Discovering optimal system configurations using decentralized probability based active sampling

1. A method for optimizing system performance, comprising:

applying sampling based optimization to identify optimal configurations of a computing system by:

selecting a number of configuration samples;

evaluating system performance based on the samples;

based on system performance feedback of evaluated samples, inferring a location of an optimal configuration; and

generating additional samples towards the location of the inferred optimal configuration to further optimize a system configuration;

wherein said selecting a number of configuration samples includes implementing a decentralized probability based active sampling (DPAS) method to employ a plurality of configuration parameters, said DPAS including a sampling strategy which utilizes experience from past samples; and the method further comprises building a probability for generating new samples based upon historic data from the past samples.

2. The method as recited in claim 1 , wherein selecting a number of configuration samples includes considering global coverage samples and local structure to determine the number of samples in a configuration space.

3. The method as recited in claim 1 , wherein building a probability includes dynamically updating each evaluated sample based on a reward-penalty strategy so that the reward-penalty strategy leads a search process towards an optimal setting.

4. The method as recited in claim 1 , further comprising decomposing the sampling into a set of separate modules, where each module is responsible for a search of a specific configuration parameter.

5. The method as recited in claim 1 , wherein evaluating system performance includes incorporating multiple performance dimensions simultaneously to permit a definition of acceptable system performance.

6. The method as recited in claim 1 , wherein based on feedback of evaluated samples, inferring a location of an optimal configuration includes:

dividing a range of a parameter into a number of bins; and

determining a best observed performance evaluation from test samples originated from that bin.

7. The method as recited in claim 6 , further comprising constructing a probability distribution over the bins based on best history performances for the bins where if one bin has encountered a higher evaluation score than other bins, that bin gets a high probability for generating new samples.

8. The method as recited in claim 7 , further comprising updating the probability distribution every time a new sample has been evaluated using a reward-penalty based method.

9. The method as recited in claim 8 , further comprising performing a local search based on a fitted function to obtain a solution.