[or-cvs] r15554: Add pydoc summary to StatsSupport package. (torflow/branches/gsoc2008/TorCtl)

[mikeperry at seul.org]

Author: mikeperry
Date: 2008-06-28 16:32:56 -0400 (Sat, 28 Jun 2008)
New Revision: 15554

Modified:
   torflow/branches/gsoc2008/TorCtl/StatsSupport.py
Log:

Add pydoc summary to StatsSupport package.



Modified: torflow/branches/gsoc2008/TorCtl/StatsSupport.py
===================================================================
--- torflow/branches/gsoc2008/TorCtl/StatsSupport.py	2008-06-28 20:16:49 UTC (rev 15553)
+++ torflow/branches/gsoc2008/TorCtl/StatsSupport.py	2008-06-28 20:32:56 UTC (rev 15554)
@@ -1,5 +1,40 @@
+#!/usr/bin/python
 #StatsSupport.py - functions and classes useful for calculating stream/circuit statistics
 
+"""
+
+Support classes for statisics gathering
+
+The StatsSupport package contains several classes that extend PathSupport to
+gather continuous statistics on the Tor network.
+
+The main entrypoint is to extend or create an instance of the StatsHandler
+class. The StatsHandler extends from TorCtl.PathSupport.PathBuilder, which is
+itself a TorCtl.EventHandler. The StatsHandler listens to CIRC and STREAM
+events and gathers all manner of statics on their creation and failure before
+passing the events back up to the PathBuilder code, which manages the actual
+construction and the attachment of streams to circuits.
+
+The package also contains a number of support classes that help gather
+additional statistics on the reliability and performance of routers.
+
+For the purpose of accounting failures, the code tracks two main classes of
+failure: 'actual' failure and 'suspected' failure. The general rule is that an
+actual failure is attributed to the node that directly handled the circuit or
+stream. For streams, this is considered to be the exit node. For circuits, it
+is both the extender and the extendee. 'Suspected' failures, on the other
+hand, are attributed to every member of the circuit up until the extendee for
+circuits, and all hops for streams.
+
+For bandwidth accounting, the average stream bandwidth and the average ratio
+of stream bandwidth to advertised bandwidth are tracked, and when the
+statistics are written, a Z-test is performed to calculate the probabilities
+of these values assuming a normal distribution. Note, however, that it has not
+been verified that this distribution is actually normal. It is likely to be
+something else (pareto, perhaps?).
+
+"""
+
 import sys
 import re
 import random