Benchmarks

In yyjsonr: Fast 'JSON', 'NDJSON' and 'GeoJSON' Parser and Generator

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 6,
  fig.height = 4
)

library(yyjsonr)
library(bench)
library(ggplot2)
library(tidyr)
library(ggbeeswarm)
library(geojsonsf)
library(sf)

Benchmark overview

• Benchmarking was done an Apple M2 Silicon.
• Test-cases were drawn from other packages and other examples seen in the wild

Validate JSON String

json_str <- write_json_str(iris)

res00 <- bench::mark(
  jsonlite     = jsonlite::validate(json_str),
  jsonify      = jsonify::validate_json(json_str),
  yyjsonr      = yyjsonr::validate_json_str(json_str),
  check = TRUE
)

res00$benchmark <- 'Validate "iris" in JSON String'
knitr::kable(res00[,1:5])
plot(res00) + theme_bw() + theme(legend.position = 'none')

To JSON String

res01 <- bench::mark(
  jsonlite = jsonlite::toJSON(iris),
  jsonify  = jsonify::to_json(iris),
  yyjsonr  = yyjsonr::write_json_str(iris),
  check = FALSE
)

res01$benchmark <- 'Convert "iris" to JSON String'
knitr::kable(res01[,1:5])
plot(res01) + theme_bw() + theme(legend.position = 'none')

From JSON String

json_str <- write_json_str(iris)

res02 <- bench::mark(
  jsonlite     = jsonlite::fromJSON(json_str),
  jsonify      = jsonify::from_json(json_str),
  yyjsonr      = yyjsonr::read_json_str(json_str),
  check = TRUE
)

res02$benchmark <- 'Read "iris" from JSON String'
knitr::kable(res02[,1:5])
plot(res02) + theme_bw() + theme(legend.position = 'none')

From JSON raw vector

# a <- nanonext::ncurl("https://postman-echo.com/get", convert = FALSE)

# raw_data <- a$data
raw_data <- as.raw(c(0x7b, 0x0a, 0x20, 0x20, 0x22, 0x61, 0x72, 0x67, 0x73, 
0x22, 0x3a, 0x20, 0x7b, 0x7d, 0x2c, 0x0a, 0x20, 0x20, 0x22, 0x68, 
0x65, 0x61, 0x64, 0x65, 0x72, 0x73, 0x22, 0x3a, 0x20, 0x7b, 0x0a, 
0x20, 0x20, 0x20, 0x20, 0x22, 0x78, 0x2d, 0x66, 0x6f, 0x72, 0x77, 
0x61, 0x72, 0x64, 0x65, 0x64, 0x2d, 0x70, 0x72, 0x6f, 0x74, 0x6f, 
0x22, 0x3a, 0x20, 0x22, 0x68, 0x74, 0x74, 0x70, 0x73, 0x22, 0x2c, 
0x0a, 0x20, 0x20, 0x20, 0x20, 0x22, 0x78, 0x2d, 0x66, 0x6f, 0x72, 
0x77, 0x61, 0x72, 0x64, 0x65, 0x64, 0x2d, 0x70, 0x6f, 0x72, 0x74, 
0x22, 0x3a, 0x20, 0x22, 0x34, 0x34, 0x33, 0x22, 0x2c, 0x0a, 0x20, 
0x20, 0x20, 0x20, 0x22, 0x68, 0x6f, 0x73, 0x74, 0x22, 0x3a, 0x20, 
0x22, 0x70, 0x6f, 0x73, 0x74, 0x6d, 0x61, 0x6e, 0x2d, 0x65, 0x63, 
0x68, 0x6f, 0x2e, 0x63, 0x6f, 0x6d, 0x22, 0x2c, 0x0a, 0x20, 0x20, 
0x20, 0x20, 0x22, 0x78, 0x2d, 0x61, 0x6d, 0x7a, 0x6e, 0x2d, 0x74, 
0x72, 0x61, 0x63, 0x65, 0x2d, 0x69, 0x64, 0x22, 0x3a, 0x20, 0x22, 
0x52, 0x6f, 0x6f, 0x74, 0x3d, 0x31, 0x2d, 0x36, 0x35, 0x33, 0x62, 
0x61, 0x33, 0x38, 0x65, 0x2d, 0x35, 0x65, 0x65, 0x66, 0x32, 0x39, 
0x64, 0x38, 0x30, 0x61, 0x35, 0x63, 0x65, 0x62, 0x32, 0x30, 0x33, 
0x65, 0x36, 0x64, 0x32, 0x64, 0x35, 0x61, 0x22, 0x0a, 0x20, 0x20, 
0x7d, 0x2c, 0x0a, 0x20, 0x20, 0x22, 0x75, 0x72, 0x6c, 0x22, 0x3a, 
0x20, 0x22, 0x68, 0x74, 0x74, 0x70, 0x73, 0x3a, 0x2f, 0x2f, 0x70, 
0x6f, 0x73, 0x74, 0x6d, 0x61, 0x6e, 0x2d, 0x65, 0x63, 0x68, 0x6f, 
0x2e, 0x63, 0x6f, 0x6d, 0x2f, 0x67, 0x65, 0x74, 0x22, 0x0a, 0x7d
))

res03 <- bench::mark(
  jsonlite     = jsonlite::fromJSON(rawConnection(raw_data)),
  yyjsonr      = yyjsonr::read_json_raw(raw_data),
  check = FALSE
)

res03$benchmark <- 'From JSON Raw Vector'
knitr::kable(res03[,1:5])
plot(res03) + theme_bw() + theme(legend.position = 'none')

To JSON File

json_file <- tempfile()

res04 <- bench::mark(
  jsonlite = jsonlite::write_json(iris, json_file),
  yyjsonr  = yyjsonr::write_json_file(iris, json_file),
  check = FALSE
)

res04$benchmark <- 'Write "iris" to JSON File'
knitr::kable(res04[, 1:5])
plot(res04) + theme_bw() + theme(legend.position = 'none')

From JSON File

json_file <- tempfile()
jsonlite::write_json(iris, json_file)

res05 <- bench::mark(
  jsonlite     = jsonlite::fromJSON(file(json_file)), 
  jsonify      = jsonify::from_json(json_file),
  yyjsonr      = yyjsonr::read_json_file(json_file),
  check = TRUE
)

res05$benchmark <- 'Read "iris" from JSON File'
knitr::kable(res05[, 1:5])
plot(res05) + theme_bw() + theme(legend.position = 'none')

Write modest data.frame to string (10 thousand rows)

n <- 1e5
df <- data.frame(
  id = 1:n
  , value = sample(letters, size = n, replace = T)
  , val2 = rnorm(n = n)
  , log = sample(c(T,F), size = n, replace = T)
  , stringsAsFactors = FALSE
)

res10 <- bench::mark(
  jsonlite = jsonlite::toJSON( df ),
  jsonify  = jsonify::to_json( df ),
  yyjsonr  = yyjsonr::write_json_str( df ),
  check = FALSE
)

res10$benchmark <- '10k row data.frame to file'
knitr::kable(res10[,1:5])
plot(res10) + theme_bw() + theme(legend.position = 'none')

Read modest data.frame from string (10 thousand rows)

str <-  jsonlite::toJSON( df )

res11 <- bench::mark(
  jsonlite = jsonlite::fromJSON( str ),
  jsonify  = jsonify::from_json( str ),
  # rcppsimdjson = RcppSimdJson::fparse( str ),
  yyjsonr  = yyjsonr::read_json_str( str ),
  check = TRUE
)

res11$benchmark <- '10k row data.frame from string'
knitr::kable(res11[,1:5])
plot(res11) + theme_bw() + theme(legend.position = 'none')

Summary

library(dplyr)
plot_df <- bind_rows(
  res00, res01, res02, res03, res04, res05,
  # res13, res14, # geojson
  #res06, res07, res08, 
  # res09, 
  res10, res11 #, res12,
)

plot_df$benchmark <- factor(
  plot_df$benchmark, 
  levels = unique(plot_df$benchmark)
)

plot_df <- plot_df %>% 
  mutate(
    package = as.character(expression),
    iters   = `itr/sec`,
    speed   = iters
  ) %>%
  select(benchmark, package, iters, speed)

plot_df <- plot_df %>%
  group_by(benchmark) %>%
  mutate(
    ref_speed = speed[which(package %in% c('jsonlite', 'geojsonsf'))],
    speed = speed / ref_speed
  ) %>%
  ungroup()

ggplot(plot_df) + 
  geom_col(aes(package, speed, fill = package), 
           position = position_dodge2(preserve = "single")) + 
  facet_wrap(~benchmark, scales = 'free_y', nrow = 2) + 
  theme_bw(15) + 
  theme(legend.position = 'none') + 
  scale_fill_manual(values = c(rep(grey(0.5), 2), 'dodgerblue3')) +
  geom_hline(yintercept = 1, color = 'red', alpha = 0.5, linetype = 2) + 
  labs(
    x = NULL,
    y = "Factor speed increase over reference implementation",
    title = "Speed-up compared to reference implementation",
    subtitle = "Red line indicates reference implementation {jsonlite}"
  ) + 
  scale_y_continuous(breaks = scales::pretty_breaks())

if (FALSE) {
  ggsave("./man/figures/benchmark-summary.png", width = 12, height = 7)
  # saveRDS(plot_df, "man/benchmark/cache-df-types.rds")
}

yyjsonr documentation built on May 29, 2024, 3:01 a.m.