|
|
* Description:
|
|
|
|
|
|
Misuse of compute resources wastes funding and reduces availability for legitimate purposes.
|
|
|
One type of misuse is crypto mining, where compute resources are used to generate crypto currency.
|
|
|
|
|
|
* Criterion:
|
|
|
|
|
|
Mining often uses specific optimized instructions that are uncommon in other HPC workloads:
|
|
|
detected = sum(counters) > threshold
|
|
|
|
|
|
* Works for shared jobs with existing metrics:
|
|
|
|
|
|
Easy, if counters are per core, so it can be associated with a job.
|
|
|
|
|
|
* Possible false positives:
|
|
|
|
|
|
There may be legitimate software that makes use of the same instructions.
|
|
|
|
|
|
Approaches to deal with them could be:
|
|
|
- whitelist of users and applications
|
|
|
- combination with other metrics
|
|
|
|
|
|
* Possible cures/workarounds:
|
|
|
|
|
|
Lock user.
|
|
|
|
|
|
* Missing Data:
|
|
|
|
|
|
TDB: How can we collect the relevant counters.
|
|
|
|
|
|
|
|
|
Crypto Miner
|
|
|
```
|
|
|
Input:
|
|
|
* Instruction use counter
|
|
|
* Threshold value
|
|
|
|
|
|
Rule:
|
|
|
detected = sum(counter) > threshold
|
|
|
|
|
|
Output: detected
|
|
|
```
|
|
|
|