Colloquium 2023 DIY job monitoring, from cache misses to CO2 footprint

We'll describe some sources of information you can access from within a job context, and show how they may offer insights into your job's behavior. The scope of this talk is relatively simple, DIY-type data collection, rather than more complicated profiling frameworks. For example, our systems expose power counters which your programs can query, if you're interested in carbon footprint. There are many other "gauges" and counters that can provide information about memory use, cache misses, sources of slowdowns, etc.