oar-team / batsim

Batsim: Infrastructure simulator for job and I/O scheduling
GNU Lesser General Public License v3.0
30 stars 15 forks source link

[pybatsim] Easy scheduler is overlaping jobs #14

Closed mickours closed 7 years ago

mickours commented 7 years ago

I dont know if it is directly related to pybatsim but I have overlapping jobs using Easy and EasyNotopo schedulers of pybatsim. It sometimes give a 200% utilisation for a short time. Just run the example from the readme: https://github.com/oar-team/batsim#run-batsim-example but change the fillerSched by easyBackfillNotopo or easyBackfill.

mpoquet commented 7 years ago

Cannot reproduce the issue. Added pybatsim_tests in c9542e1c3f but they seem to pass (https://gitlab.inria.fr/batsim/batsim/pipelines/403). Batsim is supposed to halt if the same resource is used by several jobs at the same time (unless time-sharing is enabled). Furthermore, I don't see any overlapping when I visualise the execution traces, what workload are you using? easyNOTOPO_medium.pdf easyNOTOPO_tiny.pdf easy_medium.pdf easy_tiny.pdf

mickours commented 7 years ago

In fact there, with a closer look on the data, I found the problem: Sometimes the job finish time and the next job begin time is exactly the same because the scheduler has no scheduling delay.

I tried to patch the code to add some scheduling delay but I have some assert errors... I will see with @glesserd if he can help me on this.

mpoquet commented 7 years ago

The issue seems to be more related to a data analysis issue than a Batsim one, closing it.