April, 13th - April, 19th

During this week, our work was focused on the simple predictors for the job completion times and upload times. The job completion time predictors are based on the fact that we know in advance how our jobs behave, which allows us to predict how long it will take them to finish, based on the input file size. The upload time predictor is based on a constant average upload speed previously measured after a considerate number of uploads. The use of the predictors allows our Load Balancer to use our local data center more efficiently, instead of bursting the job to the cloud.

The next step was to test if we could manage to run two jobs simultaneously without compromising the performance of both jobs. After our tests, we concluded that it is infact possible to run two jobs in simultaneous without degrading their performance and completion time. Both jobs took the exact same time to complete. So, provided that the jobs have the required amount of containers for their execution, it is possible to run at least two jobs simultaneously in our cluster without compromising performance.

So, our Load Balancer was updated to predict and allow the running of two simultaneous jobs. As of this point, the idea is to fine tune the algorithm to fit as many cases as possible.

NEXT OBJECTIVES: Finish fine tuning the solution. After the solution is adequatly fine tuned, we believe we can consider that we have hit a milestone. Thus, the writing of the sections of the dissertation regarding this milestone can start soon.