mpetri / mem_monitor

A simple C++11 memory monitor

Memory Monitor

mem_monitor is self contained header file containing a C++11 memory monitor.

to monitor memory usage, the monitor objects creates a thread which periodically reads /proc/self/status and records values for:

Metric	Description

     pid | The process ID
     VmPeak | Peak virtual memory size
     VmSize | Virtual memory size
     VmHWM | Peak resident set size ("high water mark")
     VmRSS | Resident set size
     VmData | Size of data segments
     VmPTE | Page table entries size

Usage

Here an example of how the class can be used:

#include "mem_monitor.hpp"

int main(int argc, char const *argv[])
{
	mem_monitor mm("mem-mon-out.csv"); // poll every 50 milliseconds

	mm.event("vector init"); // optional event
	std::vector<uint64_t> vec(50000);
	/* do more stuff */
	mm.event("sort vector"); // optional event
	std::sort(vec.begin(),vec.end());
}

the information are periodically written to the output file and flushed once the object is destroyed.

Additionally, the granularity with which the monitoring thread polls the /proc/self/status file can be passed to the constructor:

mem_monitor mm("mem-mon-out.csv",std::chrono::milliseconds(5));

to use the memory monitor, your program has to be linked to the pthread and rt library and the c++11 flags:

g++ -std=c++11 -o a.out example.cpp -l pthread -l rt

Note: The object does not need to be created at the beginning of main(). The monitor can also be used to track only certain aspects of the executing program.

Output

The output produced by the class consist of a CSV file:

time_ms	pid	VmPeak	VmSize	VmHWM	VmRSS	VmData	VmPTE	event
1	12382	159952896	92844032	1163264	1163264	75800576	57344	"vector init"
51	12382	159952896	114884608	22253568	21483520	97841152	110592	"vector init"
101	12382	159952896	139624448	44494848	44494848	122580992	151552	"vector init"
151	12382	160440320	160440320	67743744	67743744	143396864	196608	"sort vector"
201	12382	191279104	184373248	91262976	84426752	167329792	237568	"sort vector"
251	12382	201809920	201809920	103350272	103350272	184766464	270336	"sort vector"
301	12382	220192768	220192768	122814464	122814464	203149312	311296	"sort vector"

where time_ms corresponds to the number of milliseconds since the creation of the object.

Visualization

The output can be visualized, for example, with the following R script:

library(ggplot2)
library(reshape2)
library(sitools)
library(scales)

data <- read.csv(file="res-mon.csv",sep=";",header=TRUE);
tdata <- melt(data,id=c("time_ms","event"))
tdata <- subset(tdata,variable!="pid")
dup <-tdata[!duplicated(tdata$event),]

plot <- ggplot(tdata,aes(time_ms,value,color=variable))
plot <- plot + geom_line(size=1)
#plot <- plot + scale_y_log10(expand=c(0,0),labels=f2si,breaks = trans_breaks("log10", function(x) 10^x),"Memory Usage [Byte]")
plot <- plot + scale_y_continuous(expand=c(0,0),labels=f2si,"Memory Usage [Byte]")
#plot <- plot + annotation_logticks(sides = "lr")
plot <- plot + scale_x_continuous(expand=c(0,0),labels=f2si,"Time [Millseconds]")
plot <- plot + scale_linetype(name="Events") + scale_color_discrete(name="Metric")
plot <- plot + geom_vline(data=dup,aes(xintercept = time_ms,linetype=event),show_guide=TRUE)
plot <- plot + theme_grey() + ggtitle("Memory Usage over Time")
ggsave(plot,file="memory_usage.png")

which produces a graph such as:

License

LGPLv3