Abstract: A health of a server infrastructure is supervised by sending measurement requests, from a polling node to a server of the server infrastructure, at each of successive polling periods. Measurements received from the server in response to the measurement requests are stored in a database. A machine learning system is trained using accumulated measurements stored in the database to construct a prediction model for the server. Compliance of a latest measurement received from the server with the prediction model for the server is evaluated. A corrective action is taken if the latest measurement does not comply with the prediction model for the server. Measurements received from a group of servers may be aggregated and the machine learning system may construct a prediction model for the group of servers. Compliance to the prediction model may be evaluated for the group of servers.

Abstract: A method for formatting raw data comprises accessing the raw data, the raw data comprising sparse data segments, which are empty of any data, and non-sparse data segments, which comprise data, and generating a formatted data stream comprising one or more atomic blocks, each atomic block corresponding to a metadata file and to a portion of the non-sparse data segments of the raw data. Generating one atomic block comprises browsing the raw data and, upon locating one sparse data segment, populating the corresponding metadata file with offsets indicative of a beginning and an end of the located sparse data segment and populating the atomic block with a concatenation of at least portions of the non-sparse segments of the raw data located before and after the located sparse segment. If the atomic block exceeds a maximum size, another atomic block and another corresponding metadata file are populated.

Abstract: A health of a server infrastructure is supervised by sending measurement requests, from a polling node to a server of the server infrastructure, at each of successive polling periods. Measurements received from the server in response to the measurement requests are stored in a database. A machine learning system is trained using accumulated measurements stored in the database to construct a prediction model for the server. Compliance of a latest measurement received from the server with the prediction model for the server is evaluated. A corrective action is taken if the latest measurement does not comply with the prediction model for the server. Measurements received from a group of servers may be aggregated and the machine learning system may construct a prediction model for the group of servers. Compliance to the prediction model may be evaluated for the group of servers.