sleep_until_or_later

Overview

Is std library's sleep_for accurate to a microsecond ? Put another way, what should be our expectation of the difference between expected_count of miroceconds sleep and actual_count. It is required that actual_count be equal or greater than expected_count. Initial tests showed that std::this_thread::sleep_for* is accurate to a millisecond or so.

sleep_until_or_later is a header-only microsecond accurate alternative to std::this_thread::sleep_for.

Design

Requirements

• OS’s scheduler priority choice can not be altered.
• Runs on the thread it is called on. Does not create any threads.
• Minimal work above comparable std::this_thread::sleep_for.
• Simple API void sleep_until_or_later (unsigned int microseconds)

Design approach: Coarse duration & Fine duration

Divide the expected_count in to larger coarse and fine duration counts. Use std::this_thread::sleep_for for coarse_duration, followed by call sleeping through fine_duration be repeated calls to ::now().

*1. Initial_Mark = ::now()

*2. std::this_thread::sleep_for for a chunk_count = expected_count - delta_count

*3. Calculate remainder_count = ::now() - Initial_Mark

*4. Rem_Mark = ::now()

*5. Initialize duration_count = ::now() - Rem_Mark

*6. While (duration_count < remainder_count) {duration_count = ::now() - Rem_Mark}

Choice of fine_duration count ( microseconds)

Fine_duration count is smaller than expected_count and is set proportional to the expected_count but never larger than 10,000 ( or 10 milliseconds)

Example

We'll now see how sleep_until_or_later can be used. The header file can be fetched from the repository.

#include "sleep_util_or_later.hpp"

unsigned int in_microsecs = 1234567u;
auto mark = std::chrono::steady_clock ::now ();

sleep_until_or_later(in_microsecs.count);

auto took = std::chrono::duration_cast<std::chrono::microseconds>(std::chrono::steady_clock ::now () - mark);
auto time = std::chrono::duration_cast<std::chrono::duration<float>>(took);
auto diff = std::chrono::duration_cast<std::chrono::duration<float>>(sleep_time - time);

Executable Demo: sleep_until_or_later

Building

git clone https://github.com/gmnamra/sleep_until_or_later cd sleep_until_or_later open sleep_until_or_later.xcodeproj Set Team and build

Running

$ ./sleep_until_or_later
    Measures performance of sleep_until_or_later versus this_thread::sleep_for
    Usage:  [ optional:-[vV] verbose optional:count <microseconds>
    commandline is empty, runs standard test with verbose set to off
    If count is 0, runs built in test cases
    else sleep_until_or_later is run and performance compared with this_thread::sleep_for

$ ./sleep_until_or_later -v 10000000
sleep_until_or_later is run and performance compared with thisthread::sleep_for
Testing sleep_until_or_later 
0 out of 10     Expected     10000000 units      Observed      10000003 with -3 microseconds 
1 out of 10     Expected     10000001 units      Observed      10000002 with -1 microseconds 
2 out of 10     Expected     10000002 units      Observed      10000003 with -1 microseconds 
3 out of 10     Expected     10000003 units      Observed      10000004 with -1 microseconds 
4 out of 10     Expected     10000004 units      Observed      10000005 with -1 microseconds 
5 out of 10     Expected     10000005 units      Observed      10000007 with -2 microseconds 
6 out of 10     Expected     10000006 units      Observed      10000008 with -2 microseconds 
7 out of 10     Expected     10000007 units      Observed      10000008 with -1 microseconds 
8 out of 10     Expected     10000008 units      Observed      10000009 with -1 microseconds 
9 out of 10     Expected     10000009 units      Observed      10000011 with -2 microseconds 
Testing sleep_for  
0 out of 10     Expected     10000000 units      Observed      10005030 with -5030 microseconds 
1 out of 10     Expected     10000001 units      Observed      10001864 with -1863 microseconds 
2 out of 10     Expected     10000002 units      Observed      10000586 with -584 microseconds 
3 out of 10     Expected     10000003 units      Observed      10004994 with -4991 microseconds 
4 out of 10     Expected     10000004 units      Observed      10005036 with -5032 microseconds 
5 out of 10     Expected     10000005 units      Observed      10005004 with -4999 microseconds 
6 out of 10     Expected     10000006 units      Observed      10001017 with -1011 microseconds 
7 out of 10     Expected     10000007 units      Observed      10001751 with -1744 microseconds 
8 out of 10     Expected     10000008 units      Observed      10004987 with -4979 microseconds 
9 out of 10     Expected     10000009 units      Observed      10001330 with -1321 microseconds 
===============================================================
10 Test Cases
sleep_until_or_later was on or later at most by 1
this_thread::sleep_f0r was on or later at most by 3421
======================10000000 microseconds =========Passed