library(ggplot2) library(sqldf) library(plyr) library(cowplot) thunder_ms <- read.csv("thunder_configure_19.csv") thunder_ms <- sqldf("select run,ident,jmax,links,latency, CAST(latency as real) / 1000. as lat_ms from thunder_ms") thunder_ms$links <- as.factor(thunder_ms$links) thunder_ms$jmax <- as.factor(thunder_ms$jmax) v1 <- ggplot(data = thunder_ms, aes(x = jmax, y=lat_ms, fill=links)) + #geom_violin() + geom_boxplot(outlier.size=0.1) + geom_hline(yintercept=800, color = "red", linetype="dashed") + geom_hline(yintercept=400, color = "DarkGreen") + annotate("text", x=7, y=550, label= "high", color="DarkGreen") + annotate("text", x=7, y=1000, label= "acceptable", color="red") + coord_cartesian(xlim = c(1, 7)) + #ylim(0,1000) + scale_fill_grey() + scale_y_log10() + ylab("latency (ms)") + xlab("max allowed jitter") + theme_classic() #v1 + ggsave("thunder_configure_ms.png", dpi=300, dev='png', height=5, width=15, units="cm") thunder_rcv <- sqldf("select run,jmax,links,(9900-COUNT(latency)*1.0)/9900 as dlv from thunder_ms group by jmax,links,run") thunder_rcv$jmax <- as.factor(thunder_rcv$jmax) thunder_rcv$links <- as.factor(thunder_rcv$links) v2 <- ggplot(data = thunder_rcv, aes(x = jmax, y=dlv, fill=links)) + geom_boxplot(outlier.size=0.1) + scale_y_continuous(labels = scales::percent) + scale_fill_grey() + ylab("dropped packets") + xlab("max allowed jitter") + theme_classic() thunder_bw <- read.csv("thunder_configure_19_bw.csv") thunder_bw <- sqldf("select run,jmax,links,udp_sent,udp_rcv,cells_sent,cells_rcv,1.0*cells_sent/udp_sent as sent_ratio,1.0*cells_rcv/udp_rcv as rcv_ratio from thunder_bw where udp_sent > 4000") thunder_bw$jmax <- as.factor(thunder_bw$jmax) thunder_bw$links <- as.factor(thunder_bw$links) v3 <- ggplot(data = thunder_bw, aes(x = jmax, y=sent_ratio, fill=links)) + geom_boxplot(outlier.size=0.1) + #scale_y_log10() + #scale_y_log10(labels = scales::percent) + scale_fill_grey() + ylab("bandwidth ratio") + xlab("max allowed jitter") + theme_classic() t1 <- plot_grid(v1, v2, v3, labels = c('A', 'B', 'C'), ncol=1) t1 + ggsave("thunder_configure.png", dpi=300, dev='png', height=15, width=15, units="cm") thunder_links <- read.csv("thunder_configure_16_links.csv") links_down_at_least_once <- sqldf("select run,xp_time,link_id, COUNT(status) as downcount, SUM(delta) as elapsed from thunder_links where status='down' group by run,link_id,xp_time") links_down_at_least_once2 <- sqldf("select row_number () OVER (PARTITION BY run ORDER BY elapsed DESC) sorting,run,link_id,downcount,elapsed,xp_time,1.0*elapsed/xp_time down_ratio from links_down_at_least_once") links_down_at_least_once2$sorting <- as.factor(links_down_at_least_once2$sorting) v4 <- ggplot(data = links_down_at_least_once2, aes(x = sorting, y=down_ratio)) + #geom_violin() + #geom_boxplot(width=0.2) + #scale_y_log10() + #scale_y_log10(labels = scales::percent) + geom_bar(stat="summary") + #scale_y_log10() + scale_y_continuous(labels = scales::percent) + scale_fill_grey() + ylab("Cumulated downtime") + xlab("Sorted links") + theme_classic() downtime <- sqldf( " select sorting,1.0*duration/1000 as dur from thunder_links as tl inner join links_down_at_least_once2 as l2 on tl.run = l2.run and tl.link_id = l2.link_id where will_change='True' and status='down'") v5 <- ggplot(data = downtime, aes(x=sorting, y=dur)) + geom_violin() + geom_boxplot(width=0.1, outlier.shape = NA) + scale_y_log10() + ylab("Downtime (in sec)") + xlab("Sorted links") + theme_classic() gobal_links <- sqldf( " select ts,run,durations_global,8-COUNT(status) as down_link_count,xp_time from thunder_links where will_change_global='True' and status='up' and durations_global > 0 group by ts,run,durations_global ") down_group_ratio <- sqldf( " select run,1.0 * SUM(durations_global)/xp_time as down_ratio, down_link_count from gobal_links where down_link_count >= 0 group by run,down_link_count,xp_time ") down_group_ratio$down_link_count <- as.factor(down_group_ratio$down_link_count) v6 <- ggplot(data = down_group_ratio, aes(x=down_link_count, y=down_ratio)) + geom_bar(stat="summary") + #scale_y_log10() + scale_y_continuous(labels = scales::percent) + ylab("Cumulated downtime") + xlab("Number of links down at once") + theme_classic() downtime_group <- sqldf( " select down_link_count, 1.0*durations_global/1000 as dur from gobal_links where down_link_count >= 0 ") downtime_group$down_link_count <- as.factor(downtime_group$down_link_count) v7 <- ggplot(data = downtime_group, aes(x=down_link_count, y=dur)) + geom_violin() + geom_boxplot(width=0.1, outlier.shape = NA) + scale_y_log10() + ylab("Downtime (in sec)") + xlab("Number of links down at once") + theme_classic() t2 <- plot_grid(v4, v5, v6, v7, labels = c('A', 'B', 'C', 'D'), ncol=2) t2 + ggsave("thunder_links.png", dpi=300, dev='png', height=12, width=15, units="cm") latency_evol <- sqldf( " select sorting,lat_ms,ident,tm.jmax,tm.links from thunder_ms as tm, (select run,jmax,links,row_number () OVER (ORDER BY links DESC) sorting from thunder_ms group by run,jmax,links ORDER BY links DESC limit 0,1) as sel_run where tm.run = sel_run.run and tm.jmax = sel_run.jmax and tm.links = sel_run.links ") latency_evol$sorting <- as.factor(latency_evol$sorting) v8 <- ggplot(data=latency_evol, aes(x=ident,y=lat_ms)) + geom_line() + xlab("Packet identifier") + ylab("Latency (ms)") + theme_classic() thunder_drop <- read.csv("thunder_configure_16_drop.csv") thunder_drop_2 <- sqldf("select run, packet_range, 1.0*count / 990 as packet_ratio, row_number() OVER (partition by packet_range order by run) sorting from thunder_drop where run LIKE '%-26' ") #cats <- c("0-989","990-1979","1980-2969","2970-3959","3960-4949","4950-5939","5940-6929","6930-7919","7920-8909","8910-9899") thunder_drop_2$packet_range <- as.factor(thunder_drop_2$packet_range) thunder_drop_2$sorting <- as.factor(thunder_drop_2$sorting) thunder_drop_2$packet_range <- factor( mapvalues(thunder_drop_2$packet_range, cats, cats), levels = cats, ordered = TRUE) v9 <- ggplot(data = thunder_drop_2, aes(x=packet_range, y=packet_ratio,fill=sorting)) + geom_bar(stat="summary",position = "dodge") + #grom_bar() + #scale_y_log10() + scale_y_continuous(labels = scales::percent) + ylab("Packets dropped") + xlab("Packet identifier") + labs(fill="Run") + scale_fill_grey() + theme_classic() + theme(axis.text.x = element_text(angle = 45, hjust = 1), legend.key.size = unit(0.2, "cm")) thunder_drop_burst <- read.csv("thunder_configure_16_drop_burst.csv") tdb_ag <- sqldf("select run,count,COUNT(count) as oc from thunder_drop_burst where run LIKE '%-24' group by run,count") tdb_ag_2 <- sqldf( " select td.run as r, count, oc, total, 1.0 * oc / total as oc_ratio, row_number() OVER (partition by count order by td.run) as sorting from tdb_ag as td, (select run,SUM(oc) as total from tdb_ag group by run) as ag where td.run = ag.run ") tdb_ag_2$sorting <- as.factor(tdb_ag_2$sorting) tdb_ag_2$count <- as.factor(tdb_ag_2$count) v10 <- ggplot(data = tdb_ag_2, aes(x=count, y=oc_ratio)) + #geom_bar(stat="summary",position = "dodge") + #scale_y_log10() + geom_violin(scale='width') + geom_boxplot(width=0.1, outlier.shape=NA) + scale_y_continuous(labels = scales::percent) + ylab("% observed drops") + xlab("Packets lost during the drop") + scale_fill_grey() + theme_classic() thunder_red <- read.csv("thunder_configure_16_red.csv") tred <- sqldf( " select tr.run as r, delivered_at_once, 1.0 * occur / total as occur_ratio, occur, total, row_number() OVER (partition by delivered_at_once order by tr.run) as sorting from thunder_red tr, (select run,SUM(occur) as total from thunder_red group by run) as ag WHERE tr.run LIKE '%-26' and tr.run = ag.run ") tred$sorting <- as.factor(tred$sorting) tred$delivered_at_once <- as.factor(tred$delivered_at_once) v11 <- ggplot(data = tred, aes(x=delivered_at_once, y=occur_ratio)) + #geom_bar(stat="summary",position = "dodge") + geom_violin(scale='width') + xlab('Fresh packets per cell') + ylab('% of received cells') + scale_y_continuous(labels = scales::percent) + geom_boxplot(width=0.1, outlier.shape=NA) + theme_classic() t3 <- plot_grid(v8, v9, v10, v11, labels = c('A', 'B', 'C', 'D'), ncol=2) t3 + ggsave("thunder_packets.png", dpi=300, dev='png', height=12, width=15, units="cm") tor_multi_lat <- read.csv("tor_just_many_latencies.csv") tor_drop <- sqldf("select run,conf,1.0*MAX(ident)/33 as last_one from tor_multi_lat group by run,conf") v12 <- ggplot(data=tor_drop,aes(x=last_one)) + stat_ecdf(pad = FALSE) + ylab("% broken links") + coord_cartesian(xlim = c(0, 300), ylim = c(0,0.5)) + scale_y_continuous(labels = scales::percent) + xlab("Elapsed time (sec)") + theme_classic() v12 + ggsave("broken.png", dpi=300, dev='png', height=5, width=15, units="cm") library(dplyr) library(purrr) library(tidyr) tor_lat_stack <- tor_multi_lat %>% dplyr::group_by(run,conf) %>% dplyr::summarise( id = paste(first(run),first(conf)), min = min(latency), max = max(latency), max_sort = max(latency), q25 = quantile(latency,0.25), median = median(latency), median_sort = median(latency), q75 = quantile(latency,0.75), q95 = quantile(latency,0.95), q99 = quantile(latency,0.99) ) tor_lat_stack <- gather(tor_lat_stack, 'min', 'max', 'q25', 'median', 'q75', 'q95', 'q99', key="quantile_name", value="quantile_value") ggplot(tor_lat_stack, aes(x=id,y=quantile_value,fill=quantile_name)) + ylim(0,1000000) + geom_bar(stat="identity", position="dodge")