How could I generate steady CPU load in C#, lower than 100% for a certain time? I would also like to be able to change the load amount after a certain period of time. How do you recommend to generate usage spikes for a very short time?
First off, you have to understand that CPU usage is always an average over a certain time. At any given time, the CPU is either working or it is not. The CPU is never 40% working.
We can, however, simulate a 40% load over say a second by having the CPU work for 0.4 seconds and sleep 0.6 seconds. That gives an average utilization of 40% over that second.
Cutting it down to smaller than one second, say 100 millisecond chunks should give even more stable utilization.
The following method will take an argument that is desired utilization and then utilize a single CPU/core to that degree:
public static void ConsumeCPU(int percentage)
{
if (percentage < 0 || percentage > 100)
throw new ArgumentException("percentage");
Stopwatch watch = new Stopwatch();
watch.Start();
while (true)
{
// Make the loop go on for "percentage" milliseconds then sleep the
// remaining percentage milliseconds. So 40% utilization means work 40ms and sleep 60ms
if (watch.ElapsedMilliseconds > percentage)
{
Thread.Sleep(100 - percentage);
watch.Reset();
watch.Start();
}
}
}
I'm using a stopwatch here because it is more accurate than the the TickCount property, but you could likewise use that and use subtraction to check if you've run long enough.
Two things to keep in mind:
on multi core systems, you will have to spawn one thread for each core. Otherwise, you'll see only one CPU/core being exercised giving roughly "percentage/number-of-cores" utilization.
Thread.Sleep is not very accurate. It will never guarantee times exactly to the millisecond so you will see some variations in your results
To answer your second question, about changing the utilization after a certain time, I suggest you run this method on one or more threads (depending on number of cores) and then when you want to change utilization you just stop those threads and spawn new ones with the new percentage values. That way, you don't have to implement thread communication to change percentage of a running thread.
Just in add of the Isak response, I let here a simple implementation for multicore:
public static void CPUKill(object cpuUsage)
{
Parallel.For(0, 1, new Action<int>((int i) =>
{
Stopwatch watch = new Stopwatch();
watch.Start();
while (true)
{
if (watch.ElapsedMilliseconds > (int)cpuUsage)
{
Thread.Sleep(100 - (int)cpuUsage);
watch.Reset();
watch.Start();
}
}
}));
}
static void Main(string[] args)
{
int cpuUsage = 50;
int time = 10000;
List<Thread> threads = new List<Thread>();
for (int i = 0; i < Environment.ProcessorCount; i++)
{
Thread t = new Thread(new ParameterizedThreadStart(CPUKill));
t.Start(cpuUsage);
threads.Add(t);
}
Thread.Sleep(time);
foreach (var t in threads)
{
t.Abort();
}
}
For a uniform stressing: Isak Savo's answer with a slight tweak. The problem is interesting. In reality there are workloads that far exceed it in terms of wattage used, thermal output, lane saturation, etc. and perhaps the use of a loop as the workload is poor and almost unrealistic.
int percentage = 80;
for (int i = 0; i < Environment.ProcessorCount; i++)
{
(new Thread(() =>
{
Stopwatch watch = new Stopwatch();
watch.Start();
while (true)
{
// Make the loop go on for "percentage" milliseconds then sleep the
// remaining percentage milliseconds. So 40% utilization means work 40ms and sleep 60ms
if (watch.ElapsedMilliseconds > percentage)
{
Thread.Sleep(100 - percentage);
watch.Reset();
watch.Start();
}
}
})).Start();
}
Each time you have to set cpuUsageIncreaseby variable.
for example:
1- Cpu % increase by > cpuUsageIncreaseby % for one minute.
2- Go down to 0% for 20 seconds.
3- Goto step 1.
private void test()
{
int cpuUsageIncreaseby = 10;
while (true)
{
for (int i = 0; i < 4; i++)
{
//Console.WriteLine("am running ");
//DateTime start = DateTime.Now;
int cpuUsage = cpuUsageIncreaseby;
int time = 60000; // duration for cpu must increase for process...
List<Thread> threads = new List<Thread>();
for (int j = 0; j < Environment.ProcessorCount; j++)
{
Thread t = new Thread(new ParameterizedThreadStart(CPUKill));
t.Start(cpuUsage);
threads.Add(t);
}
Thread.Sleep(time);
foreach (var t in threads)
{
t.Abort();
}
//DateTime end = DateTime.Now;
//TimeSpan span = end.Subtract(start);
//Console.WriteLine("Time Difference (seconds): " + span.Seconds);
//Console.WriteLine("10 sec wait... for another.");
cpuUsageIncreaseby = cpuUsageIncreaseby + 10;
System.Threading.Thread.Sleep(20000);
}
}
}
Related
I'm trying to make a loop execute regularly every 50 milliseconds on an Atmega 2560. Using a simple delay function won't work, because the total loop time ends up being the time it took to execute the other functions in the loop, plus your delay time. This works even less well if your functions calls take variable time, which they usually will.
To solve this, I implemented a simple timer class:
volatile unsigned long timer0_ms_tick;
timer::timer()
{
// Set timer0 registers
TCCR0A = 0b00000000; // Nothing here
TCCR0B = 0b00000000; // Timer stopped, begin function start by setting last three bits to 011 for prescaler of 64
TIMSK0 = 0b00000001; // Last bit to 1 to enable timer0 OFV interrupt enable
sei(); // Enable global interrupts
}
void timer::start()
{
timer0_ms_tick = 0;
// Set timer value for 1ms tick (2500000 ticks/sec)*(1 OFV/250 ticks) = 1000OVF/sec
// 256ticks - 250ticks - 6 ticks, but starting at 0 means setting to 5
TCNT0 = 5;
// Set prescaler and start timer
TCCR0B = 0b00000011;
}
unsigned long timer::now_ms()
{
return timer0_ms_tick;
}
ISR(TIMER0_OVF_vect)
{
timer0_ms_tick+=1;
TCNT0 = 5;
}
The main loop uses this like so:
unsigned long startTime, now;
while(true)
{
startTime = startup_timer.now_ms();
/* Loop Functions */
// Wait time step
now = startup_timer.now_ms();
while(now-startTime < 50)
{
now = startup_timer.now_ms();
}
Serial0.print(ltoa(now,time_string, 10));
Serial0.writeChar('-');
Serial0.print(ltoa(startTime,time_string, 10));
Serial0.writeChar('=');
Serial0.println(ltoa(now-startTime,time_string, 10));
}
My output looks like this:
11600-11550=50
11652-11602=50
11704-11654=50
11756-11706=50
12031-11758=273
11828-11778=50
11880-11830=50
11932-11882=50
11984-11934=50
12036-11986=50
12088-12038=50
12140-12090=50
12192-12142=50
12244-12194=50
12296-12246=50
12348-12298=50
12400-12350=50
12452-12402=50
12504-12454=50
12556-12506=50
12608-12558=50
12660-12610=50
12712-12662=50
12764-12714=50
12816-12766=50
12868-12818=50
12920-12870=50
12972-12922=50
13024-12974=50
13076-13026=50
13128-13078=50
13180-13130=50
13232-13182=50
13284-13234=50
13336-13286=50
13388-13338=50
13440-13390=50
13492-13442=50
13544-13494=50
13823-13546=277
13620-13570=50
It seems to work well most of the time, but every once in a while something odd will happen with the timing values. I think it has something to do with the interrupt, but I'm not sure what. Any help would be greatly appreciated.
When you have code like this (written in java, but applicable to any similar language):
public static void main(String[] args) {
int total = 0;
for (int i = 0; i < 50; i++)
total += i * doStuff(i % 2); // multiplies i times doStuff(remainder of i / 2)
}
public static int doStuff(int i) {
// Lots of complicated calculations
}
You can see that there's room for improvement. doStuff(i % 2) only returns two different values - one for doStuff(0) on even numbers and one for doStuff(1) on odd numbers. Therefore you're wasting a lot of computation time/power on recalculating those values each time by saying doStuff(i % 2). You can improve like this:
public static void main(String[] args) {
int total = 0;
boolean[] alreadyCalculated = new boolean[2];
int[] results = new int[2];
for (int i = 0; i < 50; i++) {
if (!alreadyCalculated[i % 2]) {
results[i % 2] = doStuff(i % 2);
alreadyCalculated[i % 2] = true;
}
total += i * results[i % 2];
}
}
Now it accesses a stored value instead of recalculating each time. It might seem silly to keep arrays like that, but for cases like looping from, say, i = 0, i < 500 and you're checking i % 32 each time, or something, an array is an elegant approach.
Is there a term for this kind of code optimization? I'd like to read up more on the different forms and the conventions of it but I'm lacking a concise description.
Is there a term for this kind of code optimization?
Yes, there is:
In computing, memoization is an optimization technique used primarily to speed up computer programs by storing the results of expensive function calls and returning the cached result when the same inputs occur again.
https://en.wikipedia.org/wiki/Memoization
Common-subexpression-elimination (CSE) is related to this. This case is a combination of that and hoisting a loop-invariant calculation out of a loop.
I'd agree with CBroe that you could call this specific form of caching memoization, esp the way you're implementing it with the clunky alreadyCalculated array. You can optimize that away since you know which calls will be new values and which will be repeats. Normally you'd implement memoization with a static array inside the called function, for the benefit of all callers. Ideally there's a sentinel value you can use to mark entries which don't have a result computed yet, instead of maintaining a separate array for that. Or for a sparse set of input values, just use a hash (instead of e.g. an array with 2^32 entries).
You can also avoid the if in the main loop.
public class Optim
{
public static int doStuff(int i) { return (i+5) << 1; }
public static void main(String[] args)
{
int total = 0;
int results[] = new int[2];
// more interesting if we pretend the loop count isn't known to be > 1, so avoiding calling doStuff(1) for n=1 is useful.
// otherwise you'd just do int[] results = { doStuff(0), doStuff(1) };
int n = 50;
for (int i = 0 ; i < Math.min(n, 2) ; i++) {
results[i] = doStuff(i);
total += i * results[i];
}
for (int i = 2; i < n; i++) { // runs zero times if n < 2
total += i * results[i % 2];
}
System.out.print(total);
}
}
Of course, in this case we can optimize a lot further. sum(0..n) = n * (n+1) / 2, so we can use that to get a closed-form (non-looping) solution in terms of doStuff(0) (sum of the even terms) and doStuff(1) (sum of the odd terms). So we only need the two doStuff() results once each, avoiding any need to memoize.
I'm new to processing and trying to make a very simple program where i have an arduino that produces a seriel input (according to an analogue read value). The idea is a Processing window will open with a block color shown for 30 seconds. In this time all the readings from the arduino will be summed and averaged - creating an average for that color.
After 30 seconds the colour will change and a new average (for the next color) will start being calculated. This is the code i have started to write (for now focusing on just one 30 second period of green).
I realise there are likely problems with the reading/summing and averaging (i havent researched these yet so i'll put that to one side) - but my main question is why isn't the background green? When i run this program i expect the background to be green for 30 seconds - where as what happens is it is white for 30 seconds then changes to green. Can't figure out why! Thanks for any help!
import processing.serial.*;
Serial myPort;
float gsrAverage;
float greenAverage;
int gsrValue;
int greenTotal = 0;
int greenCount = 1;
int timeSinceStart = 0;
int timeAtStart;
int count=0;
color green = color(118,236,0);
void setup () {
size(900, 450);
// List all the available serial ports
//println(Serial.list());
myPort = new Serial(this, Serial.list()[0], 9600);
}
void draw () {
while (timeSinceStart < 30000) {
background(green);
greenTotal = greenTotal + gsrValue;
greenCount = greenCount + 1;
delay(500);
timeSinceStart = millis()-timeAtStart;
//println(timeSinceStart); for de bugging
}
greenAverage = greenTotal/greenCount;
//println(greenAverage); for de bugging
}
void serialEvent (Serial myPort) {
int inByte=myPort.read();
//0-255
gsrValue=inByte;
}
What I like to do for timers, is use IF statements and use millis() or a constantly updated variable 'm' right inside the condition:
int timeSinceStart;
int m;
void setup(){
timeSinceStart = millis(); // initialize here so it only happens once
}
void draw(){
m = millis(); // constantly update the variable
if(timeSinceStart + 30000 < m){
greenAverage = greenTotal/greenCount; // or whatever is outside while loop
timeSinceStart = millis();
}
//Anything that went inside the while loop can go here, or above the IF
}
This makes it so around every 30 seconds, the background will change once, and you just re-update the timeSinceStart variable in there too. This way, it will only update when you want it to update and not constantly update and break the code.
I tend not to use while loops in processing as they usually cause headaches. Hope my example helps.
May have found a way round this using an IF statement. I perhaps looked over the fact the draw function is itself a loop, so i was able to use a variation of
if (timeSinceStart < 5000) {
background(green);
}
within draw.
When dealing with timed events in Processing you should not use while loops inside the draw() function. The draw() function itself is a while loop which updates the "screen" each frame.
So, what you should do is create a timer and let it do a switch for you inside the draw() function. In your case, if you want to start with a green screen, you do that in the setup() function, and then create a method for altering according to a timer in your draw() function.
This is a suggestion on how you could solve your particular problem. Just change the cycle variable according to your need. In your case it would be 30000.
boolean isGreen = true;
int startTime = 0;
int lastTime = 0;
int cycle = 1000; //the cycle you need
void setup() {
size(200, 200);
background(0, 255, 0); //green
}
void draw() {
startTime = millis();
if (startTime > lastTime + cycle) {
if (isGreen) {
background(255); //white
isGreen = !isGreen;
} else {
background(0, 255, 0); //green
isGreen = !isGreen;
}
lastTime = millis();
}
}
Using dtrace on Solaris, I am able to accumulate on-cpu time for a given process (or execname) for the interval from start until control-C with the following script:
!/usr/sbin/dtrace -qs
dtrace:::BEGIN {
total = 0;
}
sched:::on-cpu
/execname == $$1/
{
self->start = vtimestamp;
}
sched:::off-cpu
/self->start/
{
this->time = vtimestamp - self->start;
total += this->time;
self->start = 0;
}
dtrace:::END {
printf("Total Time on CPU: %d us\n",total/1000);
}
(Accumulated time has a fine-grain granularity allowing nano/microsecond accumulation.)
Over the same timeframe, I would like to accumulate all or many processes in an array and report on all accumulated cpu time at break (^C).
What is the best way to do this?
Okay, with a bit more work I've solved my problem.
Here is the way to get microseconds for all processes (but display per process) over the interval.
#!/usr/sbin/dtrace -qs
dtrace:::BEGIN {
total = 0;
starttimestamp=timestamp;
printf("Starting...\n");
}
sched:::on-cpu
/pid!=0/
{
self->start = vtimestamp;
}
sched:::off-cpu
/self->start && pid!=0/
{
this->time = vtimestamp - self->start;
total += this->time;
#proctime[pid,uid,execname,curpsinfo->pr_psargs] = sum( this->time/1000 );
self->start = 0;
}
dtrace:::END {
printf("Elapsed time %d usec\n",(timestamp-starttimestamp)/1000);
printf("Total Time on CPU: %d us\n",total/1000);
printa(#proctime);
}
#using <mscorlib.dll>
#using <System.dll>
using namespace System;
using namespace System::Text;
using namespace System::IO;
using namespace System::Net;
using namespace System::Net::Sockets;
using namespace System::Collections;
Errors: IntelliSense: "#using" requires C++/CLI to be enabled....
how to fix this prob!?
Your project settings are wrong. Specifically Configuration Properties, General, Common Language Runtime support.
Fall in the pit of success by starting your project by picking one of the project templates in the CLR node.
Choose Project -> Properties from the menu bar. In the Project properties window, under Configuration Properties -> General, make sure that Common Language Runtime Support is set to Common Language Runtime Support (/clr)
In VS2019 it the steps would be :
1/ Right click on the project
2/ Project
3/ Properties
4/ Configuration Properties
5/ Advanced
6/ Common Language Runtime Support change it to Common Language Runtime Support(/clr)
Enable it in your project settings (right click on the projet -> settings) the first tab should provide the option.
The MSDN has a nice example for testing the difference in performance, Parse vs tryParse:
Stopwatch Example
#include <stdio.h>
#using <System.dll>
using namespace System;
using namespace System::Diagnostics;
void DisplayTimerProperties()
{
// Display the timer frequency and resolution.
if (Stopwatch::IsHighResolution)
{
Console::WriteLine("Operations timed using the system's high-resolution performance counter.");
}
else
{
Console::WriteLine("Operations timed using the DateTime class.");
}
Int64 frequency = Stopwatch::Frequency;
Console::WriteLine(" Timer frequency in ticks per second = {0}", frequency);
Int64 nanosecPerTick = (1000L * 1000L * 1000L) / frequency;
Console::WriteLine(" Timer is accurate within {0} nanoseconds", nanosecPerTick);
}
void TimeOperations()
{
Int64 nanosecPerTick = (1000L * 1000L * 1000L) / Stopwatch::Frequency;
const long numIterations = 10000;
// Define the operation title names.
array<String^>^operationNames = { "Operation: Int32.Parse(\"0\")","Operation: Int32.TryParse(\"0\")","Operation: Int32.Parse(\"a\")","Operation: Int32.TryParse(\"a\")" };
// Time four different implementations for parsing
// an integer from a string.
for (int operation = 0; operation <= 3; operation++)
{
// Define variables for operation statistics.
Int64 numTicks = 0;
Int64 numRollovers = 0;
Int64 maxTicks = 0;
Int64 minTicks = Int64::MaxValue;
int indexFastest = -1;
int indexSlowest = -1;
Int64 milliSec = 0;
Stopwatch ^ time10kOperations = Stopwatch::StartNew();
// Run the current operation 10001 times.
// The first execution time will be tossed
// out, since it can skew the average time.
for (int i = 0; i <= numIterations; i++)
{
Int64 ticksThisTime = 0;
int inputNum;
Stopwatch ^ timePerParse;
switch (operation)
{
case 0:
// Parse a valid integer using
// a try-catch statement.
// Start a new stopwatch timer.
timePerParse = Stopwatch::StartNew();
try
{
inputNum = Int32::Parse("0");
}
catch (FormatException^)
{
inputNum = 0;
}
// Stop the timer, and save the
// elapsed ticks for the operation.
timePerParse->Stop();
ticksThisTime = timePerParse->ElapsedTicks;
break;
case 1:
// Parse a valid integer using
// the TryParse statement.
// Start a new stopwatch timer.
timePerParse = Stopwatch::StartNew();
if (!Int32::TryParse("0", inputNum))
{
inputNum = 0;
}
// Stop the timer, and save the
// elapsed ticks for the operation.
timePerParse->Stop();
ticksThisTime = timePerParse->ElapsedTicks;
break;
case 2:
// Parse an invalid value using
// a try-catch statement.
// Start a new stopwatch timer.
timePerParse = Stopwatch::StartNew();
try
{
inputNum = Int32::Parse("a");
}
catch (FormatException^)
{
inputNum = 0;
}
// Stop the timer, and save the
// elapsed ticks for the operation.
timePerParse->Stop();
ticksThisTime = timePerParse->ElapsedTicks;
break;
case 3:
// Parse an invalid value using
// the TryParse statement.
// Start a new stopwatch timer.
timePerParse = Stopwatch::StartNew();
if (!Int32::TryParse("a", inputNum))
{
inputNum = 0;
}
// Stop the timer, and save the
// elapsed ticks for the operation.
timePerParse->Stop();
ticksThisTime = timePerParse->ElapsedTicks;
break;
default:
break;
}
// Skip over the time for the first operation,
// just in case it caused a one-time
// performance hit.
if (i == 0)
{
time10kOperations->Reset();
time10kOperations->Start();
}
else
{
// Update operation statistics
// for iterations 1-10001.
if (maxTicks < ticksThisTime)
{
indexSlowest = i;
maxTicks = ticksThisTime;
}
if (minTicks > ticksThisTime)
{
indexFastest = i;
minTicks = ticksThisTime;
}
numTicks += ticksThisTime;
if (numTicks < ticksThisTime)
{
// Keep track of rollovers.
numRollovers++;
}
}
}
// Display the statistics for 10000 iterations.
time10kOperations->Stop();
milliSec = time10kOperations->ElapsedMilliseconds;
Console::WriteLine();
Console::WriteLine("{0} Summary:", operationNames[operation]);
Console::WriteLine(" Slowest time: #{0}/{1} = {2} ticks", indexSlowest, numIterations, maxTicks);
Console::WriteLine(" Fastest time: #{0}/{1} = {2} ticks", indexFastest, numIterations, minTicks);
Console::WriteLine(" Average time: {0} ticks = {1} nanoseconds", numTicks / numIterations, (numTicks * nanosecPerTick) / numIterations);
Console::WriteLine(" Total time looping through {0} operations: {1} milliseconds", numIterations, milliSec);
}
}
int main()
{
DisplayTimerProperties();
Console::WriteLine();
Console::WriteLine("Press the Enter key to begin:");
Console::ReadLine();
Console::WriteLine();
TimeOperations();
getchar();
}
//Operations timed using the system's high-resolution performance counter.
//Timer frequency in ticks per second = 3319338
//Timer is accurate within 301 nanoseconds
//
//Press the Enter key to begin :
//
//
//
//Operation : Int32.Parse("0") Summary :
// Slowest time : #4483 / 10000 = 95 ticks
// Fastest time : #3 / 10000 = 0 ticks
// Average time : 0 ticks = 99 nanoseconds
// Total time looping through 10000 operations : 1 milliseconds
//
// Operation : Int32.TryParse("0") Summary :
// Slowest time : #7720 / 10000 = 187 ticks
// Fastest time : #1 / 10000 = 0 ticks
// Average time : 0 ticks = 109 nanoseconds
// Total time looping through 10000 operations : 1 milliseconds
//
// Operation : Int32.Parse("a") Summary :
// Slowest time : #3701 / 10000 = 2388 ticks
// Fastest time : #2698 / 10000 = 102 ticks
// Average time : 116 ticks = 35109 nanoseconds
// Total time looping through 10000 operations : 352 milliseconds
//
// Operation : Int32.TryParse("a") Summary :
// Slowest time : #8593 / 10000 = 23 ticks
// Fastest time : #1 / 10000 = 0 ticks
// Average time : 0 ticks = 88 nanoseconds
// Total time looping through 10000 operations : 1 milliseconds
If you are using Visual Studio, you might have to do some installations pre-hand. To install those, open the Visual Studio Installer from the Windows Start menu. Make sure that the Desktop development with C++ tile is checked, and in the Optional components section, also check C++/CLI Support.