Introduction

Problem Statement

Methods

Findings

Limitations and Future work.

Conclusion

NetApp (2019) High performance computing (HPC) is the ability to process data and perform complex calculations at high speeds. 

HPCs have become essential to scientific discovery over the last two decades, with researchers actively creating larger systems to accommodate new modes of scientific discovery with complex workflows (Kogge & Borkar, 2008).

High-performance computing is increasingly prominent in higher education, critical to attracting research grants, recruiting top faculty, and teaching data-intensive research methods.

HPC resources are underutilized by non-STEM and non-computer science users due to accessibility barriers. These users often lack the advanced Linux knowledge and command-line familiarity required to access HPC resources.

OSC (2013) found that scientists and engineers would rather advance their disciplines than learn HPC. Learning commands are a hurdle for researchers new to HPC, contributing to the small population of researchers from other disciplines.

Most researchers from other disciplines have rarely used a command line terminal, but they need more computing power to analyze large datasets (Abhinav, 2019).

UC’s ARC as part of the research 2030; therefore, the need to provide a standardized HPC platform and tools for its users.

David Hudak (2013) launched the Open OnDemand project, a web-based GUI portal launched by OSC, reduces the learning curve for HPC by making it as easy to use.

In 2013-2014, Indiana University and Purdue University began separate experiments in graphical computing. Indiana University developed the “HPC Everywhere” portal, which increased the total number of users by 50%.

LANL struggled to find student interns and early-career applicants with skills to fill its HPC workforce needs. LANL created SCC and CSI to develop its HPC workforce (Connor, 2016).

Results

Heuristic Evaluation

Data Analysis

Think-Aloud Activity

Survey

Data Collection

Data from the survey was visualized and categorized into various groups, specifically Superusers and Non-Superusers, which gave us a good understanding of why they selected their answers.

Thematic Analysis to identify themes, keywords and  concepts that were discussed by our participants, and we  refined iteratively among researchers

Based on the data gathered from the think-aloud activity, a heuristic evaluation was performed.

This is a method of judging the severity of usability flaws by comparing them with the 10 Nielson Norman usability heuristics principle.

A severity rate was also provided on a scale of 5 for each violation noted.

Our survey had 21 responses, and two incomplete submissions, 11 respondents representing (57%) out of the 19 were superusers, and the  remaining 8 (43%) were novice or non-superusers. 

Majority of the HPC Cluster survey respondents were graduate students (15) being (75%), 11 (73%) students out of the (75%) were Doctoral students, four students representing (27%) were pursuing a Masters Degree.

Faculty members represented 20% of the respondents (4), and 1 staff member at the Office of Research used the resources of the HPC for heavy data analytics and complex reporting (5%).

Five users participated in the Think Aloud Activity, and are referenced by pseudonyms such as P1, P2, P3 etc.

Three superusers and Two non-super users 

Three themes were identified 

First, Connection Methods. Second, Job Management. Third, Navigating the HPC portal 

Out of the 9 tasks assigned to our interviewees:

In total, 18 problems were identified with a severity greater or equal to 0.

8 were problems greater than or equal 2 – these were primary usability problems related to complex methods of connecting to the cluster

3 problems were rated a severity of 4 and one was rated a severity of 3. The rest were cosmetic problems which didn’t affect users or hinder their use of the system.

To address our research question on how HPCs can be made more accessible for use across disciplines at UC and its implications.

​

Our research proved the need to develop a customized Web GUI HPC Management portal for UC's HPC users based on our results. About 95% of users, both superusers and non-supers, recommend the need for one and its implications if action is not taken. 

Users also preferred to connect to the cluster directly instead of using the VPN as a means of security but an alternative 2FA.This impacts the inability to integrate third party tools into their codes.

Increase Workforce Development efforts, ARC develop a standardized semester-based class users can take for credits. Also, can be an onboarding. Method, Also introduce SCC, Summer Camps e.g. Early IT.

Very small Sample size ,Broaden the user scope.

Difficult finding users willing to participate.

Build a customized Intuitive portal for the users of the HPC Cluster.

Work with ARCC and the Office of Research ,SoIT /Engineering to design a High-performance Computing course for student credits.

Work on more grants to support ARCC to continue addressing the research computing needs of the  University  and create a niche in Higher education

Thank you

Questions