esmond is a system for collecting, storing, visualizing and analyzing large sets of timeseries data. It was driven by the needs of the ESnet engineering team but is likely useful to a much wider audience. esmond has a RESTful API which allows easy access to the data which is collected. The original focus was on collecting SNMP timeseries data which is still the system’s forte, but there is support for generalized timeseries data. The perfSONAR project has begun using esmond to store timeseries of network measurements.
esmond uses a hybrid model for storing data. Timeseries data such as interface counters is stored using Cassandra. esmond will save the raw data, and create summarizations similar to RRD. However, the system never discards data through summarization, which distinguishes it from RRD (and whisper/ceres). Metadata (such as interface description and interface types from SNMP) are stored in an SQL database. Storing this data in an SQL database allows us to use the full expressiveness of SQL to query this data. Since this data changes relatively infrequently the demands placed on the SQL server are fairly modest. Our production server uses PostgreSQL, but it’s likely that SQLite would work just fine. Data can be visualized using Graphite or through custom visualizations which can query the RESTful API.
esmond was designed to meet the needs of the Network Engineering group at ESnet. The key design goals were: