I am developing an Android application that shows the CPU load per core and memory consumption. For CPU load I am reading /proc/stat and for memory -> /proc/meminfo. However I see that the number of CPU cores in /proc/stat is changing during subsequent reading of the file.
cpu 230599 10622 84595 1892023 8236 16 285 0 0 0
cpu0 138005 7992 58080 1738918 6407 16 278 0 0 0
intr 9136791 0 0 0 0 0 0 0 0 0 0 0 0 0 0 9601 0 0 0 0 0 0 .......
ctxt 16904510
btime 1394641996
processes 16919
procs_running 2
procs_blocked 0
softirq 1688530 407 706934 422 1558 407 407 92978 324500 1267 559650
and after 5seconds it becomes:
cpu 230772 10623 84671 1890801 8236 16 286 0 0 0
cpu0 138104 7993 58126 1739267 6407 16 279 0 0 0
cpu1 92668 2630 26545 151534 1829 0 7 0 0 0
intr 9144729 0 0 0 0 0 0 0 0 0 0 0 0 0 0 9601 0 0 0 0 0 0 ........
ctxt 16923744
btime 1394641996
processes 16946
procs_running 2
procs_blocked 0
softirq 1690205 407 707396 422 1558 407 407 93311 324790 1267 560240
Does it mean that cpu cores are sleeping in some cases?
I know this is old, but there does not seem to be an answer for this, and I've been searching for a solution as well.
What I figured, is that if I'm recording each core named "cpu" that follows a number. and the cpus go from 0-3. And if I'm reading the cores in order. Then if .readline() of /proc/stat returns a string that does NOT contain cpu, than that core must not be working, and is offline. So, in theory, it would be at zero percent usage. So, return 0.
*** Completed answer with code, see below ***
Heres some code, in case what I said didn't make sense, mine is based on this: Get Memory Usage in Android
and here is how I found a new calculation that gave me more accurate representation of core readings: How to get total cpu usage in Linux using C++
First, here is some of my CPU function that displays a string after these loops and stuff to the user. I'm posting this so you have a better understanding of i, and what my code means
float[] coreValues = new float[10];
//get how many cores there are from function
int numCores = getNumCores();
for(byte i = 0; i < numCores; i++)
{
coreValues[i] = readCore(i);
}
getNumCores can be found here, I will not be posting since I feel like the person who did should get the credit for it: How can you detect a dual-core cpu on an Android device from code?
And, lastly, here's my code, I hope it makes sense, and I offered a lot of comments.
//for multi core value
private float readCore(int i)
{
/*
* how to calculate multicore
* this function reads the bytes from a logging file in the android system (/proc/stat for cpu values)
* then puts the line into a string
* then spilts up each individual part into an array
* then(since he know which part represents what) we are able to determine each cpu total and work
* then combine it together to get a single float for overall cpu usage
*/
try {
RandomAccessFile reader = new RandomAccessFile("/proc/stat", "r");
//skip to the line we need
for(int ii = 0; ii < i + 1; ++ii)
{
reader.readLine();
}
String load = reader.readLine();
//cores will eventually go offline, and if it does, then it is at 0% because it is not being
//used. so we need to do check if the line we got contains cpu, if not, then this core = 0
if(load.contains("cpu"))
{
String[] toks = load.split(" ");
//we are recording the work being used by the user and system(work) and the total info
//of cpu stuff (total)
//https://stackoverflow.com/questions/3017162/how-to-get-total-cpu-usage-in-linux-c/3017438#3017438
long work1 = Long.parseLong(toks[1])+ Long.parseLong(toks[2]) + Long.parseLong(toks[3]);
long total1 = Long.parseLong(toks[1])+ Long.parseLong(toks[2]) + Long.parseLong(toks[3]) +
Long.parseLong(toks[4]) + Long.parseLong(toks[5])
+ Long.parseLong(toks[6]) + Long.parseLong(toks[7]) + Long.parseLong(toks[8]);
try
{
//short sleep time = less accurate. But android devices typically don't have more than
//4 cores, and I'n my app, I run this all in a second. So, I need it a bit shorter
Thread.sleep(200);
}
catch (Exception e) {}
reader.seek(0);
//skip to the line we need
for(int ii = 0; ii < i + 1; ++ii)
{
reader.readLine();
}
load = reader.readLine();
//cores will eventually go offline, and if it does, then it is at 0% because it is not being
//used. so we need to do check if the line we got contains cpu, if not, then this core = 0%
if(load.contains("cpu"))
{
reader.close();
toks = load.split(" ");
long work2 = Long.parseLong(toks[1])+ Long.parseLong(toks[2]) + Long.parseLong(toks[3]);
long total2 = Long.parseLong(toks[1])+ Long.parseLong(toks[2]) + Long.parseLong(toks[3]) +
Long.parseLong(toks[4]) + Long.parseLong(toks[5])
+ Long.parseLong(toks[6]) + Long.parseLong(toks[7]) + Long.parseLong(toks[8]);
//here we find the change in user work and total info, and divide by one another to get our total
//seems to be accurate need to test on quad core
//https://stackoverflow.com/questions/3017162/how-to-get-total-cpu-usage-in-linux-c/3017438#3017438
return (float)(work2 - work1) / ((total2 - total1));
}
else
{
reader.close();
return 0;
}
}
else
{
reader.close();
return 0;
}
}
catch (IOException ex)
{
ex.printStackTrace();
}
return 0;
}
And, as a last note, my readCore function will return a value of 0.0 - 1.0, you need to multiply by 100 to get a percentage.
EDIT As requested from the comments below, Android documentation: "While active, CPUs can be brought online or put offline, change clock speeds and associated voltages (possibly also affecting memory bus speeds and other system core power state), and can enter lower power idle states while in the kernel idle loop. Not only are these different CPU power states measured for the power profile, it may be necessary to avoid the power draw variance when measuring other parameters."