Server cluster monitoring
AssigneeInternational Business Machines Corporation
InventorsFrederic J. Bauchot, Jacques Milman
Life of this patent
Estimating the term…
Abstract
A master monitor spawns a local monitor in each node of a server cluster. Each local monitor, responsive to a database event, determines a metric of a performance-related variable. A maximum period between measurements may be imposed to reduce irrelevant results, as may a minimum period. The metric may be stored in a measurement file on the respective node and/or may be reported to the master monitor. The master monitor may make any reported metrics available to a user via a front end.
Claims11
A massively parallel database server cluster monitoring method configured to monitor performance of a server cluster itself configured to host a massively parallel database, the monitoring method comprising:
- starting a master monitor in a database server;
- spawning with the master monitor a local monitor in each node of the server cluster, the local monitor being configured to monitor for a database event and to be responsive to the master monitor;
- responsive to a database event:
- obtaining a current time;
- obtaining a current value of a first performance-related variable;
- determining whether a measurement file is present;
- responsive to the measurement file not being present, creating the measurement file;
- responsive to the measurement file being present:
- reading from the measurement file a last time and a last value of the first performance-related variable;
- and
- determining a first metric based on a difference between the current value of the first performance-related variable and the last value of the first performance-related variable;
- storing in the measurement file the current time as the last time and the current value of the first performance-related variable as the last value of the first performance-related variable;
- and
- returning to monitoring for a database event.
A computer program product for monitoring performance of a massively parallel database server cluster, the server cluster including a computing device configured to execute the computer program product, the computer program product comprising instructions in the form of computer executable program code stored on a tangible computer readable storage medium such that the computer executable program code configures the computing device to:
- start a master monitor in a database server;
- spawn with the master monitor a local monitor in each node of the server cluster, the local monitor being configured to monitor for a database event and to be responsive to the master monitor;
- responsive to a database event:
- obtain a current time;
- obtain a current value of a first performance-related variable;
- determine whether a measurement file is present;
- responsive to the measurement file not being present, create the measurement file;
- responsive to the measurement file being present:
- read from the measurement file a last time and a last value of the first performance-related variable;
- and
- determine a first metric based on a difference between the current value of the first performance-related variable and the last value of the first performance-related variable;
- store in the measurement file the current time as the last time and the current value of the first performance-related variable as the last value of the first performance-related variable;
- and
- return to monitoring for a database event.
Dependencies are inferred from the claim text.