Server-Side vs Device-Side Aggregation

Many devices that are connected to Iotellect (via Agents or device drivers) provide different gauge-type or counter-type values, such as temperature readings, odometers, or fuel consumption counters. These values are normally grouped by time periods (aggregated) to allow use consolidated values (averages, minimums and maximums) inside reports, charts and other data processing tools.

Developers of device monitoring and control systems often face a necessity to choose between aggregating data inside a device and passing raw values into Iotellect Server. This article explains pros and cons of each method.

In this case device calculates consolidated values (such as daily averages or monthly minimums) by itself. It then provides the pre-calculated table of values to the server in the form of tabular variable, such as:

Pros of this method:

• Device can continue data aggregation even if connection to the server is not available for long time (however some protocols, such as Iotellect protocol, allow device-side buffering)
• Aggregated statistics is available inside device, e.g. for displaying on the device LCD or internal web interface

Cons of this method:

• Implementation of data aggregation inside the device is quite complicated
• Aggregated statistics will occupy device-side storage space
• If many aggregation types are required (e.g. for charting), such as minutely/hourly/daily/weekly/monthly/yearly averages/minimums/maximums, the device must provide multiple aggregation tables for single counter
• Aggregation tables will fully be re-read from the server during any update, partial reading is not possible
• Aggregation data will be duplicated in the device and in the server database
• Charts won't be able to auto-switch between different aggregation types (e.g. from daily to monthly samples)

In this case device provides raw values to the server in the form of variables. Server polls these variables and uses statistical channels to calculate and store aggregated data points.

Pros of this method:

• Only raw values are sent to the server and, therefore, device-to-server traffic is minimized
• Device resources are not consumed for calculating, storing and exposing aggregated data
• Server provides centralized control for aggregation parameters
• Aggregated data points from multiple devices can be easily combined within server reports of charts

Cons of this method:

• Unless buffering is used, server will not receive raw values from offline devices, what can cause gaps in high-granularity statistics (minutely, hourly) and degradation of precision in low-granularity statistics (daily, monthly, yearly)
• Device itself won't be able to profit from consolidated data, e.g. display it in its web interface

In some systems it may be reasonable to combine both methods, i.e aggregate data on both device and server sides. In this case some server-side data analysis tools will use pre-build device-side consolidation tables, and others will refers to server statistical channels.

Was this page helpful?

Tell us more

Send

Thank you!