The performance test builds random targetings (varying from 10-5000), then queries it with random bids. See the JS file for more details.
# Generate test data
setwd("~/src/incrowd/bidder")
# system('node test/targeting_collection_performance.js')
# Read basic data
basic <- read.csv("targeting_collection.csv")
head(basic)
## targetings bids time ms.per.bid
## 1 10 1000 96 0.096
## 2 20 1000 78 0.078
## 3 50 1000 100 0.100
## 4 100 1000 132 0.132
## 5 250 1000 242 0.242
## 6 500 1000 411 0.411
# Detailed profiling data:
data <- read.csv("targeting_collection_all.csv")
head(data)
## targetings bids microtime control found linear
## 1 10 1000 197 5 0 214
## 2 10 1000 64 1 0 74
## 3 10 1000 29 0 0 38
## 4 10 1000 88 1 0 97
## 5 10 1000 30 0 0 39
## 6 10 1000 82 0 0 90
# INVALID NOW Basic performance plot. Looks linear in targetings:# plot(targetings~ms.per.bid, basic)
Performance profile of individual find()'s, with control (variable assignment) for comparison.
library(ggplot2)
library(reshape)
(ggplot(melt(data[, c("targetings", "bids", "microtime", "control")],
id = c("targetings", "bids"))) + aes(value, color = factor(targetings)) +
geom_density() + scale_x_log10(name = "microseconds", breaks = c(1, 10,
100, 200, 500, 1000, 2000, 5000)) + facet_wrap(~variable, nrow = 2) + opts(title = "Latency distribution for different #'s of targetings versus control"))
Versus linear:
library(ggplot2)
library(reshape)
(ggplot(melt(data[, c("targetings", "bids", "microtime", "linear")],
id = c("targetings", "bids"))) + aes(value, color = factor(targetings)) +
geom_density() + scale_x_log10(name = "microseconds", breaks = c(1, 10,
100, 200, 500, 1000, 2000, 5000, 10000)) + facet_wrap(~variable, nrow = 2) +
opts(title = "Latency distribution for different #'s of targetings versus linear"))
Compare mean latency vs linear:
melted <- melt(data[, c("targetings", "bids", "microtime", "linear")],
id = c("targetings", "bids"))
aggregated <- aggregate(value ~ targetings + variable, melted, mean)
(ggplot(aggregated) + aes(targetings, value/1000, color = variable) +
geom_line() + scale_y_continuous(name = "mean latency (ms)") + scale_x_continuous(name = "number of targetings") +
opts(title = "Mean find() latency luis vs linear"))
Build time:
setwd("~/src/incrowd/bidder")
data <- read.csv("targeting_collection_build.csv", header = T)
(ggplot(data) + aes(size, time) + geom_point() + geom_smooth(method = lm,
fullrange = T) + scale_x_continuous(limits = c(0, 20000)) + scale_y_continuous(limits = c(0,
12000)) + opts(title = "TargetingCollection build speed for various random workloads"))
