General stogare limits
| Limit | Value | Configurable |
|---|
| Maximum available storage in TIM | 10 GB | Yes (on license level) |
Dataset limits
| Limit | Value | Configurable |
|---|
| Maximum CSV size | 100 MB | No |
| Maximum number of rows in dataset | 60 000 (default) | Yes (on license level) |
| Maximum number of columns in dataset | 40 (default) | Yes (on license level) |
Quota and throttling of ML jobs
To ensure fair distribution of compute resources between all licenses, a priority is assigned to each executed ML job. Jobs with higher priority get access to a TIM Worker's compute time sooner than jobs with lower priority. Each license can execute a certain number (the throttling threshold) of ML jobs of a defined category over a period of 60 minutes with normal priority. When the license exceeds this throttling threshold, the priority for the next executed jobs of this license is reduced. Priority returns back to normal automatically when the license has executed less jobs than the throttling threshold in the last 60 minutes.
| Job category | Job types | Normal priority | Reduced priority | Threshold |
|---|
| A | predict, detect | 5 | 3 | 60 ML jobs of category A in last 60 minutes |
| B | build-model (F and AD), rebuild-model (F and AD), retrain-model (F) | 4 | 2 | 12 ML jobs of category B in last 60 minutes |
| C | rca (F and AD), what-if (F and AD) | 1 | - | No threshold, all jobs are executed with normal priority of category C |