud Credit Craze! Stanford’s Guide to Making it Rain (in the cloud) – Lucrecia Kim-Boswell and Bruno Velazquez, Stanford University
    Discover the collaborative odyssey between Stanford University’s Hosting Services team, Burwood, and the Center for Human-Centered Artificial Intelligence (HAI) in revolutionizing access to Google Cloud research credits. This presentation delves into the dynamic partnership forged among these entities, spotlighting the fusion of technological expertise from Burwood, the academic prowess of HAI, and the operational finesse of Stanford’s Hosting Services. Unveiling the narrative of innovation, challenges surmounted, and the synergistic harmony among these partners, this session unveils how this triumvirate redefined access to cloud resources. Join us for an inspiring narrative that showcases the power of collaboration in propelling research initiatives forward within the realm of Artificial Intelligence and beyond.
  2. Lessons in Scaling Cloud Object Storage – John Bailey, Washington University in St. Louis
    In this lightning talk, I will share additional lessons learned from WashU’s journey to store on-premises server backup data in cloud object storage. Having now scaled this practice up to near petabyte scale with billions of objects, I will share stories of our battle scars with the aim of helping our peers avoid some of the pitfalls we ran into.
  3. Kubernetes in a Snap – David Lacey, The Getty
    Kubernetes is everywhere, but it can be difficult to deploy and manage. Many projects and companies have developed tools to make it easier to get a Kubernetes cluster up and running. Learn how the Getty uses IaC and ArgoCD to run and manage multiple Kubernetes clusters.
  4. Attracting and Retaining Diverse Talent – Kari Robertson, University of California Office of the President
    Research has proven that diverse teams are more productive due to the variety of skills, backgrounds, and perspectives. At University of California, we strive to create diverse teams and have developed pointers that can help any institution do the same. Let’s rethink what skills make for the ideal cloud engineer, leader, or student intern.
  5. InstaCloud, the Turn-key Solution – Richard Guo, Stanford University
    InstaCloud at Stanford is a turn-key offering to cloud resources without the cloud overhead. Get University IT approved computing resources fast and more importantly, affordably. Implemented on AWS. Obtain VMs, GPUs, and container runtime environments with a simple submission form.
  6. Zero to GCC High in 8 Weeks – Gabriel Geise, Penn State University
    Penn State and its associated UARC embarked on a plan to enable a GCC high tenant and secure landing zones in Azure GovCloud for research. After selecting an implementation partner, and completing a statement of work, the project team was tasked with having a CMMC/NIST-800-171 rev2 compliant environment completed in 8 weeks.
  7. Simplifying Cloud Operations for Research – Shruthi Sreenivasa Murthy, Saint Louis University
    At Saint Louis University (SLU), researchers leverage geospatial information to generate valuable research datasets applicable to a diverse array of investigations encompassing economics, public health, and finance, specifically focusing on social mobility trends. These datasets comprise extensive microdata files, with an average size of approximately 500 terabytes spanning five years of data. In this presentation, I will discuss how we built a substantial geospatial data repository utilizing AWS EMR serverless and how we successfully streamlined both cost and data processing durations with this workflow.

Meeting Our Users with Cloud Checkups

Matthew Rich, Northwestern University
4:00 pm | Slides
The Northwestern CloudOps team began a process in May 2023 to meet with all of our nearly 100 cloud account owners regularly. 6 months in, the effort has been a mixed success, but one worth continuing and improving. In this short talk, I will discuss motivations and objectives, the process we designed, areas for improvement and interesting takeaways from meeting with our users.

Attracting Researchers to the Cloud

Jan Cheetham, University of Wisconsin-Madison
4:30 pm | Slides
This talk will describe incentives and initiatives to encourage cloud adoption by researchers at the University of Wisconsin-Madison. Public cloud platforms represent a vast suite of DIY resources for researchers who want to host applications, compute at scale, and build complex data pipelines. In theory, the cloud has many characteristics that align well with what researchers need for research tools: nearly limitless resources, flexibility for innovation, and allocatable expenses. However, despite notable examples of adoption by the research community, cloud resources have been largely unexplored by this population. We sought to understand some of the factors that make researchers “cloud-hesitant.” Our experiences include some successful cloud adoptions and some setbacks. From both situations, we’ve learned about barriers and what can help researchers get past those barriers. We will share our initiatives, findings, and current thinking about how to get more researchers to the cloud.

May 1 Presentations

More than just Research: Understanding and Addressing Gun Violence with a Data-driven and Community-engaged Approach

Susan Burtner, Northwestern University
9:00 am | Slides
The Center for Neighborhood Engaged Research and Science (CORNERS) at Northwestern University conducts cutting-edge research into the ways that neighborhood science can help build safer, healthier, and more equitable communities. Many of our research projects focus specifically on community violence intervention (CVI), particularly the dynamics of gun violence among individuals and groups across space. We work closely with community-based organizations from the research design phase to the dissemination of findings to ensure that our partners’ knowledge and experiences are embedded in the scientific process. We then integrate data from partner organizations and publicly available sources to best understand and analyze the challenges faced by those most impacted by gun violence. Our research not only contributes to the science of neighborhoods and gun violence prevention, but it illustrates the power of data-driven and engaged research to shift the research paradigm, elevate community voices, and support gun violence preventionists in enhancing their effectiveness and reach.

Digital Healthcare using Consumer Devices and Cloud Computing

Peter Washington, University of Hawaii
9:45 am | Slides
My lab at the Computer Science department of the University of Hawaii works closely in collaboration with the Wall Lab at Stanford School of Medicine to build digital phenotyping, diagnostic, and therapeutic solutions for a variety of psychiatric conditions and pediatric developmental delays. These solutions involve a series of distributed devices (smartphones and smartwatches) collecting structured protected health data from patients in real-time, sending the data to the cloud, training deep learning models using the data, and coordinating between the cloud-hosted AI models and the lightweight digital therapies. This process involves extensive cloud computing infrastructure for data collection, data storage, machine learning model training, and real-time collaboration between the models and distributed devices. This talk will describe this process and highlight challenges and opportunities in cloud computing for digital health.

Artificial Intelligence for Animal Farming: Cloud Computing and Automation

João Dorea, University of Wisconsin-Madison
10:15 am | Slides
Our research specializes in advancing and deploying large-scale AI solutions tailored for the agricultural sector, with a particular focus on animal husbandry. We host the largest computer vision network within the United States for livestock industry, our setup incorporates over 200 cameras that gather a diverse array of data, including RGB, infrared, and depth imagery on a daily basis. This extensive surveillance allows us to monitor various aspects of animal behavior, growth patterns, and mobility, which facilitates the early detection of potential health issues, evaluates lactation efficiency, and ensures the overall welfare of the livestock. Our cutting-edge system is built upon a hybrid infrastructure of edge computing and cloud platforms, which allow us to deliver highly efficient, scalable solutions capable of real-time analysis and enhanced farm management. For the development of robust deep learning algorithms, we harness the power of local servers. These models are subsequently deployed across cloud services and edge computing devices, such as the NVIDIA Jetson Nano and TX2, among others. This framework not only supports our research but also serves as a cornerstone for educational and outreach programs. Through these initiatives, we are committed to training the next generation of students in the agricultural sciences, equipping them with the knowledge and skills to innovate in the field of smart farming.

Rapid Evaluation of AI Tools

John Bailey, Washington University in St. Louis
1:30 pm | Slides
With an AI arms race in full swing, I believe institutions need to develop a set of techniques for evaluating cloud-based AI tools as they become available. In this talk, I will discuss and share examples of how Washington University in St. Louis has evaluated various AI tools both to measure fit for purpose (how well a particular AI tool meets 1 or more use cases that matter to our institution) and level of advancement (how well does it perform the task compared to similar AI product offerings from competitors.)

New Developments and Challenges in Using Large Language Models in Academic Settings with Azure Service

Anna Alber, Chapman University
Phillip Lyle, Chapman University
2:00 pm | Slides
Our presentation will highlight the use of Large Language Models (LLMs) at our institution. We will discuss the challenges and considerations associated with the use of Generative AI in conjunction with Azure cloud services for academic and teaching purposes.

Leveraging the Cloud in Response to Campus Demands for AI

Timothy A Werth, Purdue Global
Justin Ward, Purdue Global

2:45 pm | Slides
Generative AI tools, such as large language models (LLMs) are rapidly transforming various industries, including higher education. Our institutions must react to the internal and external demands, opportunities and threats these tools present while doing so in a responsible manner. Every institution, at some point in the very near future, will have to meaningfully confront this topic with strategic and tactical activities in order to keep pace in this quickly advancing landscape. Higher education institutions are increasingly adopting these tools to enhance teaching, learning, and research. However, deploying and managing these tools on-campus infrastructure can be challenging due to computational resource requirements, expertise limitations, and scalability constraints. Current project’s underway at Purdue Global are aimed at leveraging the cloud for deploying and managing these tools with a focus on scalability, cost-effectiveness and reduced human/SME burden. In an environment where many of our institutions, big and small, are having to quickly evaluate and strategize how to proceed it is clear that understanding how the cloud can support our journey is essential to our success with AI. We intend to propose and inform how Purdue Global is leveraging and governing AI tools through: Cloud-based infrastructure: Managing workloads, storage and computations resources. Managed AI services: Cloud-based services handling tasks such as model training, deployment, and monitoring. Integration with existing systems: Learning Management System and CRM integration. Training, Support, and Governance: Governance structures, task forces and exploratory groups lead to engagement, excitement and knowledge transfer. By implementing this cloud-based strategy, higher education institutions like Purdue Global can respond to the growing demand for generative AI tools while ensuring that these tools are deployed and managed in a secure, scalable, and cost-effective manner. We all must be able to harness the power of generative AI to enhance teaching, learning, and research, ultimately improving student outcomes and institutional success.

With Great Capability Comes Great Cost (Containment)

David Love, Stanford Medicine Research Technology
3:15 pm | Slides
In 2018, the Stanford Medicine Research Technology team embarked on the creation of a petascale data lake on Google Cloud Platform. One goal in this move to the cloud was to “replace tedium with innovation,” shedding infrastructure management and focusing instead on creating new solutions with powerful technologies like BigQuery. Over time, we’ve brought more and more capability to the cloud, and our data lake has grown to multiple petabytes…and with that increase in scope came an increase in costs. We’ve learned many lessons while balancing the power of the cloud with the responsibility of containing its costs, and would like to share a number of those lessons with you.

May 2 Presentations

CloudBank at Five Years

Rob Fatland, University of Washington
11:30 am | Slides
Since 2019 three universities — UCSD, UCB, U.Washington — have collaborated to pilot CloudBank, an NSF-sponsored multi-cloud access brokerage program primarily serving academic computer science researchers (330 such projects to date). This pilot program has found challenges and successes across the cloud-as-research-platform spectrum: In terms of financial operation and in terms of case-by-case research cyberinfrastructure building. This review will cover configuration, management, and governance of cloud access (UCSD responsibilities), curriculum augmentation through the UC Berkeley Data8 project, and the education, outreach and training aspects of CloudBank implemented towards optimizing successful, efficient use of the cloud in many sub-domains of computer science (UW’s responsibility). We will particularly review the distribution of CS research subdomains supported by CloudBank by topic, frequency of occurrence and cloud spend. We are particularly interested in the central process of training program participants to use the cloud efficiently; and in the meta-topic of where and how to engage more broadly with academia on CI skill-building. We will also describe successes and challenges on the vendor side of the CloudBank proposition, including the value brought to this program by CloudBank’s partnership with cloud resource management company Strategic Blue. Finally, we will present a perspective on future direction and include time at the forum for discussion.

Flattening the learning curve: A comprehensive training paradigm

Priya Desai, Stanford University
1:00 pm | Slides
Stanford Medicine Research Technology team has developed a comprehensive training program for observational research on top of OMOP Common Data Model and OHDSI tools to make cloud computing paradigms and cross-organizational collaboration more accessible to Stanford biomedical researchers. The training program includes asynchronous learning modules, live workshops and user group meetings. In this talk, we will discuss the challenges our researchers face using cloud services. We will further showcase how we collaborate with stewards of our research computing platform on Google Cloud, and OHDSI community to streamline data science experience.

Landing Zone Accelerator at CU Boulder

Jason Armbruster, Colorado University – Boulder
1:30 pm | Slides
CU Boulder (with AWS Professional Services support) has deployed the Landing Zone Accelerator for Education to replace our homegrown “landing zone” and is in the process of migrating accounts. This session will outline our goals (Security! Compliance!) the process and our lessons learned along the way.

Azure Infrastructure as Code Framework

George Kopf, Princeton University
2:00 pm | Slides
A technical presentation on Princeton’s approach to creating a maintainable code base for instantiating Azure resources. The presentation starts by discussing IAC, then Frameworks, pros & cons of different approaches and finally presents our real-life example standing up our Azure landing zone using Bicep.

Does your DaaS still Dazzle?

Chris Manly, Cornell University
3:00 pm | Slides
At Cloud Forum ’22, I presented on a relatively new venture of offering DevOps-as-a-Service to other University Departments as a professional services engagement. Now that we’re a bit further down the road, let’s check in and find out if the shine has come off or it is still DaaS-ling our customers. What worked? What didn’t work? What are the challenges and opportunities that we face?

Overcoming Nephophobia: Case Studies

Kari Robertson, University of California Office of the President
3:30 pm | Slides
Drama, Intrigue, and Passion…Migrations from a traditional data center to the cloud can create apprehension for all involved. There is no one size fits all approach, but there are some proven strategies that can help in even the most cloud-hesitant culture. Review real world case studies from different institutions that some might consider ‘creative’ or even ‘mischievous’, but were successful in the end. Caution: None of the case studies result in termination, but similar results cannot be guaranteed.