Resolution:
One of the differences between the dashboards of FabricServer 1.3 and FabricServer 2.0 is the introduction of sparklines.  Sparklines were named by information design guru Edward Tufte; in Tufte's words, they are "intense, simple, word-sized graphics." On the FabricServer 2.0 dashboard, we use sparklines to quickly convey the recent history of a statistic. The sparklines are the little graphics that appear to the left of the numbers in the Sampled Value column. The red dot at their right end represents the most recent value, which is the reported number; the rest of the little graph indicates the values seen during the recent past. The bottom of the sparkline always corresponds to a value of 0; the top represents the largest value seen during the recent past.

Under the covers, sparkline generation can be affected by three settable parameters:

    * The Broker's statistics purge duration. This parameter sets, in minutes, the amount of time that statistics will remain in memory. That length of time is what we refer to above as the recent past. The default value is 30 minutes.
    * The Broker's statistics manager tick; this parameter, a number of milliseconds, is used by the charting pages (Reports -> Engine Charts and Reports -> Domain Charts) as the sampling window for generating data points. The default is 60000 milliseconds-one minute).
    * Each Engine Daemon's Native Statistic Reporting Frequency (per day). This parameter defines the frequency with which that Daemon's Engines report native stats to the Broker. The default is 2880 times per day, which corresponds to once every 30 seconds. (Tracked stats-those that come from the Container and are specific to particular Domains-are gathered every 10 seconds; allocation stats are gathered once per minute.)

The sparkline generator uses the tick to split the duration into a sequence of slices, while requiring each slice to be at least 2 minutes in length; for each slice it gets a stat value, then stitches that stat value into the sparkline. This works well because, in general, stats are reported more frequently than once every two minutes.

However, in FabricServer 2.0 systems upgraded from FS 1.3 and previous releases, you may see pathological behavior on the part of the sparklines (and the dashboard's values), where they vary wildly from 0 to reasonable values even though no actual variation is occurring. The tipoff is seeing a sawtooth sparkline for some or all of the native statistics (CPU Utilization, DS CPU Utilization, Free Disk, Free Memory and Total Memory), especially when other indicators are showing you a stable grid.

This occurs because before FS 2.0 the native statistic reporting frequency defaulted to six minutes (that is, stats were reported 240 times per day). To fix the misleading sparkline behavior, change the value for reporting frequency. The tick and the statistics purge duration are both defined in broker.xml. The native statistics reporting frequency is set for each Engine Daemon as part of that Daemon's Engine configuration.

For more on sparklines, see the discussion at http://www.edwardtufte.com/bboard/q-and-a-fetch-msg?msg_id=0001OR&topic_id=1.

Sparklines on the FabricServer 2.0 Dashboard