Granted Patent B2

US 12,510,943 · App. 18/425,536 · Granted Dec 30, 2025

Inventor: Kanad Ghose (Vestal, NY)

Assignee: The Research Foundation for The State University of New York

G06F1/206A61K36/185B03B1/00B03B5/02B03B5/58B03B11/00G05D23/19G06F1/20G06F1/3203G06F1/3206G06F1/3209G06F1/3228G06F9/45558G06F9/4893G06F9/5094G06F13/409G06K19/0723G06K19/07705G06K19/07722H04L69/329H05K7/20836G06F2009/4557Y02D10/00

Abstract

Claims (44)

1 . A system for managing a datacenter comprising a plurality of racks, each rack having a plurality of servers, each server supporting execution of virtual machines, comprising:

a set of stored tables, which record at least:

rack energy versus performance characteristics,

rack utilization statistics, and

rack environmental temperature;

an ordered queue of tasks to be processed by the virtual machines;

a computer implemented model configured to predict an impact of a new task in the ordered queue of tasks on the rack energy versus performance characteristics, rack utilization statistics, and rack environmental temperature, based on the set of stored tables;

a scheduler, configured to optimize a cost of operation of the plurality of racks and at least one cooling system for the plurality of racks while meeting a task performance constraint for the respective tasks and a thermal constraint, by controlling:

a sequence of, and the server employed by, the respective tasks in the ordered queue of tasks to meet the task performance constraint, and

the at least one cooling system for the plurality of racks in dependence on the sequence of tasks in the ordered queue of tasks prior to their execution, to meet predicted cooling needs to meet the thermal constraint; and

at least one virtual machine, configured to perform tasks dependent on the ordered queue of tasks.

2 . The system according to claim 1 , wherein the optimization of cost comprises energy costs, datacenter maintenance costs, datacenter operating costs, and task processing latency.

3 . The system according to claim 1 , wherein the scheduler is further configured to control an activation state of respective racks of the plurality of racks, to optimize the cost of operation of the plurality of racks and the at least one cooling system for the plurality of racks while meeting the task performance constraint for the respective tasks and the thermal constraint.

4 . The system according to claim 3 , wherein the scheduler is further configured to maximize a workload of tasks allocated to respective racks which are controlled to be active.

5 . The system according to claim 1 , further comprising:

a memory which stores energy requirement characteristics of each respective server for performing the respective tasks, an energy requirement characteristics of the at least one cooling system for cooling the servers for performing the respective tasks, and task completion rates of respective tasks; and

an input port configured to receive energy cost information,

wherein the scheduler is configured to employ the energy cost information, energy requirement characteristics of the respective servers, energy requirement characteristics of the at least one cooling system, energy cost information, and the task completion rates of the respective tasks to optimize the cost of operation of the plurality of racks and the at least one cooling system by selectively targeting execution of respective tasks on respective servers dependent on at least the energy requirement characteristics of the respective tasks and the task completion rates of the respective tasks.

6 . The system according to claim 1 , wherein the scheduler is further configured to produce an output to control a migration of a virtual machine from a first server to a second server of the plurality of racks, wherein the optimization of the cost of operation of the plurality of racks and the at least one cooling system for the plurality of racks is dependent on at least a cost of energy expended in the migration of the virtual machine.

7 . The system according to claim 1 , wherein the scheduler is further configured to change an activation state of at least one respective rack of the plurality of racks and to control a migration of a virtual machine from a first server to a second server of the plurality of racks, wherein the optimization of the cost of operation of the plurality of racks and the at least one cooling system for the plurality of racks is dependent on a cost of energy expended in the change of the activation state of the at least one respective rack and the migration of the virtual machine.

8 . The system according to claim 1 , wherein the scheduler is further configured to redirect cooling from the at least one cooling system from a first rack to a second rack of the plurality of racks dependent on cooling needs predicted from the ordered queue of tasks.

9 . The system according to claim 1 , wherein the scheduler is further configured to optimize the cost of operation by selecting between at least a first mode in which workload is concentrated on a subset of the plurality of racks, and a second mode in which workload is balanced across the plurality of racks.

10 . The system according to claim 1 , wherein the scheduler is responsive to a latency of the at least one cooling system between generation of a cooling system control signal and a cooling response.

11 . The system according to claim 1 , wherein the scheduler comprises a hierarchical set of independent control elements at a plurality of levels, wherein energy versus performance goals of a respective level below a highest level are explicitly dictated by an independent control element at a hierarchically higher level.

12 . The system according to claim 1 , wherein the scheduler is configured to selectively change an activation state of respective racks of the plurality of racks and an operational parameter of the at least one cooling system at predetermined update intervals.

13 . The system according to claim 1 , wherein each task in the ordered queue of tasks is labeled as being compute bound or input-output bound, and wherein the scheduler is further configured to:

selectively group compute bound tasks for execution on at least one server optimized for compute bound tasks;

selectively group input-output bound tasks for execution on at least one server optimized for input-output bound tasks;

the at least one server optimized for compute bound tasks having different characteristics from the at least one server optimized for input-output bound tasks; and

selectively produce a signal for optimization of a performance of a respective server for processing of compute bound tasks or input-output bound tasks by changing the characteristics of the respective server.

14 . The system according to claim 1 , wherein the scheduler is further configured to schedule operation of the cooling systems dependent on the ordered queue of tasks.

15 . The system according to claim 1 , wherein the scheduler comprises a hierarchical scheduler comprising a higher level scheduler and a lower level scheduler, each respective server has an energy budget allocation defined by the higher level scheduler, and the lower level scheduler adjusts performance settings of the respective server and performs local scheduling for the respective server to stay within the respective server's energy budget allocation defined by the higher level scheduler.

16 . The system according to claim 1 , wherein the scheduler is configured to separately control at least:

a cooling system an air flow rate;

an inlet air temperature; and

at least one moveable baffle configured to redirect air flow.

17 . A system for managing a datacenter comprising a plurality of racks, each rack having a plurality of servers, comprising:

at least one cooling system, configured to cool at least one rack;

a set of stored tables, which record actual rack energy versus performance characteristics, rack utilization statistics, and rack environmental temperature, wherein the respective servers of the plurality of servers have controllable energy versus performance characteristics;

an ordered queue of tasks to be processed by the plurality of servers;

a computer implemented model configured to predict an impact of execution of a new task in the ordered queue of tasks on the rack energy versus performance characteristics, rack utilization statistics, and rack environmental temperature, based on the set of stored tables; and

a scheduler, configured to optimize an energy cost for processing the plurality of racks while meeting a task performance constraint and the operating at least one cooling system for cooling the plurality of racks to meet a thermal constraint, by controlling:

a sequence of, and a respective server employed for execution of the tasks in the ordered queue of tasks to meet the task performance service constraint, and

the at least one cooling system for the plurality of racks in dependence on the sequence of tasks in the ordered queue of tasks prior to their execution, to meet predicted cooling needs to meet the thermal constraint.

Continuity (5)

References Cited (400)