Automatic prediction system for server failure and method of automatically predicting server failure

1. An automatic prediction system for a server failure comprising:

a data collection module ( 100 ) to collect status information of a server and operating service of a target system;

a model generation and optimization module ( 200 ) to generate a CNN-based failure prediction model by using the collected data and to optimize model parameters and hyper-parameter values; and

a prediction module ( 300 ) to perform online failure prediction by using the optimized CNN-based failure prediction model,

wherein the data collection module ( 100 ) comprises:

a monitoring agent ( 110 ) to periodically collect historical data indicating status of the server and operating service of the target system by detection plug-in;

a failure detector ( 120 ) to detect failure data when a log error of the server of the target system occurs; and

a data explorer ( 130 ) to check and analyze the collected data,

wherein the historical data are in the key-value form representing a data name and numerical value, the system sources to be collected as data are main memory, auxiliary memory, network and CPU and represent each source's source information, speed and utilization.

2. The automatic prediction system for a server failure in claim 1 , wherein the model generation and optimization module ( 200 ) comprises:

a configuration interpreter ( 210 ) to interpret model configuration information and transfer the interpreted model configuration information to an optimizer ( 220 );

the optimizer ( 220 ) to generate the prediction model based on the interpreted model configuration information, to optimize the model by applying machine learning and hyper-parameter tuning and to evaluate failure prediction performance; and

a predictor repository ( 230 ) to store the optimized model.

3. The automatic prediction system for a server failure in claim 1 , wherein the prediction module ( 300 ) comprises:

a predictor loader ( 310 ) to generate online predictors ( 320 ) by calling the relevant model from the predictor repository ( 230 ) when receiving from the optimizer ( 220 ) a notice that a new prediction model is updated; and

a number of online predictors ( 320 ) to predict a failure by receiving the information of the currently operating system and to warn a user when the probability of a failure occurrence exceeds a predetermined limit.

4. A method for automatically predicting server failure comprising the steps of:

collecting historical data which include training data, validation data and test data and indicate the status of sources and operating services of a target system, through a monitoring agent ( 110 ) of a data collection module ( 100 );

generating a CNN-based failure prediction model and optimizing the model through an optimizer ( 220 ), based on model configuration values; and

predicting a server failure of the target system by using the generated CNN-based failure prediction model and storing results,

wherein the step of generating and optimizing the failure prediction model further comprises the steps of:

generating the CNN-based failure prediction model based on prediction model configuration values;

learning parameters of the CNN-based failure prediction model by using the training data;

tuning hyper-parameters by using the validation data; and

evaluating the CNN-based failure prediction model by using the test data,

wherein the historical data are in the key-value form representing a data name and numerical value, the system sources to be collected as data are main memory, auxiliary memory, network and CPU and represent each source's source information, speed and utilization.