Dynamic autoscaling of server resources using intelligent demand analytic systems with determined baseline level of server resources

A method comprising:

• detecting, by a computer system, past computing usage of server resources of the computer system during an execution of one or more computing tasks over a previous period of time;
• determining, by the computer system using a machine learning (ML) engine comprising one or more ML models, a predicted future computing usage of the server resources at a first time during a future period of time based on the past computing usage, wherein the one or more ML models are trained based on demand metrics for a plurality of machine pools capable of processing the one or more computing tasks using the server resources of the computer system;
• determining, by the computer system, available server resources from the server resources during the future period of time, wherein the available server resources include at least a portion of the plurality of machine pools each having one or more machines operable to perform the one or more computing tasks;
• determining, by the computer system using the ML engine, a baseline level of the available server resources to allocate for the predicted future computing usage based on the past computing usage and one or more rule, wherein the baseline level comprises a threshold amount of resources that is allocated to prevent the available server resources from falling below a minimum capacity of the available server resources;
• determining, by the computer system, an allocation of at least one machine from the plurality of machine pools that is designated for handling of a computing task associated with the predicted future computing usage;
• configuring, by the computer system, a computing capacity of the available server resources allocated to one or more computing devices at a current time based on the determined allocation of the at least one machine and the baseline level;
• executing, by the computer system, the computing task using the at least one machine from at least one of the plurality of machine pools, wherein the executing the computing task comprises:
• dynamically sizing the plurality of machine pools to allocate the at least one machine to the computing task, and
• allocating a reserve machine pool from the plurality of machine pools for additional computing tasks;
• analyzing, by the computer system, a current computing usage of the server resources based on executing the computing tasks and a plurality of additional computing tasks executed using the available server resource at the current time;
• and
• retraining, by the computer system, the ML engine based on the current computing usage, wherein the one or more ML models of the ML engine comprise at least one continuous learning model, and wherein the retraining comprises:
• utilizing live data from the current computing usage to adjust the at least one continuous learning model based on changes in the past computing usage and the current computing usage over a selected period of time including the previous period of time and the current time,
• adjusting at least one of a weight or a node of the one or more ML models based on a recent change bias associated with the past computing usage and the current computing usage, and
• updating the baseline level based on the retraining for future resource allocation at one or more future periods of time.

A non-transitory machine-readable medium having stored thereon machine-readable instructions executable to cause a machine to perform operations comprising:

• analyzing, using a machine learning (ML) engine comprising one or more ML models, past computing usage of server resources of a server system during a performance of one or more computing tasks over a previous period of time, wherein the one or more ML models are trained based on demand metrics for a plurality of machine pools capable of processing the one or more computing tasks using the server resources of the server system;
• determining, based on the analyzing, a predicted future computing usage of the server resources at a first time during a future period of time;
• determining available server resources from the server resources during the future period of time, wherein the available server resources include at least a portion of the plurality of machine pools each having one or more machines operable to perform the one or more computing tasks;
• determining, using the ML engine, a baseline level of the available server resources to allocate for the predicted future computing usage based on the past computing usage and one or more rule, wherein the baseline level comprises a threshold amount of resources that is allocated to prevent the available server resources from falling below a minimum capacity of the available server resources;
• determining an allocation of at least one machine from the plurality of machine pools that is designated for handling of a computing task associated with the predicted future computing usage;
• configuring a computing capacity of the available server resources allocated to one or more computing devices at a current time based on the determined allocation of the at least one machine and the baseline level;
• executing the computing task using the at least one machine from at least one of the plurality of machine pools, wherein the executing the computing task comprises:
• dynamically sizing the plurality of machine pools to allocate the at least one machine to the computing task, and
• allocating a reserve machine pool from the plurality of machine pools for additional computing tasks;
• analyzing a current computing usage of the server resources based on executing the computing tasks and a plurality of additional computing tasks executed using the available server resource at the current time;
• and
• retraining the ML engine based on the current computing usage, wherein the one or more ML models of the ML engine comprise at least one continuous learning model, and wherein the retraining comprises:
• utilizing live data from the current computing usage to adjust the at least one continuous learning model based on changes in the past computing usage and the current computing usage over a selected period of time including the previous period of time and the current time,
• adjusting at least one of a weight or a node of the one or more ML models based on a recent change bias associated with the past computing usage and the current computing usage, and
• updating the baseline level based on the retraining for future resource allocation at one or more future periods of time.