Alert Analysis of ABwE Raw capacity data

The data analyzed for alerts is the ABwE exponentially weighted moving average of available bandwidth estimate data for about 40 hosts in 12 countries in N. America (US and Canada), Europe (.ch, .cz, .de, .fr, .it, .nl, .uk) and Asia (.jp, .pk,. su) with bottleneck capacities from 512kbps to 1000Mbps. The data points are measured from SLAC at 3 minute intervals. The Event analysis uses a modified "plateau" algorithm developed from Automated Event Detection for Active Measurement Systems McGregor A. J and Braun H-W. The history buffer duration we used in the plateau algorithm was 600 points (the interval between points was 1 minute) or 10 hours, and the trigger buffer was 60 points or 1 hour.

Connie Logg and Les Cottrell
Program Logic Manual
Perl pseudo code
This page was created from the averaged-new ABwE data on Tue Jun 8 09:17:10 2004 for iepm on nettest1 by: version 1.2, 5/26/04 of runperiod.les using threshold=0.1, statistical_limit = 2 and writing data to /nfs/slac/g/net/iepm-bw/bandwidth-tests/eventanalysis/niit-2. The data goes back 3 days to 4/21/2004 18:00 and includes 1 remote hosts.
EVENT SUMMARY

node = node1.niit.pk
starttime = 4/21/2004 18:00; endtime = 04/25/04 00:00; change threshold = 0.1, statistical_limit = 2, data_column=4
The number of standard deviations that a data value must lie outside the mean to be evaluated as a trigger value is defined by the sensitivity = 2
maximum history buffer size = 600
Number of minutes the throughput must be depressed is controlled by the trigger buffer length = 60

line color = magenta drop in bandwidth greater than 0.1 %, the statistical_limit = 2
Bandwidth Data Event Info

NODENAME lastmean laststdev trigmean trigstdev %change
last-trig F1 %change
lasttrig-trig date&time trigger buffer

triggers = 0; aborted triggers = 0; alerts = 0; datastart=1082601271; dataend=1082876342; duration=275071 ; avg. data separation= 107.3 seconds