arXiv CSopen access

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding

2026 · arxiv_cs

arXiv CS · Papers · License: Open Access · 2026

software-architecture software-engineering testing

software engineering, software architecture, testing

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding AYUSH CHATURVEDI, ROB POKORNEY, ELYN FRITZ-WATERS, CHARLOTTE ROUSE, GARY BAX, DARYL SPENCER, and CRAIG POHL, Washington University in St. Louis, USA Research computing centers around the world struggle with onboarding new users. Subject matter experts, researchers, and principal investigators are often overwhelmed by the complex infrastructure and software o!erings designed to support diverse research domains at large academic and national institutions. As a result, users frequently fall into confusion and complexity to access these resources, despite the availability of documentation, tutorials, interactive trainings and other similar resources. Through this work, we present a framework designed to improve new-user onboarding experience. We also present an empirical validation through its application within the Research Infrastructure Services at Washington University in St. Louis. CCS Concepts: • Social and professional topics → Computing literacy; Computer science education; Software engineering education; Information technology education.

Additional Key Words and Phrases: New User, Training, Onboarding, Advanced Computing, High Performance Computing, Supercomputing ACM Reference Format: Ayush Chaturvedi, Rob Pokorney, Elyn Fritz-Waters, Charlotte Rouse, Gary Bax, Daryl Spencer, and Craig Pohl. 2026. Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding. 1, 1 (April 2026), 14 pages. https://doi.org/10.1145/nnnnnnn.nnnnnnn

Introduction

Research computing centers (RCCs) or high-performance computing (HPC) centers are critical to the scienti"c community. Millions of researchers and students worldwide rely on critical computing infrastructure to advance science. Thus, it is evident that accessibility and rapid onboarding to RCC services mandate the least amount of time, e!ort, and complexity. However, this seldom happens due to computer science and engineering principles being rooted in the RCC sta! and foundation; after all, they provide computing services. As a result, end users: researchers, principal investigators (PIs), and students "nd themselves lost in a set of complex documentation, video libraries, and deviations in RCC infrastructure as compared to their local workstation. A similar case was in our center, where users often complained about a poor onboarding experience. Hence, we decided to overhaul all aspects and entry points of user onboarding journey. We carried out our e!orts for over a year, in phases. With some success, we share our journey and a novel framework dedicated to improving user experience (UX) in HPC and RCC centers in this work. Below, we outline background of our center and its user followed by problems in our onboarding services. Authors’ Contact Information: Ayush Chaturvedi, [email protected]; Rob Pokorney, [email protected]; Elyn Fritz-Waters, [email protected]; Charlotte Rouse, [email protected]; Gary Bax, [email protected]; Daryl Spencer, [email protected]; Craig Pohl, [email protected], Washington University in St. Louis, St. Louis, Missouri, USA.

2026. Manuscript submitted to ACM PEARC 2026 Manuscript submitted to ACM PEARC 2026

Chaturvedi et al.

1.1

WashU-RIS Background

Washington University in St. Louis (WashU) maintains a proli"c research ecosystem with landmark contributions to genomic sequencing [16], metabolic microbiome function [33], Alzheimer’s disease progression [4], and Martian habitability [14]. Sustaining an annual research portfolio exceeding one billion dollar necessitates sophisticated computational expertise; however, these e!orts were historically distributed and domain-focused. As a result, several internal collaborations formed with evolution of institutional computing capabilities over thirty years. In 2017, WashU leadership formed the WashU Information Technology (IT) research computing department and WashU IT Research Infrastructure Service (RIS) group. Originating from the Human Genome Project [16] and part of the McDonnell Genome Institute, WashU IT RIS brought a strong background in genomics work#ows, high-speed data transfers, high-throughput computing, container technologies, multi-petabyte parallel "lesystems, and security to support both basic science and clinical research e!orts. 1.2

WashU-Researcher Background

WashU is primarily composed of students, researchers, faculty and PIs working at two physical campuses separated by a large urban park. On one side are the medical schools and the large urban hospital, on the other are undergraduate and graduate schools. Due to such diverse user community, WashU lacked common central services until the formation of WashU IT in 2014, which consolidated IT organizations and services across both campuses. Despite central IT services, until RIS began operations, departmental IT teams and researchers were themselves responsible for technological solutions for their programs. This resulted in hundreds of di!erent technology solutions, operational standards, and funding mechanisms. 1.3

Motivation

In 2018, RIS released its "rst ‘Storage Service’ accessible from a network share or an HPC cluster followed by release of integrated HPC services in 2020. Onboarding users required researchers and students to submit a ticket for ‘Storage’ or ‘Compute’ services. Provisioning access and providing a list of instructions and references to our documentation was a standard onboarding procedure. Provisioning was catered through internal ITSM portal supported by Jenkins. For technology support, we leverage JIRA for service desk, request management and project management. Users were expected to understand our architecture and technology stack, be willing to learn by consulting our documentation, and understand the gap between our services and determine how to use them e!ectively. However, as evident from Table 3, user who request computing infrastructure services at RIS lack a formal computer science or engineering background (only ↑ 8%sponsors and ↑ 6% departments). As a result, users reported a lack of personalized

support, complex navigation, and insu$cient guidance, resulting in a steep learning curve and user dissatisfaction. Addressing these shortcomings was imperative to improve user retention and satisfaction and prompted us to revamp our onboarding processes to ensure a more intuitive, e$cient, and supportive user experience. Consequently, we ran a project for one complete year to improve onboarding experience which we share as framework for the broader community. 2

Prior Works

Computing centers around the world have made many e!orts to enhance the user onboarding experience. Julich supercomputing center [37] created dedicated courses similar to PNNL [10], while University of Colorado Boulder [18, 26] Manuscript submitted to ACM PEARC 2026

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding3 improved documentation, website and tutorials in an attempt to improve user experience. They also highlight the role of community engagement and collaborations with other local computing consortium. Although this is a great way to educate users from non-computer-science backgrounds, it may not be as useful much for users with a computer-science background. Other works such as [1, 35] attempts to provide similar frameworks as ours, however, they only cover either domain speci"c areas to onboarding or deviates from onboarding in their primary focus to beyond. Moreover, both lack testimonial evidence supported by data. 3

Framework for Improving New User Experience

With increasing complexity in infrastructure and services around it, in RCCs, the entry point for researchers often becomes opaque and confusing. This results in unwarranted speed bumps for ‘research computing users’, and in turn, delays essential scienti"c discovery.

Fig. 1. Overview of Framework for Improving User Experience at a Research Computing Center.

Therefore, we introduce a comprehensive framework designed to replace ad-hoc methods and philosophies with a streamlined, robust, and articulated set of processes to help improve user onboarding experience. Figure 1 shows a detailed single-point bird’s eye view of our framework that consists of six phases. By institutionalizing this blue print and practices outlined in this work, computing centers can drastically reduce the user learning curve and administrative overhead for sta!. Our pioneering work converges multiple aspects at a computing center to a single goal aimed to champion overall user experience. Principles and method we share can minimize ‘time-to-science’, ensuring that researchers can transition from registration to computation with e$ciency and con"dence. Below, we elaborately discuss each part in this framework. Manuscript submitted to ACM PEARC 2026

4 3.1

Chaturvedi et al. Initial Evaluation and Identification

Initial evaluation for current state of a!air builds a robust foundation for planning and evaluation. Before any onboarding interventions can be designed, a research computing center (RCC) must establish a baseline understanding of current user ecosystem with objective and subjective evaluation. Additionally, drawing from information technology service management (ITSM) [9, 22] and user experience (UX) [8, 27, 32] postulates, RCC sta! should try to quantify and diagnose friction points and align service delivery with actual ‘research user’ needs. The process and methods for identi"cation and evaluation can be outlined in three critical categories as addressed below. 3.1.1 Identify Entry Points. An RCC ecosystem always has multiple entry points for users to interact with the o!ered services. Mapping the user journey requires identifying every touch point where a current or potential user may interact or request services for the "rst time. These often include RCC’s website, ticketing portal, and other gateways and service portals, such as Open-OnDemand, ColdFront, and XDMoD, etc. Other indirect entry points can originate from RCC collaborations and science gateways. For e.g., an NSF NAIRR Pilot user may choose to use Jetstream2 cloud infrastructure services and the ‘Exoshpere’ [15] science gateway. Therefore, identi"cation of all direct and indirect entry points early on is critical. Formalizing these entry points ensures consistent information delivery and e!ective triage of how the users arrive. 3.1.2 Services O!ered: Direct and Indirect. When planning for a change in user experience, identifying all the services is a primary step. These services can be classi"ed into the following categories: • Direct Services (User-facing): Tangible resources which the user interacts with immediately, such as HPC cluster

interfaces, service portals, storage services (scratch/project), computational infrastructure (CPU/GPU/xPUs), application software, and direct hosted science gateways.

• Indirect Services (RCC facing): The "invisible" layer running in RCC that a!ects user accessibility. In-house

utility tools, security infrastructure and services extended through RCC collaboration with other organizations, for e.g., an Internet2 subscription may provide ‘Eduroam’ access to all university members. Other examples, such as Active Directory (AD) services provided by the RCC parent department or organization.

• Latent Services (Utility): The latent set of services includes other utility services to empower user productivity

such as: consultation, training materials and repositories, grant writing support, training workshops and seminars, hackathons, etc.

Latent and mission-critical services are often oblivious to end users, however, some of them serve as the spinal cord for the RCC itself. Hence, identifying them early on is extremely critical for planning, development and rollout phases in later stages. 3.1.3 "antify and Evaluate Current State. Deciding on key KPIs and formulating a data-driven strategy has been a industry standard for decades. Nonetheless, quantifying UX at an RCC can be challenging if there is no infrastructure for it. Using technologies that support analytics, capture user clicks, session timing and journey mapping features can often capture drops and friction points in user experience. Hence, quantifying and analyzing a bigger picture can reveal patterns and help in mapping areas of poor user engagement with technological bottlenecks. Manuscript submitted to ACM PEARC 2026

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding5 3.2

Data Curation

Systematic data collection, accounting, and analytics quantify an RCC’s high-impact research cataclysm role. Tracking the growth of speci"c user groups and their software usage often plays a vital role in linking HPC usage to research output acting as a testament to a great user experience. Below, we share strategies for deriving business analytics and practices for storing them for informed decision making. 3.2.1 Business Analytics. The primary motivation for collecting data is to provide a bird’s-eye view of RCC services and to understand patterns of RCC users’ usage to further improve their experience. Such business analytics will help leadership and planning committees make informed decisions and target areas with the maximum user footprint or usage. Creating dashboards for software usage, job-level heuristics, allocation-wise active usage, and classi"cation of batch-job errors with respect to software and applications are some examples. Collecting data on user a$liations and professional information can also reveal unexpected patterns of cross-disciplinary activity. Advanced RCCs correlate job accounting data with bibliometric databases (e.g., Web of Science, Scopus) and grant awards. This allows centers to quantify computing time in terms of publications with respect to federal and departmental funding. This "impact-based accounting" is often the deciding factor in securing institutional subsidies and renewing federal infrastructure grants [2]. Another important usage metric is ‘license usage’. It allows RCCs to identify underutilized licenses, facilitating the "right-sizing" of agreements and renewals that lead to potential future cost savings. 3.2.2 Data Management. Many previous scholarly works provide insight into the importance and impact of collecting telemetry data at an HPC center [7, 28]. Frameworks such as "MELT" combine metrics with telemetry to identify user usage patterns [21]. However, our framework goes beyond telemetry data. We believe in the ‘Data Triangulation’ approach that precedes the monitoring and collecting phase and covers data beyond telemetry. We believe that data from customer-facing portals plays a critical role in providing UX analytics. Surveys often complement this step, but have a thin hit rate. Identifying mechanisms to save, click paths, and session duration are some techniques used by most cloud and social media service providers [21] that operate under strict Service Level Agreements (SLAs). In the context of storing this analytics data, our framework adheres to a simple tenet: "Knowledge consuming more time to fetch than help in taking action is not worth the e!ort". Systematic, chronological storage is fundamental to extracting actionable intelligence and ensuring operational e$ciency. Chronological ordering serves as the backbone for time-series analysis [25, 36]. Long-term retention enables the discovery of high-level patterns, such as seasonal anomalies or multi-year shifts in user behavior, which are often invisible in real-time operations. To support this, an archival strategy to ensure that "cold" data remains immediately accessible for compliance and retrospective study without clogging the primary data storage utility is an e!ective approach [31]. 3.3

Planning and Orchestration

The user experience in any RCC is the "nal test of leadership and their decisions. Throughout year, the entire sta! works hard to keep millions of moving parts in the center running smoothly to provide unfettered access to RCC services. Hence, in this part of the framework, we intend to provide assistance and a checklist for the RCC leadership, the core governing body that lays the path for the center’s future. 3.3.1 Objectives and Goals. Setting organizational objectives, target goals, and timelines is not new [5]. However, in the context of improving RCC UX, we provide a guide for setting priorities and developing plans. We hypothesize Manuscript submitted to ACM PEARC 2026

Chaturvedi et al.

tents, based on our learning and experiences that shall assist the RCC leadership to de"ne objectives by assuming the following about users in an academic setting: • RCC users always want more than one entry point for software and hardware resources. • Users want a computing experience as similar as their local workstation.

• There will always be a set of users who do not like the current state a!airs.

• The user will never remember training material that is unappealing to them. • A GUI portal is never completely disliked.

Oftentimes engineering sta! at RCCs rarely interact with users. Hence, we hope that the above tents can also help translate user presumptions to such teams that are less or not involved with users on regular basis. 3.3.2 Budgeting and Discussion. HPC centers are constantly evolving their underlying hardware and software services. Millions of dollars are spent on funding the scienti"c service life cycle. Formal terms and calculations, such as TCO (total cost of ownership) and ROI (return on investment) [19] are common in the RCC business model. These existing methods and studies recognize dedicated "nancial allocations for Human capital, software, hardware, infrastructure, energy, and other similar aspects related to the RCC itself. However, currently, there are no "nancial provisions or classi"cations dedicated to improve the user experience. The "nancial and non-"nancial costs involved in developing training and technologies to educate users are often obscured within the existing categories of the RCC business model. The cost and energy used to develop indirect services discussed in Section 3.1 are also discounted in a similar manner. Therefore, our framework seek to voice for a dedicated category within RCC’s business model aiming at UX improvement. 3.3.3 Architecture & Licensing. Designing the UX architecture for an HPC center requires balancing various opposing forces: the immense complexity of the underlying infrastructure, security, compliance and the cognitive simplicity required for e$cient use of interface. Therefore, architectural planning should also account for any necessary licenses and technology. We provide four basic principles for UXDC: • HPC UX should eliminate the need for dependencies. The Open OnDemand portal is one great example; everything runs in a browser.

• Modern HPC UX architecture should focus on increased accessibility and adoption of science work#ows rather than a single unit of compute like Jobs, GPUs, CPU, etc. The "Science Gateway" model succeeds only when it covers the "Long Tail" of science; the end-to-end process, not just the calculation step [20]. • The new UX architecture should integrate CLI-GUI integration via APIs. The Web Portal and the CLI should both call the same underlying API. This allows a user to "Start a job in CLI, Monitor it on Web, and Visualize it on Tablet." • HPC interfaces and entry points often obscure operational complexity to facilitate accessibility. However, we

propose that should be made a user’s choice, and that the UX architecture should be designed with ‘Operational

Transparency’ in mind, aligning with Nielsen’s "rst heuristic ("Visibility of system status") which is often neglected in HPC architecture [24]. 3.3.4 Implementation and Phase-out Plan. In HPC centers, the implementation and phase out of services are not merely technical milestones; they are critical "User Experience (UX) Events." From a UX perspective, these phases represent the highest friction points for researchers. Poor implementation of UX tools can create a "learning cli!" that discourages adoption. Similarly, a poorly planned phase-out of old services may create "migration anxiety" or data loss. Hence, it is vital to balance and time both events: the implementation of new UX and the phase out of old UX. It is important for Manuscript submitted to ACM PEARC 2026

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding7 the plan to have two faces, one for the users and the other for internal tasks. Also, note that the implementation and roll out strategies for each of the development tasks, archival and restoration of analytics data should be documented at this stage and communicated to respective stakeholders. 3.4

Development Phase

3.4.1 Policies and Process. The very "rst step in UXDC is developing policies and processes. Most RCCs are part of a central division in the University, such as the ‘Department of Information Technology’. Hence, they are subject to the central university and departmental policies. This can be both challenging and championing for improving the UX. The most critical policy area is security. Firewalls, load balancers, and public-facing infrastructure may be subject to additional security measures and university or higher departmental policies. RCCs that have a user base working with grants from sensitive organizations such as DoD (Department of Defense), NIH, and the like must account for additional policies and security measures in place; HIPAA [34] and NIH’s latest guidelines [23] are some examples. 3.4.2 Websites and Portals. GUIs are seldom hated by RCC users from a non-computer science background. OpenOnDemand’s global acceptability and success are one such testimony. Hence, RCCs should mold and develop open-source software according to their needs. Filling blank space with something is not an ideology that this framework supports. We propose that RCCs should facilitate a UI portal based on usage patterns identi"ed, allocated budgets, and timelines to go live. RCCs can collaborate and identify existing science gateways that work for their center. Regional computing consortia are a great place to meet local RCC sta! and users to discuss patterns in region-speci"c usage and UX solutions. 3.4.3 Trials, Betas and Proof of Concepts (PoCs). Introducing software and usability change may face cultural resistance from a predominant set of practices. Hence, all development should be tested with a special set of users identi"ed in the implementation plan (Sec. 3.3) [11]. A Beta phase allows you to recruit in#uential PIs (Principal Investigators), leadership personnel, and faculty who can validate tools and act as ‘proof’ in the new UX concept. Developing ‘Beta Recruitment Emails’ alongside the ‘User Agreement,’ which may clearly set expectations for release and usage is the preferred approach. However, PoCs and trials should be complemented with a mechanism for observability and feedback to quantify impact of the change [3, 17]. Additionally, adhering to policies or taking exceptions is a step often missed or skipped under the excuse of ‘beta testing’. 3.4.4 Training Resources. Many previous attempts have been made for ’perfect training resources’. Micro-credentials [6] for users, interactive courses, video repositories, and an innovative learning platform for documentation are some recent examples [12]. However, there is no panacea or a perfect set of training resources due to the high cardinality of the worldwide HPC user base. Centers should constantly "nd new ways to train their users. However, we advocate the tents described in Sections 3.3 and 3.1. All training resources must not depend on the software environment, and the training experience should closely resonate with the user’s local workstation. 3.5

Deployments and Release

Our framework categorizes steps involved in deployment and release of services into distinct category. This process is initiated through comprehensive communication strategies that ensure the diverse RCC user set is informed of upcoming changes and maintenance windows. Manuscript submitted to ACM PEARC 2026

Chaturvedi et al.

3.5.1 Communications. RCC should engage in pre-deployment communication and stakeholder engagement to prevent a gap between systems administration and end-user productivity. Communication roll out strategy should be developed like a story of changes from a user’s perspective. The chronological sequence of events and their impact on the user is what matters. A well-thought-out plan and proposed (user-facing) architecture should be transparent to users and noti"ed in accordance with timelines. Transparency is essential for maintaining the "Availability, Accessibility, and Usability" (AAU) metrics that de"ne modern supercomputing democratization [30]. Communications should be crafted for each tool or platform leveraged in RCC. With the advent of modern generative AI tools, we believe it is now time to retire text-based communication in favor of more graphic-based material. Lastly, regular and timely reminders of critical deadlines should also be pre-planned in communication strategy. 3.5.2 Service Upgrades. Upgrading HPC services is no longer a hardware endeavor. With a rich set of software stacks and ever-evolving complex networking and storage solutions, HPC admins have to take a ‘swiss army knife’ approach to service upgrades. Timing, versions, and ordering in upgrades matter at centers working with thousands of computational nodes, software modules, and integrations with Single Sign-On (SSO) for authentication. The challenge becomes even steeper when compliance, vendor, and additional governmental policies are considered, especially at national labs and similar venues. Hence, administrations should draw up a ‘release framework’ that minimizes user-level disruptions and infrequent maintenance windows. 3.5.3 Go-Live and Releases. The "nal step in deployment is to champion the timelines and prioritize organizational goals. ‘Go-Live’ activities must also account for software provenance. Using external platforms for documentation and client-side binaries provides a vital fail-safe, ensuring researchers can access critical support resources even during primary system maintenance windows. 3.6

Feedback and Support

Retrospective analysis in HPC often focuses on identifying "cognitive friction" points where the system forces the user to divert attention from their research to handle administrative or technical overhead. Releasing products and improvements to users does not marks the end of UX improvement journey. In the context of cyberinfrastructure, User-Centeric Design provides the foundational theoretical framework for this retrospection. Studies have shown that even experienced researchers can be demotivated by the steep learning curve of parallel programming and job management [32]. Hence, RCC leadership should also focus on strategies to capture feedback, monitor user engagement and develop strategies to engage with PIs and researchers. 3.6.1 Feedback. In the HPC context, collecting UX feedback that is in alignment with previously established data analytics and storage strategies often providing meaningful insights. Training sessions and specialized courses that provide micro-credentials is a good strategy to increase user engagement [6]. These credentials provide a way for researchers to demonstrate pro"ciency in specialized domains such as GPU programming, parallel computing with MPI, or ethics in AI. Moreover, engaging with ‘User Advisory Groups’ for feature prioritization is often a most practical and direct feedback. 3.6.2 Support. There is no question to the fact that continued support is a critical piece in user experience. O$ce hours or portal requests are structured consulting mechanisms. We argue that in addition to IaaS, SaaS; RCCs also provides "Consulting-as-a-service" (CaaS) which should be put in the same tier as other service o!erings since CaaS consume a chunk from the same pile of resources, "nance and operations. Furthermore, by implementing post-release Manuscript submitted to ACM PEARC 2026

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding9 pipelines, communications and establishing timed-windows for patching, HPC centers can signi"cantly reduce the cognitive friction associated with CI/CD cycles in scienti"c computing [13, 29]. 4

User Onboarding Case Study At RIS

Fig. 2. Old Vs. New Onboarding Workflow

In Fall 2025, we launched our new Compute2 cluster with a completely new onboarding experience with entire ‘Onboarding Revamp’ project ran for entire year with a ‘User-Centric’ approach. We analyzed and quanti"ed the current challenges users experience through extensive data analytics and strategies. Finally, we formulated an implementation plan to roll out changes in our internal processes, policies, and various software infrastructure used by RIS users. 4.1

Policy Changes

With our new Compute2 cluster launch in Fall 2025, we rolled out several policy changes intended to accelerate changes in our operating process aimed toward better user experience. Several new policies such as VPN-free access to ‘Open-OnDemand’ portal, increased SSH idle-time outs, mandatory training requirement for account access, setting up of minimal viable product were laid out. These policies had direct and indirect a!ect on user experience as they detemine the course of technical architecture. Table 4 shows impact of these changes across our services and to the end user. 4.2

Website & Ticketing Portal Integration

To improve the onboarding experience, our website and ticket portal were redesigned to deliver a more integrated, user-friendly interface across all entry points, as noted in Section 3.1. This e!ort involved a comprehensive technical overhaul to help users more easily start and continue their journey with RIS. Portal interfaces and website entry points were integrated with JIRA APIs for a strong focus on accessibility, speed, and ease of use. Additionally, our goal was to Manuscript submitted to ACM PEARC 2026

Chaturvedi et al.

reduce friction and make key information easier to "nd. For that, we completely changed our ticket portal from having a ‘service-oriented’ design to ‘user-oriented’. Figure 3 shows the contrast for service-request categorization in the old and new portal. These initiatives led to reduce response time and incorrect labeling of user requests.

Fig. 3. User Request categorization in Service-oriented old portal Vs User-oriented new portal.

4.3

New Documentation Portal

Previously, RIS user-facing documentation was a standalone website lacking analytics capabilities and demanded development infrastructure and e!orts for regular updates. Consequently, the documentation was migrated to ‘Con#uence’ removing development hurdles, streamlining access management and eliminating local con"guration dependencies while promising feature integrity. Con#uence also o!ered superior integration with other atlassian products used at RIS, speci"cally with the JIRA service desk. This integration enabled ‘ROVO AI assistant’ which is capable of answering from knowledge base articles in Con#uence and raising service desk requests in JIRA that. Table 1 show metrics for AI assistance and documentation portal before and after launch of Compute2 cluster. As evident, with new cluster and its documentation, there is drop in old cluster’s view of compute Ultimately, this integration supports a more holistic onboarding process and provides an even more extensive experience to RIS services for new users. Table 1. Rovo AI and Documentation Metrics Before and A!er Compute2 Launch Component

Item

Before (Compute2 Launch)

After

Rovo AI

AI-assisted Queries Resolution AI-assisted Queries Escalation AI-assisted Queries Non-Resolution

28 70 84

10 67 116

Documentation

Compute Cluster 1 Compute Cluster 2

1106 N/A

816 1454

4.4

Creation of Mission Critical Team

With a goal to enhance user support and reduce response times and friction for user queries, RIS leadership created a dedicated ‘Mission Critical Team’ that included members from across teams with di!erent skills and expertise. The Manuscript submitted to ACM PEARC 2026

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding 11 primary responsibilities were handling outages, complex user tickets and reduce friction in ticket resolution that demanded triage from di!erent teams in the department. 5

Results & Conclusion

Improving the user experience is a cycle. Table 2 shows the impact of our changes made to website, portal and policies respectively for the last year. We measured KPIs for a period of three months each, before and after rolling out these changes with our ‘Compute2’ cluster. As evident, we see an increased footprint at our website. Similarly, improvements in o$ce availability by RIS sta! and website improvements resulted in increased engagement in o$ce hours. With ticketing portal improvements, we see a drop in incorrect request categorization in general and for onboarding requests. Customer satisfaction also increased due to automated follow ups and standardized responses which also played a part in reduction in SLA times for ‘First response and Resolution’. Departmental policy changes and creation of mission critical team are other additional factors that contributed to reduction in these times. Moreover, the leadership was also pleased to see improved communication skills and camaraderie across department. Table 2. Impact of Various Changes Done in Website and Ticketing Portal During User Onboarding Experience Component

Website

Ticketing Portal Policy

KPI

Before (1 Apr–1 Jul)

After (1 Sep–1 Dec)

Onboarding Page Visits – Average Monthly O$ce Hours Page Visits – Average Monthly

N/A 213

941 317

Booked O$ce Hours Sessions Cost Calculator – monthly downloads User Sessions – monthly average

75 N/A 1561

73 30 1934

Incorrect Request Categorization Incorrect Onboarding Requests Avg. customer satisfaction

29/115 (25%) 11 4.9/5

30/224 (14%) 14 4.96/5

Number of outages

N/A

SLA Avg. Time to First Response (Incident/SR) SLA Avg. Time to Resolution (Incident/SR)

5.93 Hrs / 7.02 Hrs 1.92 Days / 2.34 Days

4.4 Hrs / 2.38 Hrs 1.49 Days / 22.49 Hrs

Through this work we aim to share a robust framework designed to serve as a guiding lamp for those determined to enhance user experience at their HPC centers. We hope that our experience and insights gained through our journey shall provide a valuable resource to the wider research community. References [1] Gladys Andino et al. 2024. Onboarding Research Computing and Data Professionals. In Practice and Experience in Advanced Research Computing 2024. doi:10.1145/3626203.3670582 [2] Amy Apon et al. 2014. Assessing the e!ect of high performance computing capabilities on academic research output. Empirical Economics (2014). doi:10.1007/s00181-014-0833-7 [3] Team Atriade. 2024. 7 Steps for Successful Proof of Concept in IT Infrastructure. White Paper. Atriade Consulting. https://atriade.com/7-proof-ofconcept-steps/ Industry-standard framework for running PoCs, emphasizing that "Success Criteria" must be de"ned *before* the PoC starts.. [4] Randall J Bateman, Tammie LS Benzinger, Scott Berry, David B Cli!ord, David M Holtzman, Mathias Jucker, John C Morris, et al. 2012. Clinical and biomarker changes in dominantly inherited Alzheimer’s disease. New England Journal of Medicine 367, 9 (2012), 795–804. doi:10.1056/NEJMoa1202753 [5] Mary Jo Bitner et al. 2008. Service Blueprinting: A Practical Technique for Service Innovation. California Management Review (2008). doi:10.2307/ 41166446 Manuscript submitted to ACM PEARC 2026

Chaturvedi et al.

[6] Dhruva Chakravorty et al. 2024. BRICCs: Building Pathways to Research Cyberinfrastructure at Under Resourced Institutions. In Practice and Experience in Advanced Research Computing 2024. doi:10.1145/3626203.3670535 [7] Liuhua Chen and Haiying Shen. 2017. Considering resource demand misalignments to reduce resource over-provisioning. IEEE INFOCOM 2017 (2017). doi:10.1109/infocom.2017.8057084 [8] Hashim Chunpir et al. 2017. User Experience (UX) of a Big Data Infrastructure. In Lecture Notes in Computer Science. doi:10.1007/978-3-319-58524-6_37 [9] Alice E. Cook et al. 2021. ADVANTAGES, CHALLENGES, AND SUCCESS FACTORS IN IMPLEMENTING INFORMATION TECHNOLOGY INFRASTRUCTURE LIBRARY. Issues In Information Systems (2021). [10] Ryan Danehy and Jaelyn Litzinger. 2025. Review of the Documentation Activities and Training for HPC at PNNL. In Practice and Experience in Advanced Research Computing 2025: The Power of Collaboration. doi:10.1145/3708035.3736005 [11] Sandra Gesing and Katherine Lawrence. 2018. Science Gateways: Leveraging Modeling and Simulations in HPC Infrastructures via Increased Usability. In Proceedings of the 2018 IEEE International Conference on Systems, Man, and Cybernetics (SMC). [12] Ben Godfrey. 2024. Practical advice for the creation of e!ective HPC user documentation. In Practice and Experience in Advanced Research Computing 2024. doi:10.1145/3626203.3670621 [13] Ben Godfrey. 2024. Practical advice for the creation of e!ective HPC user documentation. In PEARC ’24. doi:10.1145/3626203.3670621 [14] JP Grotzinger, DY Sumner, LC Kah, K Stack, S Gupta, L Edgar, RE Arvidson, et al. 2014. A habitable #uvio-lacustrine environment at Yellowknife Bay, Gale Crater, Mars. Science 343, 6169 (2014), 1242777. doi:10.1126/science.1242777 [15] David Y. Hancock et al. 2021. Jetstream2: Accelerating cloud computing via Jetstream. In Practice and Experience in Advanced Research Computing 2021. doi:10.1145/3437359.3465565 [16] International Human Genome Sequencing Consortium. 2001. Initial sequencing and analysis of the human genome. Nature 409, 6822 (2001), 860–921. doi:10.1038/35057062 [17] G. Kandaswamy et al. 2010. Software Engineering for Scienti"c Computing: The Role of PoCs and Prototyping. Computing in Science & Engineering (2010). [18] Shelley L. Knuth et al. 2019. An expansion of the user support services for the Research Computing group at the University of Colorado Boulder. In Practice and Experience in Advanced Research Computing 2019. doi:10.1145/3332186.3332229 [19] Jonathan Koomey et al. 2007. A Simple Model for Determining True Total Cost of Ownership for Data Centers. Technical Report. Uptime Institute. [20] Katherine A Lawrence et al. 2015. Science gateways: The long road to the "long tail" of science. Proceedings of the 11th IEEE International Conference on e-Science (2015). [21] J. Miller, K. Agrawal, and A. Panyala. 2025. Navigating Exascale Operational Data Analytics: From Inundation to Insight. IEEE Transactions on Parallel and Distributed Systems 36, 2 (2025), 210–225. doi:10.1109/TPDS.2025.10820578 Provides the MELT framework for HPC telemetry.. [22] Mirghani S. Mohamed et al. 2008. The restructuring of the information technology infrastructure library (ITIL) implementation using knowledge management framework. VINE (2008). doi:10.1108/03055720810904835 [23] National Institutes of Health. 2025. NIH Public Access Policy (2024 Update). https://grants.nih.gov/policy-and-compliance/policy-topics/publicaccess/nih-public-access-policy-overview E!ective July 1, 2025; Accessed: 2026-03-25. [24] Jakob Nielsen. 1994. Usability Engineering. Morgan Kaufmann. [25] Sai Dikshit Pasham. 2018. Dynamic Resource Provisioning in Cloud Environments Using Predictive Analytics. International Journal of Engineering and Computer Science (2018). [26] Dylan Perkins et al. 2022. Challenges and Lessons Learned of Formalizing the Partnership Between Libraries and Research Computing Groups. In Practice and Experience in Advanced Research Computing 2022. doi:10.1145/3491418.3535165 [27] Sebastian A. C. Perrig et al. 2024. Measurement practices in user experience (UX) research. Frontiers in Computer Science (2024). doi:10.3389/fcomp. 2024.1368860 [28] István Pintye et al. 2024. Enhancing Machine Learning-Based Autoscaling for Cloud Resource Orchestration. Journal of Grid Computing (2024). doi:10.1007/s10723-024-09783-1 [29] Zebula Sampedro et al. 2018. Continuous Integration and Delivery for HPC. In Proceedings of PEARC ’18. doi:10.1145/3219104.3219147 [30] Jim Samuel, Margaret Brennan-Tonetta, Yana Samuel, Pradeep Subedi, and Jack Smith. 2022. Strategies for Democratization of Supercomputing: Availability, Accessibility and Usability of High Performance Computing for Education and Practice of Big Data Analytics. Journal of Big Data: Theory and Practice 1 (06 2022). doi:10.54116/jbdtp.v1i1.16 [31] M. Schwartz. 2018. Data Archiving for the Enterprise. O’Reilly Media. [32] Kanu Priya Singh. 2024. A systematic literature review of the application of user experience studies in cyberinfrastructure. Quality and User Experience (2024). doi:10.1007/s41233-024-00069-8 [33] Peter J Turnbaugh, Ruth E Ley, Michael A Mahowald, Vincent Magrini, Elaine R Mardis, Vickie G Musick, and Je!rey I Gordon. 2006. An obesity-associated gut microbiome with increased capacity for energy harvest. Nature 444, 7122 (2006), 1027–1031. doi:10.1038/nature05414 [34] U.S. Congress. 1996. Health Insurance Portability and Accountability Act of 1996 (HIPAA). https://www.congress.gov/bill/104th-congress/housebill/3103 Public Law 104-191. [35] A. Wong et al. 2024. A Uni"ed Framework for the Deployment and Access of HPC Applications as Services in Clouds. Journal of Grid Computing (2024). Manuscript submitted to ACM PEARC 2026

Institutionalizing Best Practices in Research Computing: A Framework and Case Study for Improving User Onboarding 13 [36] Yunqi Zhang et al. 2016. History-Based Harvesting of Spare Cycles and Storage in Large-Scale Datacenters. In 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16). [37] Ilya Zhukov and Jolanta Zjupa. 2025. Onboarding of HPC users: hands-on approach at Juelich Supercomputing Centre. In Practice and Experience in Advanced Research Computing 2025: The Power of Collaboration. doi:10.1145/3708035.3736077

A A.1

Appendix A Use of Generative AI

The work uses generative AI platform Gemini-AI provided to WashU in partnership with Google for correcting grammar, syntax and spelling errors. Gemini was also used to "nd a few scholarly articles and publications for the section 2 to reduce human labor intended in searching the vast internet. Gemini-AI partnered exclusively with WashU does not use or curate any data for improving the Gemini-AI. A.2

Tables School