wolfganghuber / speedtest

Periodically measure your internet speed using speedtest command line interface and crontab, and visualize the data with R

title	author	output
Speedtest	Wolfgang Huber	html_document keep_md true

Install the speedtest command line interface

The tool is called speedtest-cli. The script also calls gdate, which homebrew provides in coreutils. The below shell commands are for Mac OS X and homebrew. Replace by your favourite package manager.

brew install speedtest-cli coreutils

Set up crontab to run it in regular intervals

This works on Unix-derived systems like Linux and Mac OS X. Edit the crontab using

env EDITOR=nano crontab -e

and add a line similar to:

 */5 * * * * /Users/whuber/svnco/speedtest/run-speedtest-cli.sh

This will run the shell script run-speedtest-cli.sh, which is provided in a separate file, every 5 minutes. Of course you'll have to adapt the file path to your local setup. Check the documentation of crontab if you want to run it at different times or intervals. Make sure the file is executable, with e.g.

chmod 755 /Users/whuber/svnco/speedtest/run-speedtest-cli.sh

Visualize the data

library("readr")
library("tidyr")
library("dplyr")
library("magrittr")
library("ggplot2")

Get the CSV header and read the CSV file

hosts = c("spinoza", "boltzmann")[1]
logfile = "/Users/whuber/Dropbox/speedtest/speedtest-%s.csv"
  
header = system2("speedtest-cli", args = c("--csv-header"), stdout = TRUE)  %>%
  strsplit(split = ",") %>% `[[`(1)

st = lapply(hosts, function(h) {
  read_csv(sprintf(logfile, h), col_names = FALSE) %>%
  `colnames<-`(header)  %>% 
  mutate(hostname = h)
}) %>% bind_rows 

stopifnot(all(is.finite(st$Timestamp)))
is_fin = is.finite(st$Download)
stopifnot(identical(is_fin, is.finite(st$Upload)),
          identical(is_fin, is.finite(st$Ping)))
with(st,
  table(ifelse(is.na(Download), "NA", ifelse(Download==0, "0", ">0")), 
        hostname))

##     hostname
##      spinoza
##   >0    1684
##   0        3
##   NA      21

negping = (!is.na(st$Ping) & (st$Ping <= 0))
if (any(negping)) {
  print(st[which(negping), ])
  st = filter(st, (!negping) | is.na(st$Ping))  
}

## # A tibble: 1 x 11
##   `Server ID` Sponsor `Server Name` Timestamp           Distance  Ping Download
##         <dbl> <chr>   <chr>         <dttm>                 <dbl> <dbl>    <dbl>
## 1       38469 Diamon… Limburg An D… 2020-12-27 12:59:02     115. -64.0   2.53e7
## # … with 4 more variables: Upload <dbl>, Share <lgl>, `IP Address` <chr>,
## #   hostname <chr>

variables = c("Download (Mb/s)", "Upload (Mb/s)", "log10 Ping (ms)")
st %<>% mutate(
  `Download (Mb/s)` = Download / 2^20,
  `Upload (Mb/s)` = Upload / 2^20,
  `log10 Ping (ms)` = log10(Ping))

Stratify by "IP Address", which indicates whether the measurement was taken in Unitymedia broadband, LTE Telekom or at work at EMBL.

The stopifnot call makes sure that all instances of IP Address that are not NA are one of the three addresses below. Subsequently we assume that the cases where IP Address is NA were at home, i.e. in the Unitymedia network.

servers = c(Unitymedia    = "176.199.211.22", 
            EMBL          = "194.94.44.220", 
            `Telekom-LTE` = "80.187.97.*")
st$network = rep(NA_character_, nrow(st))
for (nm in names(servers)) 
  st$network[ grep(servers[nm], st$`IP Address`) ] = nm
stopifnot(all(!is.na(st$network) | is.na(st$`IP Address`)))

Epochs. The running median line will be plotted separately for each epoch. Here the assumption is that consecutive measurements were taken about 5 minutes apart (i.e., well less than 7 minutes).

newepoch = (diff(st$Timestamp) > as.difftime(7,  units = "mins"))
st$epoch = as.factor(c(0L, cumsum(as.numeric(newepoch))))

Pivot to long format for plotting.

st %<>% pivot_longer(cols = all_of(variables))
st$name %<>% factor(levels = variables)

Smooth. Running median with window size +/- 12 min, i.e., in practice these are the 5 measurements t-10, t-5, t, t+5, t+10. The for loop and esp. computation of k is a bit clumsy and inefficient here; but it works.

delta = as.difftime(12,  units = "mins")
st$value_s = rep(NA_real_, nrow(st))

for (i in which(!is.na(st$network))) {
  k = which(with(st,  
      (abs(Timestamp - Timestamp[i]) <= delta) &
      (name == name[i]) &
      (network == network[i])))
  stopifnot(length(k)>=1)
  st$value_s[i] = median(st$value[k], na.rm = TRUE)    
}

makeplot = function(x, networks, from = "2021-01-07", to = "2021-12-31") {
  x = filter(x, (Timestamp >= from) & (Timestamp <= to) & (network %in% networks)) 
  ggplot(x, aes(x = Timestamp)) + 
    # Sadly, it seems that the below timezone setting is ignored by ggplot2
    scale_x_datetime(timezone = "CET") + 
    xlab("Time") + ylab("Value") + 
    geom_point(aes(y = ifelse(is.na(value), 0, value), 
                   col = network, shape = is.na(value), size = is.na(value))) + 
    geom_line(aes(y = ifelse(is.na(value_s), 0, value_s), group = epoch), col = "#777777", size = 0.4) + 
    facet_grid(rows = vars(name), scales = "free_y") +
    scale_shape_manual(values = c(`FALSE` = 16, `TRUE` = 4)) +
    scale_size_manual(values = c(`FALSE` = 0.75, `TRUE` = 1.5)) +
    scale_color_manual(values = c(Unitymedia = "black", `Telekom-LTE` = "magenta")) +
    theme(legend.position = "bottom") + 
    guides(shape = "none", size = "none")
 }
makeplot(st, networks = c("Unitymedia", "Telekom-LTE")[1])

#dev.copy(pdf, file = "speedtest.pdf", width = 16, height = 8)