Anomaly detector for a group of hosts

Abstract

Techniques for anomaly detection within a group of hosts are described. As one example, a computer-implemented method includes performing a single, distributed task with a plurality of hosts, sending, by each host of the plurality of hosts, processing utilization data for a period of time to an anomaly detector, generating a Fourier component vector for each host that comprises a Fourier component value in each element from the processing utilization data of that host with the anomaly detector, generating a mean value vector that comprises a mean value for each element position of the Fourier component vectors in a respective element position of the mean value vector with the anomaly detector, generating a standard deviation vector that comprises a standard deviation for each element position of the Fourier component vectors in a respective element position of the standard deviation vector with the anomaly detector, generating a host standard deviation vector for each host that comprises a deviation of each element of the Fourier component vector of that host from a respective element of the mean value vector by a number of standard deviations read from a respective element of the standard deviation vector with the anomaly detector, determining, with the anomaly detector, an anomalous host of the plurality of hosts from the host standard deviation vectors when at least one threshold is exceeded in the host standard deviation vectors, and causing an indication for the anomalous host to be sent by the anomaly detector.