The Metrics Facility balances flexibility and minimizing the impact on overall Orchestration Server performance. There is no continuous parsing of XML on the server. Instead, parsing of gmond-generated XML is performed by each managed resource.
This section includes the following information:
By default, RRD-based data aggregation is file-based. Because of the frequency of updates and queries of RRD files, this poses a significant performance issue. The Metrics Facility minimizes I/O contention by using in-memory caching and batched write operations to avoid I/O contention and the resulting performance degradation.
The the Orchestration Server Monitoring Server uses a “one RRD per resource” approach, where a RRD contains the AVERAGE, MIN, and MAX RRAs for multiple metrics (DS). In contrast, the Metrics Facility takes an “inside out” approach, which results in “one RRD per metric.” For 1,000 agents reporting 24 Ganglia metrics, this reduces the number of files dramatically (from 24,000 in the “one RRD per metric, per resource” case, and 1,000 in the “one RRD per resource” case, to 24 in the “one RRD per metric” case). This approach avoids a “too many open files” condition.
NOTE:A “too many open files” condition occurs when the default maximum file descriptors available to a process launched from the Linux shell is exceeded.