summaryrefslogtreecommitdiff
path: root/doc/design-paper
diff options
context:
space:
mode:
authorSteven Murdoch <Steven.Murdoch@cl.cam.ac.uk>2008-12-25 12:10:27 +0000
committerSteven Murdoch <Steven.Murdoch@cl.cam.ac.uk>2008-12-25 12:10:27 +0000
commit3ba7a6e219b3a341bb650cdc37b3ca1435a1a8a2 (patch)
treed5eccfd07595e80e24c3f1e6e3f7238386bdd33b /doc/design-paper
parenta32477db65f9925708d36cc1217a9997d1ca7d58 (diff)
downloadtor-3ba7a6e219b3a341bb650cdc37b3ca1435a1a8a2.tar.gz
tor-3ba7a6e219b3a341bb650cdc37b3ca1435a1a8a2.zip
Add R script for estimating average node latency at different levels of network load
svn:r17768
Diffstat (limited to 'doc/design-paper')
-rw-r--r--doc/design-paper/node-selection/vary-network-load.R93
1 files changed, 93 insertions, 0 deletions
diff --git a/doc/design-paper/node-selection/vary-network-load.R b/doc/design-paper/node-selection/vary-network-load.R
new file mode 100644
index 0000000000..e2b96dd01f
--- /dev/null
+++ b/doc/design-paper/node-selection/vary-network-load.R
@@ -0,0 +1,93 @@
+## The waiting time for a node (assuming no overloaded nodes)
+## x: 1/bandwidth
+## q: selection probability
+## L: network load
+wait <- function(x,q,L) {
+ a <- q*L*x*x
+ b <- 2*(1-q*x*L)
+ return (x + a/b)
+}
+
+## The weighted wait time
+wwait <- function(x,q,L) {
+ return (q*wait(x,q,L))
+}
+
+## Average latency, returning NA for infinite
+netLatency <- function(x, q, L) {
+ if (any(x*q*L <0 | x*q*L >1)) {
+ return (NA)
+ } else {
+ return (sum(wwait(x, q, L)))
+ }
+}
+
+## Load in data files
+t1 <- read.table("opt_1e-6.pickle.dat", header=TRUE)
+t2 <- read.table("opt_1e-3.pickle.dat", header=TRUE)
+t3 <- read.table("opt_1e-1.pickle.dat", header=TRUE)
+t4 <- read.table("opt_0.75.pickle.dat", header=TRUE)
+t5 <- read.table("opt_0.5.pickle.dat", header=TRUE)
+t6 <- read.table("opt_0.25.pickle.dat", header=TRUE)
+t7 <- read.table("opt_0.1.pickle.dat", header=TRUE)
+tt <- read.table("opt_tor.pickle.dat", header=TRUE)
+
+## Node bandwidth and reciprocal
+bw <- t1$bw
+x <- 1/bw
+
+## Calculate network capcity
+capacity <- sum(bw)
+
+## Calculate selection probabilties that Tor uses
+torProb <- bw/sum(bw)
+
+## Load values to try
+varyLoad <- seq(0.01,0.93,0.01)
+latencyTor <- c()
+latency3 <- c()
+latency4 <- c()
+latency5 <- c()
+for (L in varyLoad) {
+ latencyTor <- append(latencyTor,
+ netLatency(x, torProb, capacity*L))
+ latency3 <- append(latency3,
+ netLatency(x, t3$prob, capacity*L))
+ latency4 <- append(latency4,
+ netLatency(x, t4$prob, capacity*L))
+ latency5 <- append(latency5,
+ netLatency(x, t5$prob, capacity*L))
+}
+
+## Output graph
+pdf("vary-network-load.pdf")
+
+## Set up axes
+yFac <- 1000
+xFac <- 100
+
+ylim <- range(na.omit(c(latencyTor, latency3, latency4, latency5)))
+ylim <- c(0,0.015) * yFac
+xlim <- c(0,1) * xFac
+plot(NA, NA,
+ xlim=xlim, ylim=ylim,
+ frame.plot=FALSE,
+ xlab = "Network load (%)",
+ ylab = "Average queuing delay (ms)",
+ main = "Latency for varying network loads")
+
+## Plot data
+col <- rainbow(8)
+lines(varyLoad*xFac, latency3*yFac, col=col[3])
+lines(varyLoad*xFac, latency4*yFac, col=col[4])
+lines(varyLoad*xFac, latency5*yFac, col=col[5])
+lines(varyLoad*xFac, latencyTor*yFac)
+
+## Plot points at which selection probabilities are optimal
+par(xpd=TRUE)
+points(c(0.9, 0.75, 0.5, 1)*xFac, rep(par("usr")[3], 4),
+ col=c(col[3:5], "black"), pch=20,
+ cex=2)
+
+## Close output device
+dev.off()