CAF memory consumption

memory-consumption.R

library(dplyr)
library(tidyr)
library(ggplot2)

parse_measurement_filename <- function(path) {
  filename <- strsplit(basename(path), "\\.")[[1]][1]
  s <- strsplit(filename, "_")[[1]]
  list(Cores=as.factor(s[1]), Run=as.factor(s[3]), Type=as.factor(s[5]))
}

parse_runtime_filename <- function(path) {
  filename <- strsplit(basename(path), "\\.")[[1]][1]
  s <- strsplit(filename, "_")[[1]]
  list(Cores=as.factor(s[1]), Type=as.factor(s[3]))
}

ingest_measurement <- function(path) {
  l <- parse_measurement_filename(path)
  x <- read.table(path, col.names=c("Time", "Memory"))
  x$Cores <- rep(l$Cores, nrow(x))
  x$Run <- rep(l$Run, nrow(x))
  x$Type <- rep(l$Type, nrow(x))
  x
}

ingest_runtime <- function(path) {
  l <- parse_runtime_filename(path)
  x <- read.table(path, col.names=c("Runtime"))
  x$Cores <- rep(l$Cores, nrow(x))
  x$Type <- rep(l$Type, nrow(x))
  x
}

# Import all files.
path <- "data"
files <- list.files(path, ".*_run_.*\\.txt", full.names=TRUE)
measurements <- do.call(rbind, lapply(files, ingest_measurement))
files <- list.files(path, ".*_runtime_.*\\.txt", full.names=TRUE)
runtimes <- do.call(rbind, lapply(files, ingest_runtime))

# Global theme options
theme_set(theme_bw())
theme_update(legend.key=element_rect(color="white"))
options(scipen=100000)

# Boxplot workaround: ensure their outlier have the same color as the contour.
update_geom_defaults("point", list(colour=NULL))

# Plot a boxplot per core for all values across all runs.
p.box <- measurements %>%
  ggplot(aes(x=Cores, y=Memory, color=Type)) + 
    geom_boxplot() #+ facet_wrap(~Type)

# Plot the maximum memory consumption across all runs.
p.box.max <- measurements %>% 
  group_by(Cores, Type, Run) %>% 
  summarize(Memory=max(Memory)) %>% 
  ggplot(aes(x=Cores, y=Memory, color=Type)) + 
    geom_boxplot()

# Plot the memory consumption over time for each run.
p.time.points <- measurements %>%
  ggplot(aes(x=Time, y=Memory, color=Cores)) + 
    geom_point(size=1) +
    facet_wrap(~Type)

# Same as above, but connect points from one run.
cores <- length(levels(measurements$Cores))
p.time.lines <- measurements %>%
  unite(c_r, Cores, Run, remove=FALSE) %>%
  ggplot(aes(x=Time, y=Memory, group=c_r, color=Cores)) + 
    geom_line() + 
    geom_point(aes(fill=Cores), size=1, shape=21) + 
    scale_fill_manual(values=rep("white", cores)) +
    facet_wrap(~Type)

# Same as above, but use one shape per run and lower alpha of lines.
runs <- length(levels(measurements$Run))
p.time.shapes <- measurements %>%
  unite(c_r, Cores, Run, remove=FALSE) %>%
  ggplot(aes(x=Time, y=Memory, group=c_r, color=Cores)) + 
    geom_line(alpha=.5) + 
    geom_point(aes(shape=Run), size=1) + 
    scale_shape_manual(values=1:runs) +
    facet_wrap(~Type)

p.runtime.box <- runtimes %>%
  ggplot(aes(x=Cores, y=Runtime, color=Type)) +
    geom_boxplot()

p.runtime.ecdf <- runtimes %>% 
  ggplot(aes(x=Runtime, color=Cores)) + 
    stat_ecdf() + 
    ylab("ECDF") + 
    facet_wrap(~Type, ncol=1)

p.runtime.ecdf2 <- runtimes %>% 
  ggplot(aes(x=Runtime, color=Type)) + 
    stat_ecdf() + 
    ylab("ECDF") + 
    facet_wrap(~Cores, ncol=1)

# Save boxplots plots
ggsave("measure-box.pdf", p.box)
ggsave("measure-box-max.pdf", p.box.max)
ggsave("runtime-box.pdf", p.runtime.box)

# Undo boxplot outlier workaround
update_geom_defaults("point", list(colour="black"))

# Save other plots
ggsave("measure-time-points.pdf", p.time.points)
ggsave("measure-time-lines.pdf", p.time.lines)
ggsave("measure-time-shapes.pdf", p.time.shapes)
ggsave("runtime-ecdf.pdf", p.runtime.ecdf)
ggsave("runtime-ecdf2.pdf", p.runtime.ecdf2)