System and method for proactively provisioning resources to an application
In one embodiment, a method includes executing at least one application. The method further includes collecting, at periodic intervals as the least one application executes, an application load of each of the at least one application. In addition, the method includes generating, based at least in part on the collecting, an application-behavior baseline for each of the at least one application. The application-behavior baseline includes a plurality of projected application loads over a future period of time. The application-behavior baseline has a configurable time resolution. The method also includes automatically determining, for a subperiod of the future period, a quantity of compute resources required to manage the at least one application in satisfaction of at least one performance criterion. Moreover, the method includes causing the determined quantity of compute resources to be proactively provisioned to the at least one application in advance of a start of the subperiod.
1. A method comprising:
executing at least one application on a computer system;
the computer system collecting, at periodic intervals as the at least one application executes, values of an application load of each of the at least one application;
retrieving, by the computer system, a prediction configuration previously associated with the application load of the at least one application, the prediction configuration comprising a selected period-combination data structure from among a plurality of alternative period combination data structures, wherein each period-combination data structure of the plurality of alternative period-combination data structures represents a same set of the collected values of the at least one application over the same historical period of time;
wherein, for each period-combination data structure of the plurality of alternative period combination data structures:
the period-combination data structure comprises a different selection of two or more time periods of a plurality of time periods;
each time period of the two or more time periods comprises segments corresponding to divisions of said same historical period of time;
the two or more time periods each span said same historical period of time at a different sampling rate, such that each of the two or more time periods includes a different number of segments; and
for each collected value of the collected values, the collected value has been incrementally inserted into corresponding segments of the two or more time periods, the incremental insertion comprising, for each time period of the two or more time periods, iterating from highest-frequency time period to lowest-frequency time period:
identifying a corresponding segment of the time period for the collected value;
computing a predicted value for the corresponding segment;
adapting the collected value based on any higher-frequency time periods of the period-combination data structure;
wherein, if the time period is the highest-frequency time period of the two or more time periods, the adapted collected value comprises the collected value;
wherein, if the time period is not the highest-frequency time period of the two or more time periods, the adapted collected value comprises a difference between the adapted collected value and the predicted value for the corresponding segment of the immediately higher-frequency time period; and
inserting the adapted collected value for the time period into the corresponding segment of the time period;
the computer system generating, using the adapted historical collected values of the selected period-combination data structure, an application-behavior baseline for each of the at least one application, the application-behavior baseline comprising a plurality of projected application loads over a future period of time;
the computer system automatically determining, for each subperiod of a plurality of subperiods of the future period, using the application-behavior baseline, a quantity of compute resources required to manage the at least one application in satisfaction of at least one performance criterion; and
for each subperiod of the plurality of subperiods, the computer system causing the determined quantity of compute resources to be proactively provisioned to the at least one application in advance of a start of the subperiod.
2. The method of claim 1 , wherein, for each subperiod of the plurality of subperiods:
the automatically determining comprises automatically determining a number of machine instances required to manage the at least one application over the subperiod; and
the causing comprises causing the determined number of machine instances to be proactively provisioned to the at least one application in advance of the start of the subperiod.
3. The method of claim 2 , wherein the machine instances comprise virtual machine instances.
4. The method of claim 1 , wherein the plurality of subperiods cover an entirety of the future period.
5. The method of claim 1 , wherein the plurality of projected application loads are measured using one or more application-performance metrics selected from the group consisting of: memory utilization, processor utilization, operation completion rate, storage utilization, network bandwidth, and end-user experience management metrics.
6. The method of claim 1 , comprising:
prior to the causing, sending a notification to an authorized user, the notification comprising the determined quantity of compute resources;
receiving authorization from the authorized user; and
wherein the causing is performed responsive to the received authorization.
7. The method of claim 1 , wherein the at least one application comprises a plurality of applications.
8. The method of claim 7 , comprising performing the method for each of the at least one application.
9. The method of claim 1 , comprising:
based at least in part on the collecting, identifying at least one statistical deviation from the application-behavior baseline;
responsive to the identifying, reactively determining a modified quantity of compute resources; and
causing the modified quantity of compute resources to be provisioned to the at least one application.
10. The method of claim 1 , wherein the causing comprises causing the at least one application to be at least of one scaled up and scaled out.
11. The method of claim 1 , wherein the causing comprises causing the at least one application to be at least one of scaled down and scaled in.
12. The method of claim 1 , wherein:
the collecting comprises collecting, at periodic intervals as the at least one application executes, an application load of at least one subunit of the at least one application;
the generating comprises generating the application-behavior baseline for the at least one subunit;
the automatically determining comprises automatically determining, for each subperiod of the plurality of subperiods, a quantity of compute resources required to manage the at least one subunit in satisfaction of at least one performance criterion; and
the causing comprises, for each subperiod of the plurality of subperiods, causing the determined quantity of compute resources to be proactively provisioned to the at least one subunit in advance of the start of the subperiod.
13. The method of claim 12 , wherein the at least one subunit comprises a role instance of the at least one application.
14. The method of claim 12 , comprising:
wherein the at least one subunit comprises a plurality of subunits of the at least one application; and
performing the method for each subunit of the plurality of subunits.
15. The method of claim 1 , wherein the automatically determining comprises:
testing a correlation between a user-centric metric and a resource-utilization metric; and
responsive to a determination that the correlation between the user-centric metric and the resource-utilization metric is sufficient, computing the quantity of compute resources based, at least in part, on the user-centric metric.
16. The method of claim 15 , wherein the automatically determining comprises:
responsive to a determination that the correlation between the user-centric metric and the resource-utilization metric is not sufficient, computing the quantity of compute resources independently of the user-centric metric.
17. An information handling system comprising at least one server computer and memory, wherein the at least one server computer and memory in combination are operable to implement a method comprising:
executing at least one application;
collecting, at periodic intervals as the at least one application executes, values of an application load of each of the at least one application;
retrieving a prediction configuration previously associated with the application load of the at least one application, the prediction configuration comprising a selected period-combination data structure from among a plurality of alternative period-combination data structures, wherein each period-combination data structure of the plurality of alternative period-combination data structures represents a same set of the collected values of the at least one application over the same historical period of time;
wherein, for each period-combination data structure of the plurality of alternative period-combination data structures:
the period-combination data structure comprises a different selection of two or more time periods of a plurality of time periods;
each time period of the two or more time periods comprises segments corresponding to divisions of said same historical period of time;
the two or more time periods each span said same historical period of time at a different sampling rate, such that each of the two or more time periods includes a different number of segments; and
for each collected value of the collected values, the collected value has been incrementally inserted into corresponding segments of the two or more time periods, the incremental insertion comprising, for each time period of the two or more time periods, iterating from highest-frequency time period to lowest-frequency time period:
identifying a corresponding segment of the time period for the collected value;
computing a predicted value for the corresponding segment;
adapting the collected value based on any higher-frequency time periods of the period-combination data structure;
wherein, if the time period is the highest-frequency time period of the two or more time periods, the adapted collected value comprises the collected value;
wherein, if the time period is not the highest-frequency time period of the two or more time periods, the adapted collected value comprises a difference between the adapted collected value and the predicted value for the corresponding segment of the immediately higher-frequency time period; and
inserting the adapted collected value for the time period into the corresponding segment of the time period;
generating, using the adapted historical values of the selected period-combination data structure, an application-behavior baseline for each of the at least one application, the application-behavior baseline comprising a plurality of projected application loads over a future period of time;
automatically determining, for each subperiod of a plurality of subperiods of the future period, using the application-behavior baseline, a quantity of compute resources required to manage the at least one application in satisfaction of at least one performance criterion; and
for each subperiod of the plurality of subperiods, causing the determined quantity of compute resources to be proactively provisioned to the at least one application in advance of a start of the subperiod.
18. The information handling system of claim 17 , wherein, for each subperiod of the plurality of subperiods:
the automatically determining comprises automatically determining a number of machine instances required to manage the at least one application over the subperiod; and
the causing comprises causing the determined number of machine instances to be proactively provisioned to the at least one application in advance of the start of the subperiod.
19. A computer-program product comprising a non-transitory computer-usable medium having computer-readable program code embodied therein, the computer-readable program code adapted to be executed to implement a method comprising:
executing at least one application;
collecting, at periodic intervals as the at least one application executes, values of an application load of each of the at least one application;
retrieving a prediction configuration previously associated with the application load of the at least one application, the prediction configuration comprising a selected period-combination
data structure from among a plurality of alternative period-combination data structures, wherein each period-combination data structure of the plurality of alternative period-combination
data structures represents a same set of the collected values of the at least one application over the same historical period of time;
wherein, for each period-combination data structure of the plurality of alternative period-combination data structures:
the period-combination data structure comprises a different selection of two or more time periods of a plurality of time periods;
each time period of the two or more time periods comprises segments corresponding to divisions of said same historical period of time;
the two or more time periods each span said same historical period of time at a different sampling rate, such that each of the two or more time periods includes a different number of segments; and
for each collected value of the collected values, the collected value has been incrementally inserted into corresponding segments of the two or more time periods, the incremental insertion comprising, for each time period of the two or more time periods, iterating from highest-frequency time period to lowest-frequency time period:
identifying a corresponding segment of the time period for the collected value;
computing a predicted value for the corresponding segment;
adapting the collected value based on any higher-frequency time periods of the period-combination data structure;
wherein, if the time period is the highest-frequency time period of the two or more time periods, the adapted collected value comprises the collected value;
wherein, if the time period is not the highest-frequency time period of the two or more time periods, the adapted collected value comprises a difference between the adapted collected value and the predicted value for the corresponding segment of the immediately higher-frequency time period; and
inserting the adapted collected value for the time period into the corresponding segment of the time period;
generating, using the adapted historical values of the selected period-combination data structure, an application-behavior baseline for each of the at least one application, the application-behavior baseline comprising a plurality of projected application loads over a future period of time;
automatically determining, for each subperiod of a plurality of subperiods of the future period, using the application-behavior baseline, a quantity of compute resources required to manage the at least one application in satisfaction of at least one performance criterion; and
for each subperiod of the plurality of subperiods, causing the determined quantity of compute resources to be proactively provisioned to the at least one application in advance of a start of the subperiod.