CODEWITHSUNDEEP
c#datetime
Cylindric
Cylindric

Reputation: 5894

Predicting process end time

I have a simple problem that I'm having more trouble with than I should.

It's pretty straight forward: I have a time the process started _startTime, and the current time. I also have the total number of records I need to process _records, and the current record number _current.

How can I get a simple linear prediction of the end time? I've ended up in a rabbit hole of differencing ticks and stuff, but I imagine there's some simple DateTime or TimeSpan trickery I could be using,

TimeSpan delta = DateTime.Now - _startTime;
float progress = (float)_records / (float)_current;

Upvotes: 0

Views: 146

Answers (3)

peter
peter

Reputation: 15139

hmm really sounds much simpler than it is probably

TimeSpan timeTakenSoFar = DateTime.Now - _startTime;
double percentageDoneSoFar = (double)_current / (double)_records;
TimeSpan timeTakingInTotal = timeTakenSoFar / percentageDoneSoFar;
DateTime finishTime = _startTime.Add(timeTakingInTotal);

This should work.

Update:

According to MSDN you cannot divide TimeSpan, but in that case you can use ticks: (I don't have a C# compiler at hand to check if it's 100% syntactically correct)

TimeSpan timeTakenSoFar = DateTime.Now - _startTime;
double percentageDoneSoFar = (double)_current / (double)_records;
double timeTakingInTotalInTicks = timeTakenSoFar.Ticks / percentageDoneSoFar;
DateTime finishTime = _startTime.AddTicks((long)timeTakingInTotal);

Upvotes: 1

user2160375
user2160375

Reputation: 

var restRecords = _records - _current;
var averageTimeForRecord = delta.TotalMilliseconds / _current;
var restTime = restRecords * averageTimeForRecord;
var finishDate = DateTime.Now.AddMilliseconds(restTime);

Maybe that is what you need?

Upvotes: 0

Louis van Tonder
Louis van Tonder

Reputation: 3710

Take a sample every few cycles, and determine how long they took from start time.

now.subtract(_startTtime)

Divide that by the _current pointer, and get the time per cycle.

Multiple the time/cycle with ((_records - _current) <- cycles you have left to do)

The higher your "sample" resolution, the more accurate your result... but it is still just a prediction.

Upvotes: 1

Related Questions