Provisioning service requests in a computer system

Patent metadata

Field	Value
Publication number	US-10217053-B2
Application number	US-201514747062-A
Country	US
Kind code	B2
Filing date	Jun 23, 2015
Priority date	Jun 23, 2015
Publication date	Feb 26, 2019
Grant date	Feb 26, 2019

Abstract

Disclosed is a system, computer program product, and method for provisioning a new service request. The computer-implemented method begins with receiving a new service request for computational resources in a computing system. The required computational resources are memory usage, storage usage, processor usage, or a combination thereof to fulfill the new service request. Next a sandbox computing environment is used to operate the new service request. The sandbox computing environment is used to isolate the computing system. The sandbox computing environment produces a current computational resources usage data to fulfill the new service request in the sandbox computing environment. The current sandbox computational resources usage data and historical computational resources usage data are both used by a machine learning module to create a prediction of the computational resources that will be required in the computing system to fulfill the new service request.

First claim

What is claimed is:

1. A computer-implemented method comprising:
receiving at least one new service request for computational resources in a computing system, wherein required computational resources include memory usage, storage usage, processor usage, or a combination thereof to fulfill the new service request;
operating the new service request in a sandbox computing environment, the sandbox computing environment is used to execute a new service requests without harming the computing system by restricting access to the computer system and network, the sandbox computing environment performing pre-execution testing by two or more approximate versions of services models for the new service request, where each service model uses different data sizes, and executing the new service request on each of the two or more approximate versions of services models in the sandbox computing environment to record predicted computation resources usage data including predicted execution time required, predicted memory space required, and predicted number of processors required;
machine learning module utilizing both the predicted computational resources usage data and historical computational resources usage data to create a prediction of the computational resources that will be required in the computing system to fulfill the new service request;
based on the predicted computational resources required being above a definable quality of service threshold, performing at least one of
rejecting the new service request,
providing the predicted computational resources based on each of the service models to a requestor of the new service, and
transforming the new service request into sub-requests, where each sub-request is limited to a sub-set of computational resources, or
a combination thereof; and

based on the predicted computational resources required being below the definable quality of service threshold, deploying in the computing system which is separate from the sandbox computing environment, the computational resources which have been predicted based upon the new service request subject to pricing policies and resource constraints, and recording actual computation resources usage data as historical computation resources usage data including actual execution time required, actual memory space required, and actual number of processors required.

2. The method of claim 1 , further comprising:
analyzing the new service request to identify an amount computational resources, as a function of size of data, required to fulfill the new service request; and
based on the prediction from the machine learning module and in response to the amount being greater than a definable threshold,
(a) limiting the new service request to at least a sub-set of the computational resources in the computing system;
(b) internally dividing the new service requests into at least two or more new individual service sub-requests to implement an equivalent service request; or
a combination of a and b; and

based on the prediction from the machine learning module and in response to the amount not exceeding a definable threshold, executing the new service request in the computing system.

3. The method of claim 2 , further comprising:
executing the new service request; and
determining an excess amount of computational resources being consumed for the new service request; and based on the excess amount of computational resource consumed to limit the computational resources made available for the new service request.

4. The method of claim 1 , further comprising:
sending an estimate for the new service request or a rejection for the new service request to a requestor of the new service request; and
receiving authorization from the requestor whether to continue executing the new service requests or stop executing the new service request.

5. The method of claim 1 , further comprising:
using the prediction of the computational resources required in the computing system to implement the new service request to set an estimated price for the new service.

6. The method of claim 1 , further comprising:
fulfilling the new service request by
analyzing the new service request to determine if the new service request is divisible into a set of two or more new individual service sub-requests to implement an equivalent service request; and
in response to the new service request being divisible, internally dividing the new service requests into at least two or more new individual service sub-requests to implement the equivalent service request.

7. The method of claim 6 , further comprising:
using the two or more new individual service sub-requests to set an estimated price for the new service request.

8. The method of claim 7 , further comprising:
placing the new service request into a user class depending on the prediction of the computational resources that will be required in the computing system to fulfill the new service request; and
using the user class to limit the computational resources made available for the new service.

9. The method of claim 6 , further comprising:
sending a result of an individual service sub-request to a requestor of the new service; and
receiving authorization from the requestor whether to continue executing the sub-requests or stop executing the new service request.

10. The method of claim 6 , wherein the new service request includes a query and the set of two or more new individual service sub-requests are partial queries.

11. The method of claim 1 , further comprising:
implementing the new service request by
analyzing the new service request to determine if the new service request is divisible into a set of two or more new individual service sub-requests to implement the new service request with lower accuracy; and
in response to the new service request being divisible, breaking the new service request into at least two or more new individual service sub-requests to implement the new service request with lower accuracy.

12. The method of claim 11 , further comprising:
using the two or more new individual service sub-requests to set a price for the new service.

13. The method of claim 1 , wherein the new service request is transferred to at least one lower class with fewer resources after excess resource consumption is detected.

14. The method of claim 1 , wherein an amount of overall resources allowed for the new service request is adjusted, once or more, to a percentage of its original resource estimate once an excess consumption is detected.

15. A system for managing new service requests, the system comprising:
a memory;
a processor communicatively coupled to the memory, where the processor is configured to perform
receiving at least one new service request for computational resources in a computing system, wherein required computational resources include memory usage, storage usage, processor usage, or a combination thereof to fulfill the new service request;
operating the new service request in a sandbox computing environment, the sandbox computing environment is used to execute a new service requests without harming the computing system by restricting access to the computer system and network, the sandbox computing environment performing pre-execution testing by two or more approximate versions of services models for the new service request, where each service model uses different data sizes, and executing the new service request on each of the two or more approximate versions of services models in the sandbox computing environment to record predicted computation resources usage data includ

Assignees

• IBM

Inventors

• Bordawekar Rajesh
• Kundu Ashish
• Shmueli Oded

Classifications

• G06F9/5027

  the resource being a machine, e.g. CPUs, Servers, Terminals · CPC title

• G06N7/01 Primary

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06F9/5011

  the resources being hardware resources other than CPUs, Servers and Terminals · CPC title

• G06N5/04

  Inference or reasoning models · CPC title

• G06N7/005 Primary

  Physics · mapped topic