I have had some interesting back-end questions posted to me recently.
Implementing a time series store / and sum_submerge method.
In this particular vein, I felt like solutions similar to ReductStore and PyStore were worth a look.
But I felt that I was at a loss for the overall theory in terms of time-series data vs a more traditional relational data used to model and create SaaS like I have been building most of my life.
I can definitely see how having a fleet of GPU’s one would want to collect telemetry data, and then use said data to quantify their performance and lifespan.
Using predictive analytics to predict the failure of a device, and preemptively removing it a top tier where the best clients are paying top dollar for said fleet of devices.
Would seem like a good idea to try and create a dataset of devices, and optimal telemetry vs thresholds for failure.
Also tracking deltas on things which could signify performance degradation.
Another crushing boulder has been dropped on me, with all the stuff that I don’t know being tacked on. Feels like atlas has become a splatter.