[tor-commits] [torflow/master] Clean up the spec a tad.
mikeperry at torproject.org
mikeperry at torproject.org
Sun Dec 4 02:13:50 UTC 2011
commit 2d4b7ba817e6c398f5b0137d893d96957ac2adf7
Author: Mike Perry <mikeperry-git at fscked.org>
Date: Sat Dec 3 18:13:37 2011 -0800
Clean up the spec a tad.
---
NetworkScanners/BwAuthority/README.spec.txt | 15 ++++++++-------
1 files changed, 8 insertions(+), 7 deletions(-)
diff --git a/NetworkScanners/BwAuthority/README.spec.txt b/NetworkScanners/BwAuthority/README.spec.txt
index 5eb0661..a1b732e 100644
--- a/NetworkScanners/BwAuthority/README.spec.txt
+++ b/NetworkScanners/BwAuthority/README.spec.txt
@@ -371,10 +371,6 @@
The bandwidth authorities measure F_node: the filtered stream
capacity through a given node (filtering is described in Section 1.6).
- In PID control, we add in this extra failure rate as a damper, to prevent
- the PID control system from driving nodes to CPU overload. Once nodes
- begin failing circuits, we want to stop devoting more capacity.
-
The PID Setpoint, or target for each node is F_avg: the average F_node
value observed across the entire network.
@@ -412,9 +408,9 @@
hours. Since the bandwidth authorities also take on the order of
hours to measure a slice of nodes, we do nothing special here.
- For Guard nodes however, clients have minimum rotation rate of 4-6
+ For Guard nodes however, clients try to keep their Guards for 4-6
weeks. However, on the assumption that they rotate more frequently
- than this, we set our Guard feedback interval to 2 weeks.
+ than this in practice, we set our Guard feedback interval to 2 weeks.
Guard measurements are also used without feedback whenever new
measurements are available, to compensate for changes in Guard flag
@@ -428,7 +424,7 @@
determining when to report a measurement.
The integral component, pid_error_sum is subjected to a decay factor
- of 20% per interval, to prevent unbounded growth in cases without
+ per each interval, to prevent unbounded growth in cases without
convergence.
The differential component is a simple delta, calculating by
@@ -521,6 +517,11 @@
of the circ_error and the original pid_error as the new
pid_error.
+ We use this to prevent the PID control system from driving nodes
+ to CPU overload. Once nodes begin failing circuits, we want to stop
+ devoting additional capacity to them (and decrease it proportional
+ to their failure rate relative to the rest of the network).
+
"bwauthbestratio=0"
If absent, the larger of stream bandwidth vs filtered bandwidth
is used to compute F_node.
More information about the tor-commits
mailing list