XDMoD has recently been enhanced with a Job Viewer feature that provides users, HPC support specialists, and others with access to detailed job level information. This information includes: accounting data, details about the application being run, summary and detailed job performance metrics, and time series plots describing CPU user, memory usage, memory bandwidth, network interconnect, parallel file system and flops for individual jobs. Virtually all XSEDE affiliated resources have been instrumented with TACC_stats, and as a result are now supplying job level performance data directly to XDMoD. Users and support personnel can use this information to determine the efficiency of a given job, and to guide the improvement of job efficiency for subsequent jobs. This tutorial will instruct the user in how to use the XDMoD Job Viewer, describe the type of job-level information that it includes, and guide participants in the use of the Job Viewer as a user support and analysis tool. This tutorial will be beneficial to all levels of users from beginners looking to understand how to launch parallel HPC jobs properly to advanced users trying to optimize job performance.
Tentative Outline for XSEDE16 Job Viewer Tutorial:
1. Overview of XDMoD (Thomas Furlani) a. Focus on impact to improve code efficiency/Performance b. Examples c. Introduce Job Viewer as a tool
2. Brief Demo of XDMoD (Matthew Jones or Robert DeLeon) a. Overview b. Application Kernels c. Job level monitoring (SUPReMM) d. Job Viewer Introduction
3. Detailed Demo of XDMoD Job Viewer (Joseph White or Matthew Jones) a. How to use the Job Viewer b. Information provided by the Job Viewer c. How to use the Job Viewer as a user support tool
4. Hands-on use of XDMoD and the Job Viewer (CCR personnel) a. Pick a few example cases to show in greater detail b. Emphasis on Job Viewer workflow c. Secondary emphasis on using XDMoD for efficiency and quality of service
