Superdoop
At the direction of a research team at Intel Labs in Pittsburgh, our team studied the ecosystem and administration of large network data nodes and focused on the monitoring and maintenance of those systems. From this research it was our intent to reveal inefficiencies and ultimately design a better tool for system root-cause analysis and system management. The final design incorporates features from multiple levels of system management.

This research has been accepted to the ACM CHIMIT Conference (Boston, December 2011)

Features of the new system include direct control over primary actions on running jobs:


Process

We studied the workflow and process of three primary systems users and two administrator level users. We learned of their tools and work practice and came to understand their needs. Using a combination of contextual inquiry, workflow analysis, and scenario walkthrough we gathered data and created consolidated Flow, Sequence, and Cultural models. Our design goals emerged from the research:

• Provide a single point for diagnosis-relevant information.
• Offer prioritization of information.
• Increase awareness among end-users of the computational cost of their jobs.
• Avoid information overload.

We developed a click-able prototype using Fireworks and Cogtool (Keystroke Level Modeling software). Simulating a real system, we studied four users stepping through assigned tasks.

The prototype has a more polished look than we would normally implement on a first round but we needed higher fidelity visualizations and markings which could not be achieved as easily without a matching context. Aesthetic choices beyond the web 2.0 look were derived from needs observed during the research phase, including special coloring for alerts and status.

Older visualizations were inadequate for comparing processing performance. New visualizations were added to offer multiple dimensions of comparison:


An example screenshot of the new interface:


Examples of the current interface:
/ (1 of 1)


by James Mulholland, Ben Gotow, Mark Shuster, Sriram Ramasubramanian, and Arun Ganesan
in cooperation with Intel Labs Pittsburgh
Coursework for Methods Lab
Masters of Human-Computer Interaction
Carnegie Mellon University, Fall 2010