- tor-commits - lists.torproject.org

[metrics-tasks/master] Add graph code for #2394.
by karsten＠torproject.org 20 Apr '11

20 Apr '11

commit c6887f322a243ef5526e2b793c5e461936722080 Author: Karsten Loesing <karsten.loesing(a)gmx.net> Date: Wed Apr 20 12:01:00 2011 +0200 Add graph code for #2394. --- task-2394/.gitignore | 1 + task-2394/bandwidth-comparison.R | 153 ++++++++++++++++++++++++++++++++++++++ 2 files changed, 154 insertions(+), 0 deletions(-) diff --git a/task-2394/.gitignore b/task-2394/.gitignore index bef5dec..1e3c939 100644 --- a/task-2394/.gitignore +++ b/task-2394/.gitignore @@ -3,4 +3,5 @@ descriptors/ *.pdf *.csv *.jar +*.png diff --git a/task-2394/bandwidth-comparison.R b/task-2394/bandwidth-comparison.R new file mode 100644 index 0000000..77a5842 --- /dev/null +++ b/task-2394/bandwidth-comparison.R @@ -0,0 +1,153 @@ +options(warn = -1) +suppressPackageStartupMessages(library("ggplot2")) + +b <- read.csv("bandwidth-comparison.csv", stringsAsFactors = FALSE) + +# Plot ECDF to compare categories +cdf_relays_category <- function(data, category) { + d <- data[data$category == category & data$descriptorbandwidth > 0, ] + d <- sort(d$consensusbandwidth * 1000 / d$descriptorbandwidth) + d <- data.frame(x = d, y = (1:length(d)) / length(d), + category = category) + d +} +relays_category <- rbind( + cdf_relays_category(b, "Guard & Exit (default policy)"), + cdf_relays_category(b, "Exit (default policy)"), + cdf_relays_category(b, "Guard & Exit (non-default policy)"), + cdf_relays_category(b, "Exit (non-default policy)"), + cdf_relays_category(b, "Guard"), + cdf_relays_category(b, "Middle")) +ggplot(relays_category, aes(x = x, y = y, colour = category)) + +geom_line() + +scale_x_log10("\nRatio of measured by self-reported bandwidth", + limits = c(0.1, 10), breaks = c(0.1, 0.2, 0.5, 1, 2, 5, 10), + labels = c("0.1", "0.2", "0.5", "1", "2", "5", "10")) + +scale_y_continuous("Fraction of relays\n", limits = c(0, 1), + formatter = "percent") + +scale_colour_hue("") + +geom_vline(xintercept = 1, legend = FALSE, linetype = "dashed") + +opts(title = "Ratio between measured and self-reported relay bandwidth", + legend.position = "top") +ggsave(filename = "bandwidth-comparison-relays.png", + width = 8, height = 5, dpi = 150) + +stopit + +# Plot ECDFs to compare consensus to votes +cdf_relays_category_votes <- function(data, category) { + d <- data[data$category == category & data$descriptorbandwidth > 0, ] + consensus <- sort(d$consensusbandwidth * 1000 / d$descriptorbandwidth) + ides <- sort(d$idesbandwidth * 1000 / d$descriptorbandwidth) + urras <- sort(d$urrasbandwidth * 1000 / d$descriptorbandwidth) + moria1 <- sort(d$moria1bandwidth * 1000 / d$descriptorbandwidth) + gabelmoo <- sort(d$gabelmoobandwidth * 1000 / d$descriptorbandwidth) + d <- rbind( + data.frame(x = consensus, + y = (1:length(consensus)) / length(consensus), + source = "consensus", + category = category), + data.frame(x = ides, + y = (1:length(ides)) / length(ides), + source = "ides", + category = category), + data.frame(x = urras, + y = (1:length(urras)) / length(urras), + source = "urras", + category = category), + data.frame(x = moria1, + y = (1:length(moria1)) / length(moria1), + source = "moria1", + category = category), + data.frame(x = gabelmoo, + y = (1:length(gabelmoo)) / length(gabelmoo), + source = "gabelmoo", + category = category)) + d +} +relays_category_votes <- rbind( + cdf_relays_category_votes(b, "Guard & Exit (default policy)"), + cdf_relays_category_votes(b, "Exit (default policy)"), + cdf_relays_category_votes(b, "Guard & Exit (non-default policy)"), + cdf_relays_category_votes(b, "Exit (non-default policy)"), + cdf_relays_category_votes(b, "Guard"), + cdf_relays_category_votes(b, "Middle")) +ggplot(relays_category_votes, aes(x = x, y = y, colour = source)) + +geom_line() + +facet_wrap(~ category, ncol = 3) + +scale_x_log10("\nRatio of measured by self-reported bandwidth", + limits = c(0.1, 10), breaks = c(0.1, 1, 10), + labels = c("0.1", "1", "10")) + +scale_y_continuous("Fraction of relays\n", limits = c(0, 1), + formatter = "percent") + +scale_colour_manual("", + breaks = c("consensus", "ides", "urras", "moria1", "gabelmoo"), + values = c("black", rep(alpha("black", 0.25), 4))) + +geom_vline(xintercept = 1, legend = FALSE, linetype = "dotted") + +opts(title = paste("Measured vs. self-reported bandwidth ratios in", + "consensus and votes\n"), legend.position = "none") +ggsave(filename = "bandwidth-comparison-relays-votes.png", + width = 8, height = 5, dpi = 150) + +# Plot _weighted_ ECDFs +wecdf <- function(data, source, category) { + data <- data[with(data, order(ratio)), ] + sum_measured <- sum(data$measured, na.rm = TRUE) + cur_measured <- data$measured[1] + res <- data.frame(x = data$ratio[1], + y = cur_measured / sum_measured, + source = source, category = category) + for (i in 2:length(data$ratio)) { + cur_measured <- cur_measured + data$measured[i] + res <- rbind(res, data.frame(x = data$ratio[i], + y = cur_measured / sum_measured, + source = source, category = category)) + } + res +} +cdf_measured_category_votes <- function(data, category) { + d <- data[data$category == category & data$descriptorbandwidth > 0, ] + d <- rbind( + wecdf(data.frame( + ratio = d$consensusbandwidth * 1000 / d$descriptorbandwidth, + measured = d$consensusbandwidth), "consensus", category), + wecdf(data.frame( + ratio = d$idesbandwidth * 1000 / d$descriptorbandwidth, + measured = d$idesbandwidth), "ides", category), + wecdf(data.frame( + ratio = d$urrasbandwidth * 1000 / d$descriptorbandwidth, + measured = d$urrasbandwidth), "urras", category), + wecdf(data.frame( + ratio = d$moria1bandwidth * 1000 / d$descriptorbandwidth, + measured = d$moria1bandwidth), "moria1", category), + wecdf(data.frame( + ratio = d$gabelmoobandwidth * 1000 / d$descriptorbandwidth, + measured = d$gabelmoobandwidth), "gabelmoo", category)) + d +} +measured_category_votes <- rbind( + cdf_measured_category_votes(b, "Guard & Exit (default policy)"), + cdf_measured_category_votes(b, "Exit (default policy)"), + cdf_measured_category_votes(b, "Guard & Exit (non-default policy)"), + cdf_measured_category_votes(b, "Exit (non-default policy)"), + cdf_measured_category_votes(b, "Guard"), + cdf_measured_category_votes(b, "Middle")) +ggplot(measured_category_votes, aes(x = x, y = y, colour = source)) + +geom_line() + +facet_wrap(~ category, ncol = 3) + +scale_x_log10("\nRatio of measured by self-reported bandwidth", + limits = c(0.1, 10), breaks = c(0.1, 1, 10), + labels = c("0.1", "1", "10")) + +scale_y_continuous("Fraction of measured bandwidth\n", limits = c(0, 1), + formatter = "percent") + +scale_colour_manual("", + breaks = c("consensus", "ides", "urras", "moria1", "gabelmoo"), + values = c("black", rep(alpha("black", 0.25), 4))) + +geom_vline(xintercept = 1, legend = FALSE, linetype = "dotted") + +opts(title = paste("Measured vs. self-reported bandwidth ratios in", + "consensus and votes\n"), legend.position = "none") +ggsave(filename = "bandwidth-comparison-measured-votes.png", + width = 8, height = 5, dpi = 150) +write.csv(measured_category_votes, "measured_category_votes-temp.csv", + quote = FALSE, row.names = FALSE) +

1 0

[tor/master] Prevent hugely inflated observed bandwidth values
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 3f7f96d9e7aa2254d62d63530c37ee96d4b3abc8 Author: Sebastian Hahn <sebastian(a)torproject.org> Date: Tue Apr 19 16:00:41 2011 +0200 Prevent hugely inflated observed bandwidth values When reading the bw history from the state file, we'd add the 900-second value as traffic that occured during one second. Fix that by adding the average value to each second. This bug was present since 0.2.0.5-alpha, but was hidden until 0.2.23-alpha when we started using the saved values. --- changes/bug2704_part2 | 5 +++++ src/or/rephist.c | 19 ++++++++++++++++--- 2 files changed, 21 insertions(+), 3 deletions(-) diff --git a/changes/bug2704_part2 b/changes/bug2704_part2 new file mode 100644 index 0000000..962c8b7 --- /dev/null +++ b/changes/bug2704_part2 @@ -0,0 +1,5 @@ + o Major bugfixes: + - Prevent relays that read their bandwidth history from their state file + from arbitrarily inflating that value. Fixes the second half of bug + 2704, bugfix on tor-0.2.2.23-alpha. + diff --git a/src/or/rephist.c b/src/or/rephist.c index 6034bbc..fb091d5 100644 --- a/src/or/rephist.c +++ b/src/or/rephist.c @@ -1686,11 +1686,24 @@ rep_hist_load_bwhist_state_section(bw_array_t *b, } if (start < now) { - add_obs(b, start, v); + time_t cur_start = start; + time_t actual_interval_len = s_interval; + uint64_t cur_val = 0; + /* Calculate the average per second. This is the best we can do + * because our state file doesn't have per-second resolution. */ + if (start + s_interval > now) + actual_interval_len = now - start; + cur_val = v / actual_interval_len; + /* This is potentially inefficient, but since we don't do it very + * often it should be ok. */ + while (cur_start < start + actual_interval_len) { + add_obs(b, cur_start, cur_val); + ++cur_start; + } b->max_total = mv; /* This will result in some fairly choppy history if s_interval - * is notthe same as NUM_SECS_BW_SUM_INTERVAL. XXXX */ - start += s_interval; + * is not the same as NUM_SECS_BW_SUM_INTERVAL. XXXX */ + start += actual_interval_len; } } SMARTLIST_FOREACH_END(cp); }

1 0

[tor/master] Merge remote-tracking branch 'origin/maint-0.2.2'
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 7c83d4043d7f336c14a53418ce91b12335626f6b Merge: 4a7f979 3f7f96d Author: Nick Mathewson <nickm(a)torproject.org> Date: Tue Apr 19 15:39:04 2011 -0400 Merge remote-tracking branch 'origin/maint-0.2.2' changes/bug2704_part1 | 5 +++++ changes/bug2704_part2 | 5 +++++ src/or/rephist.c | 21 +++++++++++++++++---- 3 files changed, 27 insertions(+), 4 deletions(-)

1 0

[tor/master] Don't sometimes undercount bw average
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 13c3884ff60aa0e58464f379f988683f1c30438a Author: Sebastian Hahn <sebastian(a)torproject.org> Date: Tue Apr 19 17:12:27 2011 +0200 Don't sometimes undercount bw average This fixes the first part of bug 2704. It should occur only rarely when no bw maxima are known. Bugfix on 0.2.2.23-alpha. Fixes bug 2704. --- changes/bug2704_part1 | 5 +++++ src/or/rephist.c | 2 +- 2 files changed, 6 insertions(+), 1 deletions(-) diff --git a/changes/bug2704_part1 b/changes/bug2704_part1 new file mode 100644 index 0000000..eaf2281 --- /dev/null +++ b/changes/bug2704_part1 @@ -0,0 +1,5 @@ + o Minor bugfixes: + - Fix an issue causing calculation of Tor's average bandwidth as saved + in the state file to be 10 times smaller than it should be. Fixes the + first part of bug 2704, bugfix on tor-0.2.2.23-alpha. + diff --git a/src/or/rephist.c b/src/or/rephist.c index 9b7eefe..6034bbc 100644 --- a/src/or/rephist.c +++ b/src/or/rephist.c @@ -1673,7 +1673,7 @@ rep_hist_load_bwhist_state_section(bw_array_t *b, mv *= NUM_SECS_ROLLING_MEASURE; } else { /* No maxima known; guess average rate to be conservative. */ - mv = v / s_interval; + mv = (v / s_interval) * NUM_SECS_ROLLING_MEASURE; } if (!ok) { retval = -1;

1 0

[tor/maint-0.2.2] Don't sometimes undercount bw average
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 13c3884ff60aa0e58464f379f988683f1c30438a Author: Sebastian Hahn <sebastian(a)torproject.org> Date: Tue Apr 19 17:12:27 2011 +0200 Don't sometimes undercount bw average This fixes the first part of bug 2704. It should occur only rarely when no bw maxima are known. Bugfix on 0.2.2.23-alpha. Fixes bug 2704. --- changes/bug2704_part1 | 5 +++++ src/or/rephist.c | 2 +- 2 files changed, 6 insertions(+), 1 deletions(-) diff --git a/changes/bug2704_part1 b/changes/bug2704_part1 new file mode 100644 index 0000000..eaf2281 --- /dev/null +++ b/changes/bug2704_part1 @@ -0,0 +1,5 @@ + o Minor bugfixes: + - Fix an issue causing calculation of Tor's average bandwidth as saved + in the state file to be 10 times smaller than it should be. Fixes the + first part of bug 2704, bugfix on tor-0.2.2.23-alpha. + diff --git a/src/or/rephist.c b/src/or/rephist.c index 9b7eefe..6034bbc 100644 --- a/src/or/rephist.c +++ b/src/or/rephist.c @@ -1673,7 +1673,7 @@ rep_hist_load_bwhist_state_section(bw_array_t *b, mv *= NUM_SECS_ROLLING_MEASURE; } else { /* No maxima known; guess average rate to be conservative. */ - mv = v / s_interval; + mv = (v / s_interval) * NUM_SECS_ROLLING_MEASURE; } if (!ok) { retval = -1;

1 0

[tor/maint-0.2.2] Prevent hugely inflated observed bandwidth values
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 3f7f96d9e7aa2254d62d63530c37ee96d4b3abc8 Author: Sebastian Hahn <sebastian(a)torproject.org> Date: Tue Apr 19 16:00:41 2011 +0200 Prevent hugely inflated observed bandwidth values When reading the bw history from the state file, we'd add the 900-second value as traffic that occured during one second. Fix that by adding the average value to each second. This bug was present since 0.2.0.5-alpha, but was hidden until 0.2.23-alpha when we started using the saved values. --- changes/bug2704_part2 | 5 +++++ src/or/rephist.c | 19 ++++++++++++++++--- 2 files changed, 21 insertions(+), 3 deletions(-) diff --git a/changes/bug2704_part2 b/changes/bug2704_part2 new file mode 100644 index 0000000..962c8b7 --- /dev/null +++ b/changes/bug2704_part2 @@ -0,0 +1,5 @@ + o Major bugfixes: + - Prevent relays that read their bandwidth history from their state file + from arbitrarily inflating that value. Fixes the second half of bug + 2704, bugfix on tor-0.2.2.23-alpha. + diff --git a/src/or/rephist.c b/src/or/rephist.c index 6034bbc..fb091d5 100644 --- a/src/or/rephist.c +++ b/src/or/rephist.c @@ -1686,11 +1686,24 @@ rep_hist_load_bwhist_state_section(bw_array_t *b, } if (start < now) { - add_obs(b, start, v); + time_t cur_start = start; + time_t actual_interval_len = s_interval; + uint64_t cur_val = 0; + /* Calculate the average per second. This is the best we can do + * because our state file doesn't have per-second resolution. */ + if (start + s_interval > now) + actual_interval_len = now - start; + cur_val = v / actual_interval_len; + /* This is potentially inefficient, but since we don't do it very + * often it should be ok. */ + while (cur_start < start + actual_interval_len) { + add_obs(b, cur_start, cur_val); + ++cur_start; + } b->max_total = mv; /* This will result in some fairly choppy history if s_interval - * is notthe same as NUM_SECS_BW_SUM_INTERVAL. XXXX */ - start += s_interval; + * is not the same as NUM_SECS_BW_SUM_INTERVAL. XXXX */ + start += actual_interval_len; } } SMARTLIST_FOREACH_END(cp); }

1 0

[metrics-tasks/master] Add code to generate the #1991 graphs.
by karsten＠torproject.org 19 Apr '11

19 Apr '11

commit 66474d34ad189d7ab6e6f0ad6359452bbed77e89 Author: Karsten Loesing <karsten.loesing(a)gmx.net> Date: Tue Apr 19 21:21:05 2011 +0200 Add code to generate the #1991 graphs. --- task-1991/.gitignore | 6 ++ task-1991/Aggregate.java | 131 ++++++++++++++++++++++++++++ task-1991/Merge.java | 104 ++++++++++++++++++++++ task-1991/README | 46 ++++++++++ task-1991/torperf-guard-bandwidths-ranks.R | 40 +++++++++ task-1991/torperf-guard-quantiles.R | 57 ++++++++++++ 6 files changed, 384 insertions(+), 0 deletions(-) diff --git a/task-1991/.gitignore b/task-1991/.gitignore new file mode 100644 index 0000000..a0fc089 --- /dev/null +++ b/task-1991/.gitignore @@ -0,0 +1,6 @@ +*.mergedata +*.class +*.png +*.csv +Rplots.pdf + diff --git a/task-1991/Aggregate.java b/task-1991/Aggregate.java new file mode 100644 index 0000000..682a553 --- /dev/null +++ b/task-1991/Aggregate.java @@ -0,0 +1,131 @@ +import java.io.*; +import java.util.*; + +public class Aggregate { + public static void main(String[] args) throws Exception { + BufferedReader br = new BufferedReader(new FileReader( + "torperf-guard-bandwidths-ranks.csv")); + String line = br.readLine(); + List<String> sortedByBandwidth = new ArrayList<String>(), + sortedByRank = new ArrayList<String>(); + while ((line = br.readLine()) != null) { + sortedByBandwidth.add(line); + sortedByRank.add(line); + } + br.close(); + Collections.sort(sortedByBandwidth, new Comparator<String>() { + public int compare(String a, String b) { + return Integer.parseInt(a.split(",")[0]) - + Integer.parseInt(b.split(",")[0]); + } + }); + Collections.sort(sortedByRank, new Comparator<String>() { + public int compare(String a, String b) { + return (int) (100.0 * (Double.parseDouble(a.split(",")[1]) - + Double.parseDouble(b.split(",")[1]))); + } + }); + + List<Integer> percentiles = new ArrayList<Integer>(); + for (int percentile = 1; percentile < 100; percentile += 1) { + percentiles.add(percentile); + } + + BufferedWriter bw = new BufferedWriter(new FileWriter( + "torperf-guard-rank-quantiles.csv")); + bw.write("filesize,rank"); + for (int percentile : percentiles) { + bw.write(",p" + percentile); + } + bw.write(",len\n"); + for (String filesize : Arrays.asList("50kb,1mb,5mb".split(","))) { + double rankPercentInterval = 0.02; + double rankPercent = rankPercentInterval; + List<Integer> times = new ArrayList<Integer>(); + for (int i = 0; i <= sortedByRank.size(); i++) { + double rank = -1.0; + int completionTime = -1; + if (i < sortedByRank.size()) { + String[] parts = sortedByRank.get(i).split(","); + rank = Double.parseDouble(parts[1]); + completionTime = Integer.parseInt(parts[2]); + if (!parts[4].equals(filesize)) { + continue; + } + } + if (i == sortedByRank.size() || rank > rankPercent) { + String rankString = String.format("%.3f", + rankPercent - 0.5 * rankPercentInterval); + bw.write(filesize + "," + rankString); + if (times.size() > 1) { + Collections.sort(times); + for (int percentile : percentiles) { + bw.write("," + times.get(times.size() * percentile / 100)); + } + } else { + for (int percentile : percentiles) { + bw.write(",NA"); + } + } + bw.write("," + times.size() + "\n"); + times.clear(); + rankPercent += rankPercentInterval; + } + if (i == sortedByRank.size()) { + break; + } + times.add(completionTime); + } + } + bw.close(); + + bw = new BufferedWriter(new FileWriter( + "torperf-guard-bandwidth-quantiles.csv")); + bw.write("filesize,bandwidth"); + for (int percentile : percentiles) { + bw.write(",p" + percentile); + } + bw.write(",len\n"); + for (String filesize : Arrays.asList("50kb,1mb,5mb".split(","))) { + int bandwidthInterval = 2500; + int curBandwidth = bandwidthInterval; + List<Integer> times = new ArrayList<Integer>(); + for (int i = 0; i <= sortedByBandwidth.size(); i++) { + int bandwidth = -1; + int completionTime = -1; + if (i < sortedByBandwidth.size()) { + String[] parts = sortedByBandwidth.get(i).split(","); + bandwidth = Integer.parseInt(parts[0]); + completionTime = Integer.parseInt(parts[2]); + if (!parts[4].equals(filesize)) { + continue; + } + } + if (i == sortedByBandwidth.size() || bandwidth > curBandwidth) { + String bandwidthString = String.format("%d", + curBandwidth - (bandwidthInterval / 2)); + bw.write(filesize + "," + bandwidthString); + if (times.size() > 1) { + Collections.sort(times); + for (int percentile : percentiles) { + bw.write("," + times.get(times.size() * percentile / 100)); + } + } else { + for (int percentile : percentiles) { + bw.write(",NA"); + } + } + bw.write("," + times.size() + "\n"); + times.clear(); + curBandwidth += bandwidthInterval; + } + if (i == sortedByBandwidth.size()) { + break; + } + times.add(completionTime); + } + } + bw.close(); + } +} + diff --git a/task-1991/Merge.java b/task-1991/Merge.java new file mode 100644 index 0000000..622cbf6 --- /dev/null +++ b/task-1991/Merge.java @@ -0,0 +1,104 @@ +import java.io.*; +import java.text.*; +import java.util.*; + +public class Merge { + public static void main(String[] args) throws Exception { + + System.out.println("Reading guard node bandwidths..."); + SortedMap<String, String> bandwidthRanks = + new TreeMap<String, String>(); + BufferedReader br = new BufferedReader(new FileReader( + "bandwidths-sql.csv")); + String line = br.readLine(), lastDateTime = null; + List<String> currentRelays = new ArrayList<String>(); + while ((line = br.readLine()) != null) { + if (line.startsWith("fingerprint") || line.startsWith("(")) { + continue; + } + String[] parts = line.split(","); + String fingerprint = parts[0], dateTime = parts[1], + bandwidth = parts[2]; + if (lastDateTime != null && !dateTime.equals(lastDateTime)) { + Collections.sort(currentRelays, new Comparator<String>() { + public int compare(String a, String b) { + return Integer.parseInt(a.split(",")[2]) - + Integer.parseInt(b.split(",")[2]); + } + }); + for (int i = 0; i < currentRelays.size(); i++) { + String relay = currentRelays.get(i); + String relayParts[] = currentRelays.get(i).split(","); + String relayFingerprint = relayParts[0]; + String relayBandwidth = relayParts[2]; + bandwidthRanks.put(relayFingerprint + "," + lastDateTime, + String.format("%s,%.6f", relayBandwidth, (double) i / + (double) (currentRelays.size() - 1))); + } + currentRelays.clear(); + } + lastDateTime = dateTime; + currentRelays.add(line); + } + br.close(); + + System.out.println("Reading .mergedata file and writing completion " + + "time, guard bandwidth, and rank to disk..."); + SortedMap<Integer, List<Long>> aggregatedResults = + new TreeMap<Integer, List<Long>>(); + BufferedWriter bw = new BufferedWriter(new FileWriter( + "torperf-guard-bandwidths-ranks.csv")); + bw.write("bandwidth,rank,completiontime,guards,filesize\n"); + SimpleDateFormat dateTimeFormat = + new SimpleDateFormat("yyyy-MM-dd HH:mm:ss"); + dateTimeFormat.setTimeZone(TimeZone.getTimeZone("UTC")); + for (File mergedataFile : new File(".").listFiles()) { + String filename = mergedataFile.getName(); + if (!filename.endsWith(".mergedata")) { + continue; + } + String guards = filename.substring(0, filename.indexOf("80cbt")); + String filesize = filename.split("-")[1].split("\\.")[0]; + br = new BufferedReader(new FileReader(mergedataFile)); + while ((line = br.readLine()) != null) { + String path = null; + long started = 0L, completed = 0L; + for (String part : line.split(" ")) { + String key = part.substring(0, part.indexOf("=")); + String value = part.substring(part.indexOf("=") + 1); + if (key.equals("PATH")) { + path = value; + } else if (key.equals("STARTSEC")) { + started += Long.parseLong(value) * 1000L; + } else if (key.equals("STARTUSEC")) { + started += Long.parseLong(value) / 1000L; + } else if (key.equals("DATACOMPLETESEC")) { + completed += Long.parseLong(value) * 1000L; + } else if (key.equals("DATACOMPLETEUSEC")) { + completed += Long.parseLong(value) / 1000L; + } else if (key.equals("DIDTIMEOUT")) { + if (value.equals("1")) { + continue; + } + } + } + if (path == null || started == 0L || completed == 0L) { + continue; + } + String dateTime = dateTimeFormat.format(started); + String fingerprint = path.split(",")[0].substring(1).toLowerCase(); + String guardKey = fingerprint + "," + dateTime; + String previousGuardKey = bandwidthRanks.headMap(guardKey).lastKey(); + if (previousGuardKey.startsWith(fingerprint)) { + String bandwidthRank = bandwidthRanks.get(previousGuardKey); + long completionTime = completed - started; + bw.write(bandwidthRank + "," + completionTime + "," + guards + + "," + filesize + "\n"); + } + } + } + br.close(); + bw.close(); + } +} + diff --git a/task-1991/README b/task-1991/README new file mode 100644 index 0000000..8cf5548 --- /dev/null +++ b/task-1991/README @@ -0,0 +1,46 @@ +Visualize influence of guard bandwidth on Torperf completion time + + - Generate one or more .mergedata files, e.g., + + $ ./consolidate_stats.py 50kb.data 50kb.extradata + regular80cbt-50kb.mergedata + + - Extract consensus bandwidths of all relays and write them to a CSV + file. The easiest way to do this is to run an SQL query on the metrics + database: + + => \f ',' + => \a + => \t + => \o bandwidths-sql.csv + => SELECT fingerprint, validafter, bandwidth FROM statusentry + WHERE validafter >= '2011-02-23' ORDER BY validafter, fingerprint; + => \o + => \t + => \a + + - Merge the .mergedata file with the consensus bandwidth file to have a + single CSV file that contains Torperf completion times and Guard node + consensus bandwidths and ranks: + + $ javac Merge.java && java -Xmx2048m Merge + + The result is a file torperf-guard-bandwidths-ranks.csv. + + - Aggregate the results to obtain percentiles: + + $ javac Aggregate.java && java Aggregate + + The result are two files torperf-guard-bandwidth-quantiles.csv and + torperf-guard-rank-quantiles.csv. + + - Plot Torperf results by guard consensus bandwidth and ranks for each + guard selection strategy separately: + + $ R --slave -f torperf-guard-bandwidths-ranks.R + + - Plot Torperf results by guard consensus bandwidth and ranks for all + guard selection strategies and including quantile lines: + + $ R --slave -f torperf-guard-quantiles.R + diff --git a/task-1991/torperf-guard-bandwidths-ranks.R b/task-1991/torperf-guard-bandwidths-ranks.R new file mode 100644 index 0000000..8e7eb0a --- /dev/null +++ b/task-1991/torperf-guard-bandwidths-ranks.R @@ -0,0 +1,40 @@ +library(ggplot2) + +# Read data +data <- read.csv("torperf-guard-bandwidths-ranks.csv", + stringsAsFactors = FALSE) + +data <- data[(data$filesize == "50kb" & data$completiontime < 60000) | + (data$filesize == "1mb" & data$completiontime < 120000) | + (data$filesize == "5mb" & data$completiontime < 300000), ] + +data[data$guards == "slow", "guards"] <- "a) slowest overall" +data[data$guards == "slowratio", "guards"] <- "b) slowest ratio" +data[data$guards == "regular", "guards"] <- "c) default" +data[data$guards == "fastratio", "guards"] <- "d) fastest ratio" +data[data$guards == "fast", "guards"] <- "e) fastest overall" +data[data$filesize == "50kb", "filesize"] <- "a) 50 KB" +data[data$filesize == "1mb", "filesize"] <- "b) 1 MB" +data[data$filesize == "5mb", "filesize"] <- "c) 5 MB" + +ggplot(data, aes(x = bandwidth / 1000, y = completiontime / 1000)) + +geom_point(alpha = 0.05) + +scale_x_continuous("\nGuard consensus bandwidth in MB/s") + +scale_y_continuous("Torperf completion time in seconds\n") + +facet_grid(filesize ~ guards, scale = "free_y") + +opts(legend.position = "none") +ggsave(filename = "torperf-guard-bandwidths.png", + width = 8, height = 5, dpi = 150) + +ggplot(data, aes(x = as.numeric(rank), y = completiontime / 1000)) + +geom_point(alpha = 0.05) + +scale_x_continuous(paste("\nGuard rank by consensus bandwidth from", + "slowest (0) to fastest (1)"), limits = c(0, 1), + breaks = c(0.25, 0.5, 0.75)) + +scale_y_continuous("Torperf completion time in seconds\n") + +facet_grid(filesize ~ guards, scale = "free_y") + +opts(legend.position = "none") +ggsave(filename = "torperf-guard-ranks.png", + width = 8, height = 5, dpi = 150) + + diff --git a/task-1991/torperf-guard-quantiles.R b/task-1991/torperf-guard-quantiles.R new file mode 100644 index 0000000..2554c34 --- /dev/null +++ b/task-1991/torperf-guard-quantiles.R @@ -0,0 +1,57 @@ +library(ggplot2) + +# Read data +data <- read.csv("torperf-guard-bandwidths-ranks.csv", + stringsAsFactors = FALSE) +data <- data[(data$filesize == "50kb" & data$completiontime < 60000) | + (data$filesize == "1mb" & data$completiontime < 600000) | + (data$filesize == "5mb" & data$completiontime < 1500000), ] +data[data$filesize == "50kb", "filesize"] <- "a) 50 KB" +data[data$filesize == "1mb", "filesize"] <- "b) 1 MB" +data[data$filesize == "5mb", "filesize"] <- "c) 5 MB" + +percentiles <- paste("p", seq(90, 90, 1), sep = "") + +rq <- read.csv("torperf-guard-rank-quantiles.csv", + stringsAsFactors = FALSE) +rq[(rq$len < 30 & rq$filesize == "50kb") | + (rq$len < 10 & rq$filesize == "1mb") | + (rq$len < 5 & rq$filesize == "5mb"), percentiles] <- NA +rq <- rq[, c("filesize", "rank", percentiles)] +rq <- melt(rq, id = c("filesize", "rank")) +rq[rq$filesize == "50kb", "filesize"] <- "a) 50 KB" +rq[rq$filesize == "1mb", "filesize"] <- "b) 1 MB" +rq[rq$filesize == "5mb", "filesize"] <- "c) 5 MB" +ggplot(data, aes(x = as.numeric(rank), y = completiontime / 1000)) + +geom_point(alpha = 0.05) + +scale_x_continuous(paste("\nGuard rank by consensus bandwidth from", + "slowest (0) to fastest (1)"), limits = c(0, 1)) + +scale_y_continuous("Torperf completion time in seconds\n") + +geom_line(data = rq, aes(x = as.numeric(rank), y = value / 1000, + colour = variable)) + +facet_grid(filesize ~ ., scale = "free_y") + +opts(legend.position = "none") +ggsave(filename = "torperf-guard-rank-quantiles.png", + width = 8, height = 5, dpi = 150) + +bq <- read.csv("torperf-guard-bandwidth-quantiles.csv", + stringsAsFactors = FALSE) +bq[(bq$len < 30 & bq$filesize == "50kb") | + (bq$len < 10 & bq$filesize == "1mb") | + (bq$len < 5 & bq$filesize == "5mb"), percentiles] <- NA +bq <- bq[, c("filesize", "bandwidth", percentiles)] +bq <- melt(bq, id = c("filesize", "bandwidth")) +bq[bq$filesize == "50kb", "filesize"] <- "a) 50 KB" +bq[bq$filesize == "1mb", "filesize"] <- "b) 1 MB" +bq[bq$filesize == "5mb", "filesize"] <- "c) 5 MB" +ggplot(data, aes(x = bandwidth / 1000, y = completiontime / 1000)) + +geom_point(alpha = 0.05) + +scale_x_continuous("\nGuard consensus bandwidth in MB/s") + +scale_y_continuous("Torperf completion time in seconds\n") + +geom_line(data = bq, aes(x = bandwidth / 1000, y = value / 1000, + colour = variable)) + +facet_grid(filesize ~ ., scale = "free_y") + +opts(legend.position = "none") +ggsave(filename = "torperf-guard-bandwidth-quantiles.png", + width = 8, height = 5, dpi = 150) +

1 0

[tor/master] Standardize our printf code on %d, not %i.
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 5cc322e5472996fda6095e54de302c0f5c68ea92 Author: Nick Mathewson <nickm(a)torproject.org> Date: Tue Apr 19 12:40:26 2011 -0400 Standardize our printf code on %d, not %i. --- src/common/address.c | 2 +- src/or/routerparse.c | 4 ++-- src/test/test_addr.c | 4 ++-- 3 files changed, 5 insertions(+), 5 deletions(-) diff --git a/src/common/address.c b/src/common/address.c index adc0ef0..aff517c 100644 --- a/src/common/address.c +++ b/src/common/address.c @@ -604,7 +604,7 @@ tor_addr_parse_mask_ports(const char *s, tor_addr_t *addr_out, if (family == AF_INET6 && v4map) { if (bits > 32 && bits < 96) { /* Crazy */ log_warn(LD_GENERAL, - "Bad mask bits %i for V4-mapped V6 address; rejecting.", + "Bad mask bits %d for V4-mapped V6 address; rejecting.", bits); goto err; } diff --git a/src/or/routerparse.c b/src/or/routerparse.c index dd72eb6..ba29f05 100644 --- a/src/or/routerparse.c +++ b/src/or/routerparse.c @@ -4641,8 +4641,8 @@ rend_parse_v2_service_descriptor(rend_service_descriptor_t **parsed_out, if (eos-desc > REND_DESC_MAX_SIZE) { /* XXX023 If we are parsing this descriptor as a server, this * should be a protocol warning. */ - log_warn(LD_REND, "Descriptor length is %i which exceeds " - "maximum rendezvous descriptor size of %i bytes.", + log_warn(LD_REND, "Descriptor length is %d which exceeds " + "maximum rendezvous descriptor size of %d bytes.", (int)(eos-desc), REND_DESC_MAX_SIZE); goto err; } diff --git a/src/test/test_addr.c b/src/test/test_addr.c index 20ffaa0..6db4ee2 100644 --- a/src/test/test_addr.c +++ b/src/test/test_addr.c @@ -477,9 +477,9 @@ test_addr_ip6_helpers(void) i = get_interface_address6(LOG_DEBUG, AF_INET6, &t2); #if 0 tor_inet_ntop(AF_INET, &t1.sa.sin_addr, buf, sizeof(buf)); - printf("\nv4 address: %s (family=%i)", buf, IN_FAMILY(&t1)); + printf("\nv4 address: %s (family=%d)", buf, IN_FAMILY(&t1)); tor_inet_ntop(AF_INET6, &t2.sa6.sin6_addr, buf, sizeof(buf)); - printf("\nv6 address: %s (family=%i)", buf, IN_FAMILY(&t2)); + printf("\nv6 address: %s (family=%d)", buf, IN_FAMILY(&t2)); #endif done:

1 0

[tor/master] Merge remote-tracking branch 'origin/maint-0.2.2'
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 4a7f979b54c1400190e037b818a241b92e40aabd Merge: 9eeb902 5cc322e Author: Nick Mathewson <nickm(a)torproject.org> Date: Tue Apr 19 12:41:01 2011 -0400 Merge remote-tracking branch 'origin/maint-0.2.2' src/common/address.c | 2 +- src/or/routerparse.c | 4 ++-- src/test/test_addr.c | 4 ++-- 3 files changed, 5 insertions(+), 5 deletions(-)

1 0

[tor/maint-0.2.2] Standardize our printf code on %d, not %i.
by nickm＠torproject.org 19 Apr '11

19 Apr '11

commit 5cc322e5472996fda6095e54de302c0f5c68ea92 Author: Nick Mathewson <nickm(a)torproject.org> Date: Tue Apr 19 12:40:26 2011 -0400 Standardize our printf code on %d, not %i. --- src/common/address.c | 2 +- src/or/routerparse.c | 4 ++-- src/test/test_addr.c | 4 ++-- 3 files changed, 5 insertions(+), 5 deletions(-) diff --git a/src/common/address.c b/src/common/address.c index adc0ef0..aff517c 100644 --- a/src/common/address.c +++ b/src/common/address.c @@ -604,7 +604,7 @@ tor_addr_parse_mask_ports(const char *s, tor_addr_t *addr_out, if (family == AF_INET6 && v4map) { if (bits > 32 && bits < 96) { /* Crazy */ log_warn(LD_GENERAL, - "Bad mask bits %i for V4-mapped V6 address; rejecting.", + "Bad mask bits %d for V4-mapped V6 address; rejecting.", bits); goto err; } diff --git a/src/or/routerparse.c b/src/or/routerparse.c index dd72eb6..ba29f05 100644 --- a/src/or/routerparse.c +++ b/src/or/routerparse.c @@ -4641,8 +4641,8 @@ rend_parse_v2_service_descriptor(rend_service_descriptor_t **parsed_out, if (eos-desc > REND_DESC_MAX_SIZE) { /* XXX023 If we are parsing this descriptor as a server, this * should be a protocol warning. */ - log_warn(LD_REND, "Descriptor length is %i which exceeds " - "maximum rendezvous descriptor size of %i bytes.", + log_warn(LD_REND, "Descriptor length is %d which exceeds " + "maximum rendezvous descriptor size of %d bytes.", (int)(eos-desc), REND_DESC_MAX_SIZE); goto err; } diff --git a/src/test/test_addr.c b/src/test/test_addr.c index 20ffaa0..6db4ee2 100644 --- a/src/test/test_addr.c +++ b/src/test/test_addr.c @@ -477,9 +477,9 @@ test_addr_ip6_helpers(void) i = get_interface_address6(LOG_DEBUG, AF_INET6, &t2); #if 0 tor_inet_ntop(AF_INET, &t1.sa.sin_addr, buf, sizeof(buf)); - printf("\nv4 address: %s (family=%i)", buf, IN_FAMILY(&t1)); + printf("\nv4 address: %s (family=%d)", buf, IN_FAMILY(&t1)); tor_inet_ntop(AF_INET6, &t2.sa6.sin6_addr, buf, sizeof(buf)); - printf("\nv6 address: %s (family=%i)", buf, IN_FAMILY(&t2)); + printf("\nv6 address: %s (family=%d)", buf, IN_FAMILY(&t2)); #endif done:

1 0