Method of managing electricity providing in a computers cluster

1. Method of managing electricity providing in a computers cluster ( 9 ), comprising:

a process of prediction of need of electricity ( 8 ) provided by at least one renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ),

a process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), including:

a step of managing failure risk of said renewable electricity source ( 1 , 2 , 3 ), by lowering said predicted availability, so as to:

increase life expectancy of said renewable electricity source ( 1 , 2 , 3 ),

and/or lower maintenance frequency of said renewable electricity source ( 1 , 2 , 3 ),

a process of scheduling tasks ( 7 ) in said computers cluster ( 9 ), based on both said prediction processes ( 6 , 8 ),

a process of dynamical reconfiguration of said renewable electricity source ( 1 , 2 , 3 ), according both to last need prediction and to last availability prediction, said process of prediction of availability ( 6 ) then taking into account renewable electricity source reconfiguration update, and

wherein:

in case of need prediction decrease because of workload decrease, said renewable electricity source ( 1 , 2 , 3 ) will be reconfigured so as:

not running all the time,

and/or not running at full speed when running, and preferably wherein:

in case of need prediction decrease because of workload decrease, said renewable electricity source ( 1 , 2 , 3 ) will be reconfigured so as:

provide extra electricity to the grid if it accepts it at that time, preferably in a stable way,

and/or provide extra electricity to backup rechargeable batteries if they are not sufficiently filled,

or, in case both said grid does not accept and said backup rechargeable batteries are sufficiently filled, said renewable electricity source ( 1 , 2 , 3 ) will be reconfigured so as to automatically fit actual need prediction but not more.

2. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 1 , wherein:

said process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), also includes:

a step of prediction of electricity price ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ).

3. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 1 , wherein:

said process of prediction of need of electricity ( 8 ) provided by said renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ), also includes:

a step of prediction of workload ( 8 ) in said computers cluster ( 9 ).

4. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 3 , wherein said step of prediction of workload ( 8 ) in said computers cluster ( 9 ) is partly based on historical data of said computers cluster ( 9 ).

5. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 1 , wherein said renewable electricity source ( 1 , 2 , 3 ) reconfiguration control is performed by real time sensors ( 5 ) integrated in said renewable electricity source ( 1 , 2 , 3 ).

6. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 1 , wherein:

said process of dynamical reconfiguration of said renewable electricity source ( 1 , 2 , 3 ) uses a smart grid components manager ( 4 ),

said process of scheduling tasks ( 7 ) in said computers cluster ( 9 ) uses a workload scheduler ( 7 ) holding a queue of pending jobs and mixing all predictions to schedule job executions in computers cluster ( 9 ) and to trigger said renewable electricity source ( 1 , 2 , 3 ) reconfiguration by said smart grid components manager ( 4 ) and preferably also maintenance operations on said renewable electricity source ( 1 , 2 , 3 ).

7. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 1 , wherein at least one of said prediction processes ( 6 , 8 ), preferably both said prediction processes ( 6 , 8 ), are based on using support vector machines.

8. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 1 , wherein at least one of said prediction processes ( 6 , 8 ), preferably both said prediction processes ( 6 , 8 ), are based either on using a supervised learning algorithm, preferably a nearest neighbor algorithm or a random forest algorithm, or on using a deep learning algorithm.

9. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 1 , wherein said process of scheduling tasks ( 7 ) is based on using a greedy algorithm, preferably a backfilling algorithm.

10. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 1 , wherein said process of scheduling tasks ( 7 ) is based on using an integer programming algorithm.

11. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 1 , wherein said process of scheduling tasks ( 7 ) takes into account an objective of increasing renewable electricity source ( 1 , 2 , 3 ) life expectancy and/or lowering renewable electricity source ( 1 , 2 , 3 ) maintenance frequency while, at the same time, maintaining high computers cluster ( 9 ) utilization.

12. Method of managing resources providing in a computers cluster ( 9 ) running jobs according to claim 1 , wherein said computers cluster ( 9 ) comprises more than 1000 terminal nodes, preferably more than 10000 terminal nodes.

13. Method of managing electricity providing in a computers cluster ( 9 ), comprising:

a process of prediction of need of electricity ( 8 ) provided by at least one renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 )

a process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), including:

a step of managing failure risk of said renewable electricity source ( 1 , 2 , 3 ), by lowering said predicted availability, so as to:

increase life expectancy of said renewable electricity source ( 1 , 2 , 3 ),

and/or lower maintenance frequency of said renewable electricity source ( 1 , 2 , 3 ),

a process of scheduling tasks ( 7 ) in said computers cluster ( 9 ), based on both said prediction processes ( 6 , 8 ),

a process of dynamical reconfiguration of said renewable electricity source ( 1 , 2 , 3 ), according both to last need prediction and to last availability prediction, said process of prediction of availability ( 6 ) then taking into account renewable electricity source reconfiguration update,

wherein:

in case of availability prediction decrease because of extreme meteorological events, said renewable electricity source ( 1 , 2 , 3 ) will be reconfigured so as:

either not running all the time,

and/or not running at full speed when running, and preferably wherein in case of availability prediction decrease because of extreme meteorological events, said renewable electricity source ( 1 , 2 , 3 ) will be reconfigured so as to automatically fit actual need prediction but not more.

14. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 13 , wherein:

said process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), also includes:

a step of prediction of electricity price ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ).

15. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 13 , wherein:

said process of prediction of need of electricity ( 8 ) provided by said renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ), also includes:

a step of prediction of workload ( 8 ) in said computers cluster ( 9 ).

16. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 13 , wherein:

said process of prediction of availability ( 6 ) also includes:

a step of evaluation of electricity providing at full speed by said renewable electricity source ( 1 , 2 , 3 ), based on weather forecasting data,

and preferably wherein said weather forecasting data come from a source external to said computers cluster ( 9 ), preferably from a web application programming interface.

17. Method of managing electricity providing in a computers cluster ( 9 ) according to claim 13 , wherein:

said process of dynamical reconfiguration of said renewable electricity source ( 1 , 2 , 3 ) uses a smart grid components manager ( 4 ),

said process of scheduling tasks ( 7 ) in said computers cluster ( 9 ) uses a workload scheduler ( 7 ) holding a queue of pending jobs and mixing all predictions to schedule job executions in computers cluster ( 9 ) and to trigger said renewable electricity source ( 1 , 2 , 3 ) reconfiguration by said smart grid components manager ( 4 ) and preferably also maintenance operations on said renewable electricity source ( 1 , 2 , 3 ).

18. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 13 , wherein said process of scheduling tasks ( 7 ) takes into account an objective of increasing renewable electricity source ( 1 , 2 , 3 ) life expectancy and/or lowering renewable electricity source ( 1 , 2 , 3 ) maintenance frequency while, at the same time, maintaining high computers cluster ( 9 ) utilization.

19. Method of managing resources providing in a computers cluster ( 9 ) running jobs according to claim 13 , wherein said computers cluster ( 9 ) comprises more than 1000 terminal nodes, preferably more than 10000 terminal nodes.

20. Method of managing electricity providing in a computers cluster ( 9 ), comprising:

a process of prediction of need of electricity ( 8 ) provided by at least one renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ),

a process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), including:

a step of managing failure risk of said renewable electricity source ( 1 , 2 , 3 ), by lowering said predicted availability, so as to:

increase life expectancy of said renewable electricity source ( 1 , 2 , 3 ),

and/or lower maintenance frequency of said renewable electricity source ( 1 , 2 , 3 ),

a process of scheduling tasks ( 7 ) in said computers cluster ( 9 ), based on both said prediction processes ( 6 , 8 ),

wherein it manages several renewable electricity sources ( 1 , 2 , 3 ), preferably different types of renewable electricity sources ( 1 , 2 , 3 ), said step of managing failure risk is performed for at least one, preferably several, more preferably most of, even more preferably all of, said renewable electricity sources ( 1 , 2 , 3 ),

wherein said renewable electricity sources ( 1 , 2 , 3 ) include one or more: wind turbines ( 2 ), solar panels ( 1 ), hydraulic dams ( 3 ), and

wherein said renewable electricity source ( 1 , 2 , 3 ) reconfiguration includes:

for wind turbine ( 2 ):

blade orientation modification,

neighbor wind turbine ( 2 ) synchronizing modification,

and/or for solar panel ( 1 ):

sensing surface orientation modification,

and/or for hydraulic dam ( 3 ):

dam gate opening modification.

21. Method of managing electricity providing in a computers cluster ( 9 ), comprising:

a process of prediction of need of electricity ( 8 ) provided by at least one renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ),

a process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), including:

a step of managing failure risk of said renewable electricity source ( 1 , 2 , 3 ), by lowering said predicted availability, so as to:

increase life expectancy of said renewable electricity source ( 1 , 2 , 3 ),

and/or lower maintenance frequency of said renewable electricity source ( 1 , 2 , 3 ),

a process of scheduling tasks ( 7 ) in said computers cluster ( 9 ), based on both said prediction processes ( 6 , 8 ),

wherein it manages several renewable electricity sources ( 1 , 2 , 3 ), preferably different types of renewable electricity sources ( 1 , 2 , 3 ), said step of managing failure risk is performed for at least one, preferably several, more preferably most of, even more preferably all of, said renewable electricity sources ( 1 , 2 , 3 ),

wherein said renewable electricity sources ( 1 , 2 , 3 ) include one or more: wind turbines ( 2 ), solar panels ( 1 ), hydraulic dams ( 3 ), and

wherein said renewable electricity source ( 1 , 2 , 3 ) reconfiguration in case of extreme meteorological events includes:

stopping wind turbine ( 2 ) in case of storm,

shielding solar panel ( 1 ) in case of hail,

closing dam ( 3 ) gate in case of overflow.

22. Method of managing electricity providing in a computers cluster ( 9 ), comprising:

a process of prediction of need of electricity ( 8 ) provided by at least one renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ),

a process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), including:

a step of managing failure risk of said renewable electricity source ( 1 , 2 , 3 ), by lowering said predicted availability, so as to:

increase life expectancy of said renewable electricity source ( 1 , 2 , 3 ),

and/or lower maintenance frequency of said renewable electricity source ( 1 , 2 , 3 ),

a process of scheduling tasks ( 7 ) in said computers cluster ( 9 ), based on both said prediction processes ( 6 , 8 ),

wherein said process of prediction of need of electricity ( 8 ) provided by said renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ), also includes a step of prediction of workload ( 8 ) in said computers cluster ( 9 ),

wherein in said step of prediction of electricity availability ( 6 ), electricity availability prediction is computed as a decreasing function of electricity price prediction,

wherein said step of prediction of electricity price ( 6 ) receives a feedback from said process of scheduling tasks ( 7 ), said step of prediction of electricity price ( 6 ) improves, based on said feedback, and

wherein said feedback, to said step of prediction of electricity price ( 6 ), is based on an evaluation of scheduling performances in said computers cluster ( 9 ) more than on an evaluation, if any, of electricity price prediction precision, preferably said feedback, to said step of prediction of electricity price ( 6 ), updates a cost function of said step of prediction of electricity price which is based on an evaluation of scheduling performances in said computers cluster ( 9 ) more than on an evaluation, if any, of electricity price prediction precision, more preferably said feedback, to said step of prediction of electricity price ( 6 ), is based on an evaluation of scheduling performances in said computers cluster ( 9 ), and not on an evaluation of electricity price prediction precision.

23. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 22 , wherein said process of scheduling tasks ( 7 ) takes into account an objective of increasing renewable electricity source ( 1 , 2 , 3 ) life expectancy and/or lowering renewable electricity source ( 1 , 2 , 3 ) maintenance frequency while, at the same time, maintaining high computers cluster ( 9 ) utilization.

24. Method of managing resources providing in a computers cluster ( 9 ) running jobs according to claim 22 , wherein said computers cluster ( 9 ) comprises more than 1000 terminal nodes, preferably more than 10000 terminal nodes.

25. Method of managing electricity providing in a computers cluster ( 9 ), comprising:

a process of prediction of need of electricity ( 8 ) provided by at least one renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ),

a process of prediction of availability of said electricity ( 6 ) provided by said renewable electricity source ( 1 , 2 , 3 ), including:

a step of managing failure risk of said renewable electricity source ( 1 , 2 , 3 ), by lowering said predicted availability, so as to:

increase life expectancy of said renewable electricity source ( 1 , 2 , 3 ),

and/or lower maintenance frequency of said renewable electricity source ( 1 , 2 , 3 ),

a process of scheduling tasks ( 7 ) in said computers cluster ( 9 ), based on both said prediction processes ( 6 , 8 ),

wherein said process of prediction of need of electricity ( 8 ) provided by said renewable electricity source ( 1 , 2 , 3 ) in said computers cluster ( 9 ), also includes a step of prediction of workload ( 8 ) in said computers cluster ( 9 ),

wherein said step of prediction of workload ( 8 ) receives a feedback from said process of scheduling tasks ( 7 ), and said step of prediction of workload ( 8 ) improves, based on said feedback, and

wherein said feedback, to said step of prediction of workload ( 8 ), is based on an evaluation of scheduling performances in said computers cluster ( 9 ) more than on an evaluation, if any, of workload prediction precision, preferably said feedback, to said step of prediction of workload ( 8 ), updates a cost function of said step of prediction of workload which is based on an evaluation of scheduling performances in said computers cluster ( 9 ) more than on an evaluation, if any, of workload prediction precision, more preferably said feedback, to said step of prediction of workload ( 8 ), is based on an evaluation of scheduling performances in said computers cluster ( 9 ), and not on an evaluation of workload prediction precision.

26. Method of managing electricity providing in a computers cluster ( 9 ), according to claim 25 , wherein said process of scheduling tasks ( 7 ) takes into account an objective of increasing renewable electricity source ( 1 , 2 , 3 ) life expectancy and/or lowering renewable electricity source ( 1 , 2 , 3 ) maintenance frequency while, at the same time, maintaining high computers cluster ( 9 ) utilization.

27. Method of managing resources providing in a computers cluster ( 9 ) running jobs according to claim 25 , wherein said computers cluster ( 9 ) comprises more than 1000 terminal nodes, preferably more than 10000 terminal nodes.