Systems and methods for tuning an operating system, application, or network component
A method, system, apparatus, and media are directed to tuning a device. Values are stored for each of a plurality of device parameters based on device metrics. The device parameters are constraint-matched over changes to settings of the device parameters to identify at least one point within each range of the device parameters that optimizes a performance measure for the device. Tuning values are selected for each device parameter based on the identified point. The tuning values are provided for tuning the device. The device is tuned with roll-back based on the running values.
1. A processor-implemented method for tuning a device, comprising:
storing, on a computer memory, values for each of a plurality of device parameters based on device metrics;
constraint-searching, by a processor, over changes to settings of the device parameters to identify at least one point within each range of the device parameters that optimizes an improvement function of a performance measure for the device, wherein constraint-searching comprises: (a) modifying, on the device, each said device parameters within a subset of the device parameters; and (b) recursively performing the steps of modifying and calculating the performance measure until the improvement function of the performance measure is determined to reach an optimal point and after the optimal point stabilizes during the recursive constraint-searching;
selecting tuning values for each said device parameter based on the identified point;
providing the tuning values for tuning the device; and
tuning the device based on the provided tuning values;
wherein the constraint-searching further comprises determining a relationship between first device parameter and a second device parameter based on whether changing the first and second device parameters relative to each other changes the calculated performance measure.
2. The method of claim 1 , wherein the optimal point comprises a local minimum or maximum, or a global minimum or maximum for a number of iterations of the recursive performance of the step of modifying and calculating.
3. The method of claim 1 , further comprising:
in response to determining whether the relationship, further constraint-searching based on the determined relationship, wherein the device parameters within the subset are related to each other by the determined relationship of related or unknown.
4. The method of claim 1 , wherein modifying, on the device, each said device parameters within the subset of the device parameters comprises:
changing the device parameters that are within the subset, wherein these device parameters are related to each other or are unknown to be related to each other, while keeping a device parameter outside the subset constant.
5. The method of claim 1 , wherein modifying, on the device, each said device parameters within the subset of the device parameters comprises performing on the device parameters within the subset at least one of:
a binary search;
a random walk;
a simulated annealing;
a gradient descent;
a change based on a schedule; or
a min-max search.
6. The method of claim 1 , wherein the plurality of device parameters comprises a different device parameter for a plurality of types for each of a plurality of services, wherein the plurality of types is representative of device configurations for performing a system operation or networking within a plurality of Open Systems Interconnection (OSI) layers, and wherein the services comprise a use of an application, a use of a type of application, a type of user, or system usage.
7. The method of claim 1 , further comprising:
filtering the metric based on a service related to the device parameter before storing the values for each of the plurality of device parameters.
8. The method of claim 1 , further comprising:
saving a state of the device;
tuning the device based on the selected tuning values; and
if a self-test on the device fails for at least one of the tuning values, rolling back to the saved state.
9. A non-transitory system for tuning over a network, comprising: a server computer including memory configured to perform actions comprising:
storing tuning values for each of a plurality of device parameters and for each service of the device based on device metrics;
constraint-searching using a heuristic over changes to settings of the device parameters to identify a point within each range of the device parameters, wherein the constraint-searching calculates a performance measure for the device and the identified point optimizes the performance measure for the device; and
providing the tuning values for each said device parameter based on the identified point; and
another device configured to perform actions comprising:
self-tuning based on the provided tuning values; and
rolling-back to a saved state for at least one of the tuning values if a self-test for the at least one tuning values fails;
wherein the constraint-searching further comprises determining a relationship between a first device parameter and a second device parameter based on whether changing the first and second device parameters relative to each other changes the calculated performance measure.
10. The system of claim 9 , wherein constraint-searching comprises:
performing an intersection of points for each of the ranges of the device parameters and for each said service.
11. The system of claim 9 , wherein constraint-searching comprises:
performing a statistical analysis of changes in the device parameters.
12. The system of claim 9 , wherein the actions of the other device further comprises:
generating information usable to determine a metric of the other device; and
sending the metric to the server for generating the points within the ranges.
13. The system of claim 9 , further comprising:
an administrator interface configured to perform actions comprising:
receiving a tuning policy usable in constraining the provided tuning values to satisfy the tuning policy.