Failure Prediction on google cluster-usage trace, 2019 version

[Adrian Tuns]

Hello,

I am working on my Computer Science master's thesis, on job and task failure prediction with Machine Learning, based on the 2019 dataset. I am trying to follow a similar approach with the one from this paper https://journalofcloudcomputing.springeropen.com/articles/10.1186/s13677-022-00327-0, where the 2011 dataset was analyzed. 

There are two features that I am not sure how should be taken into consideration, namely the Alloc Set and Alloc Instance. I understand that in the previous traces, they were considered the same as Jobs and Tasks.

If anyone could provide any insight regarding how the alloc set/instance would influence a ML failure prediction, it would be very much appreciated.