I experienced unexpected performance behavior of my code which uses a queue. I realized that performance degraded when more elements were in the queue. It turned out that usage of the size()
method was the reason. Here is some code that shows the problem:
#include <queue>
#include <list>
#include <iostream>
#include "Stopwatch.h"
using namespace std;
struct BigStruct
{
int x[100];
};
int main()
{
CStopwatch queueTestSw;
typedef BigStruct QueueElementType;
typedef std::queue<QueueElementType, std::list<QueueElementType> > QueueType;
//typedef std::queue<QueueElementType > QueueType; //no surprise, this queue is fast and constant
QueueType m_queue;
for (int i=0;i<22000;i++)
m_queue.push(QueueElementType());
CStopwatch sw;
sw.Start();
int dummy;
while (!m_queue.empty())
{
//remove 1000 elements:
for (int i=0;i<1000;i++)
{
m_queue.pop();
}
//call size() 1000 times and see how long it takes
sw = CStopwatch();
sw.Start();
for (int i=0;i<1000;i++)
{
if (m_queue.size() == 123456)
{
dummy++;
}
}
std::cout << m_queue.size() << " items left. time: " << sw.GetElapsedTimeInSeconds() << std::endl;
}
return dummy;
}
Where CStopwatch
is a class which uses clock_gettime(CLOCK_REALTIME, ..)
. The result is:
21000 items left. time: 1.08725
20000 items left. time: 0.968897
19000 items left. time: 0.818259
18000 items left. time: 0.71495
17000 items left. time: 0.583725
16000 items left. time: 0.497451
15000 items left. time: 0.422712
14000 items left. time: 0.352949
13000 items left. time: 0.30133
12000 items left. time: 0.227588
11000 items left. time: 0.178617
10000 items left. time: 0.124512
9000 items left. time: 0.0893425
8000 items left. time: 0.0639174
7000 items left. time: 0.0476441
6000 items left. time: 0.033177
5000 items left. time: 0.0276136
4000 items left. time: 0.022112
3000 items left. time: 0.0163013
2000 items left. time: 0.0101932
1000 items left. time: 0.00506138
This seems to contradict http://www.cplusplus.com/reference/stl/queue/size/:
Complexity: Constant.
The problem is worse if the struct is bigger. I am using GCC 4.3.2.
Can you explain this? Is there a way to solve this using the list?
(Please note that I need to use the list as underlying container of the queue because I need constant time insertion complexity.)
The std::queue is probably slower because it may free some memory when doing all these pop() while yours doest release anything.
So swapping a queue, even at a relatively small size of 100 items, is 5x faster. And the larger the queue, the more performance we can squeeze out of not using a list.
Queue is significantly faster than List , where memory accesses are 1 vs. n for List in this use case.
queue::size() is an inbuilt function in C++ STL which is declared in <queue> header file. queue::size() is used to check whether the size of the associated queue container. This function returns an unsigned int value, i.e the size of the queue container, or the number of elements present in a queue container.
queue
is a container adaptor, so you have to understand that the complexity descriptions may refer only to the work the adaptor does itself (which is indeed constant, namely just passing the call through to the underlying container).
For example, see this reference: size()
calls the underlying container's size()
function. For a list
, this has complexity O(n) in C++98/03, and O(1) in C++11.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With