Techniques for self-tuning of computing systems
A self-tuning computing system and a method for self-tuning a computing system. The method includes measuring a current operation metric representing a current performance of the computing system; determining, based on the current operation metric and a target metric, at least one optimization scheme for improving the current operation metric, wherein the at least one optimization scheme includes at least a plurality of system knobs each having a respective optimal value; and setting each of the system knobs listed in the at least one determined optimization scheme to its respective optimal value.
1. A method comprising:
mapping a current workload of a computing system to a first bucket of related workloads;
wherein the related workloads mapped to the first bucket behave similarly with respect to a target metric, and the mapping is performed using a locality based hashing function applied to a plurality of workload-related values;
measuring a current operation metric representing a current performance of the computing system;
determining, by a processor, based on the current operation metric and the target metric, a first scheme for the first bucket for improving the current operation metric, wherein the first scheme includes a plurality of system configuration parameters each having a respective value, wherein the first scheme is determined based further on a current execution phase of an application currently executed by the computing system; and
setting each of the system configuration parameters listed in the first scheme to its respective value during the current execution phase.
2. The method of claim 1 , wherein the first scheme is determined using machine learning.
3. The method of claim 1 , further comprising:
predicting a future workload based on a current workload of the computing system, wherein the first scheme is determined based further on the predicted future workload.
4. The method of claim 1 , further comprising:
testing a configuration for each of the plurality of system configuration parameters; and
selecting the respective value for each of the plurality of system configuration parameters based on the testing of the configuration.
5. The method of claim 4 , wherein testing the configuration for each of the plurality of system configuration parameters further comprises:
iteratively setting each of a subset of the plurality of system configuration parameters to a temporary value.
6. The method of claim 5 , further comprising:
continuously monitoring an operation of the computing system; and
dynamically tuning the plurality of system configuration parameters based on changes in the operation of the computing system.
7. The method of claim 5 , wherein the first scheme is determined using reinforcement learning.
8. The method of claim 1 , wherein the first scheme is determined based further on a cost of changing values of the plurality of configuration parameters according to the first scheme and an estimated benefit of applying the first scheme.
9. The method of claim 1 , wherein the current operation metric is of a current workload of the computing system, wherein the current workload of the computing system includes at least one application.
10. The method of claim 1 , wherein measuring the current operation metric further comprises:
running the computing system, wherein the current operation metric is measured when the computing system has run for a predetermined time interval.
11. The method of claim 1 , the computing system including a plurality of components, wherein setting each of the configuration parameters listed in the first scheme to its respective value causes reconfiguration of the plurality of components.
12. A non-transitory computer readable medium having stored thereon instructions for causing a processing circuitry to execute a process for self-tuning of a computing system, the process comprising:
mapping a current workload of a computing system to a first bucket of related workloads;
wherein the related workloads mapped to the first bucket behave similarly with respect to a target metric, and the mapping is performed using a locality based hashing function applied to a plurality of workload-related values;
measuring a current operation metric representing a current performance of the computing system;
determining, by a processor, based on the current operation metric and the target metric, a first scheme for the first bucket for improving the current operation metric, wherein the first scheme includes a plurality of system configuration parameters each having a respective value, wherein the first scheme is determined based further on a current execution phase of an application currently executed by the computing system; and
setting each of the system configuration parameters listed in the first scheme to its respective value during the current execution phase.
13. A self-tuning computing system, comprising:
a processing circuitry; and
a memory, the memory containing instructions that, when executed by the processing circuitry, configure the computing system to:
map a current workload of a computing system to a first bucket of related workloads;
wherein the related workloads mapped to the first bucket behave similarly with respect to a target metric, and the mapping is performed using a locality based hashing function applied to a plurality of workload-related values;
measure a current operation metric representing a current performance of the computing system;
determine, based on the current operation metric and the target metric, a first scheme for the first bucket for improving the current operation metric, wherein the first scheme includes a plurality of system configuration parameters each having a respective value, wherein the first scheme is determined based further on a current execution phase of an application currently executed by the computing system; and
set each of the system configuration parameters listed in the first scheme to its respective value during the current execution phase.
14. The self-tuning computing system of claim 13 , wherein the computing system is further configured to:
predict a future workload based on a current workload of the computing system, wherein the first scheme is determined based further on the predicted future workload.
15. The self-tuning computing system of claim 13 , wherein the computing system is further configured to:
test a configuration for each of the plurality of system configuration parameters; and
select the respective value for each of the plurality of system configuration parameters based on the testing of the configuration.
16. The self-tuning computing system of claim 15 , wherein the first scheme is determined using reinforcement learning.
17. The self-tuning computing system of claim 13 , wherein the first scheme is determined based further on a cost of changing values of the plurality of configuration parameters according to the first scheme and an estimated benefit of applying the first scheme.