Fortsätt till huvudinnehåll

Some reflections on time consumption creating new threads

Today I wrote some tests with the goal to verify thread safety on a part of production code. Using TDD I wrote a test that started two new threads that both updated the same object, and while they were running I verified the object's data. By putting this in a loop I was able to get a unit test that failed on each run.

By adding locks at the correct places I got the test to pass. But looking at the time it took to run the test I noticed that it was quite slow. I knew that the updates to the object's data were fast and started wondering if it was the start and join of the threads that were slowing it down.

So, instead of using new Thread(task) I switched to using Task.Run(task) and task.Wait() instead of thread.Join(). The main difference here being task adding a job to a pool of worker threads while new Thread creates a completely new thread that is stopped and disposed after it has run to completion.

By doing this small change the test time decreased by 25% (from 800 ms to 600 ms). So, lesson learned, do not start new threads if there is a threadpool available that you can use.

Kommentarer

Skicka en kommentar

Populära inlägg i den här bloggen

Does TDD really improve software quality?

I have asked myself this question several times, and searched for answers, without coming up with any clear answer. Therefore I have decided to go hard core TDD for a longer period of time (at least 6 months) to really evaluate the effects. There are several things that I find confusing when it comes to TDD. One example is what actually defines a unit test. What is a "unit" anyway? After reading a bit about it I found a text claiming that the "unit" is "a unit of work", i.e. something quite small. Like converting a string to UPPERCASE or splitting a string into an ['a','r', 'r', 'a', 'y'] of chars. This work is usually performed by a single call to a single method in a single, isolated, class. So, what does it mean that a class is isolated? Does it mean that it doesn't have any dependencies to other classes? NO! In the context of TDD it means that any dependencies are supplied by the test environment, for exa...
3 kommentarer
Läs mer

Codility tasks - Part I

I was recently faced with two codility tasks when applying for a job as an Embedded Software Engineer. For those of you who arn't familiar with Codility you can check out their website here:  www.codility.com Task one - Dominator The first task was called Dominator. The goal was to, given a std::vector of integers, find an integer that occurs in more than half of the positions in the vector. If no dominator was found -1 should be returned. My approach was to loop through the vector from the first to the last element, using a std::map to count the number of occurences of each integer. If the count ever reached above half the size of the vector I stopped and returned that integer and if I reached the end without finding a dominator I returned -1. So was that a good approach? Well, the reviewer at the company rated the solution as 'pretty ok'. His preferred solution was store the first integer in the array and set a counter to 1. Then loop through the remaining i...
7 kommentarer
Läs mer

Codility tasks - Part II

Now, the second codility task I was faced with was a bit tougher. The goal was to create a function that, given a vector of integers A and an integer K, returned the number of integer pairs in the vector that, when added, sums up to K. Let me give you an example. Assume that you are given a vector A = [0, -1, 3, 2, -5, 7] and K = 2. Possible combinations to get K are (0, 2), (-1, 3), (3, -1), (2, 0),  (-5, 7), and (7, -5). In other words, the function should return 6. Now, how did I solve this task? The first solution that came to mind involved nested for-loops. The outer loop picking one integer at the time from the vector and the inner loop adding the integer to the others one by one to see if the result is K. This solution works, but it does not scale well. Time complexity will be O(N**2) ,   something that for large vectors will result in very long execution times. My second approach was to use my old friend, the integer counter, and count all occurences of each...
2 kommentarer
Läs mer