C / C++ / MFC

15-Nov-07 4:49

Hello everyone,

I read MSDN document about reserve method of vector, but I can not find answers to the following 2 questions after reading it,

1. if we reserve 10,000 size, will the memory actually be occupied 10,000 size (like committed memory of Windows Virtual Memory management) or the memory will just be reserved for future use (like reserved memory of Windows Virtual Memory management);

2. Will reserve method of vector improve performance? If yes, could anyone share some experiences please?

If reserve method has any additional benefits beyond performance, please also let me know. Smile | :)

thanks in advance,
George

Nemanja Trifunovic15-Nov-07 4:56

Nemanja Trifunovic

15-Nov-07 4:56

George_George wrote:
1. if we reserve 10,000 size, will the memory actually be occupied 10,000 size (like committed memory of Windows Virtual Memory management)

Yes.

George_George wrote:
2. Will reserve method of vector improve performance?

It may if you are adding elements at the end.

Programming Blog
utf8-cpp

George_George15-Nov-07 15:54

15-Nov-07 15:54

Thanks Nemanja,

regards,
George

led mike15-Nov-07 4:58

15-Nov-07 4:58

When n supposes an expansion in capacity, a reallocation happens during the call to this function, granting that no further automatic reallocations will happen because of a call to vector::insert or vector::push_back until the vector size surpasses at least n

link[^]

George_George15-Nov-07 16:07

15-Nov-07 16:07

Thanks led,

I can understand reserve will improve performance of push_back, but how will reserve improve the performance of insert? Seems you (or the article) means that reserve will improve the performance of both insert and push_back.

regards,
George

led mike16-Nov-07 5:04

16-Nov-07 5:04

George_George wrote:
but how will reserve improve the performance of insert?

again, only if the insert does not cause the current allocation to be exceeded. That is the basis of the optimization, the minimization of heap allocations. On an insert if adequate memory has already been allocated by the reserver() call, then a memmove followed by the assignment of the inserting value could accomplish the insert.

George_George18-Nov-07 1:25

18-Nov-07 1:25

Thanks led,

Your reply is clear.

regards,
George

Matthew Faithfull15-Nov-07 5:07

Matthew Faithfull

15-Nov-07 5:07

In terms of Virtual Memory it will be committed. This doesn't mean it won't end up being paged out to a swap file/page file though.

Any perfromance improvement will depend on how you're using the vetor and on what else the machine is doing. In general if you're working with a larger number of items but only 1 at a time, adding and deleting more or less at random then reserving will give you the greatest performance gain. If you're performing entirely linear operations, add 10,000 items, then iterate over them a few times in order, then free them, it will give you very little or nothing.
Testing memory performance is tricky because new and delete do 50x and more as much work in Debug as in Release. Other factors like low memory or task switching or heavy use of non-paged pool in comms apps can easily swamp and distort your figures beyond use. Reserving of course also has the down side of using more memory than you actually need. It all depends whether you're optimising for memory footprint or speed. Of course you could write your own heap manager and override the global new and delete operators. If it's good enough for the .NET team it must be OK right? Big Grin | :-D

Nothing is exactly what it seems but everything with seems can be unpicked.

George_George15-Nov-07 16:29

15-Nov-07 16:29

Thanks Matthew,

I do not quite agree with your following scenarios, Smile | :)

Please feel free to correct me if I am wrong?

1. In the following scenario, you mentioned there will be performance improvement,

Matthew Faithfull wrote:
In general if you're working with a larger number of items but only 1 at a time, adding and deleting more or less at random then reserving will give you the greatest performance gain.

Suppose we reserved a large number of memory, but we only do small operation (e.g. adding/deleting one item).
Why will performance be improved? On the contrary, I think the following scenario you mentioned below will improve performance if we reserve a large number of memory in advance. Smile | :)

2. I think in the following scenario, since we reserved a large number of memory in advance, so the performance should be improved. But why you mentioned there will be almost no performance improvements?

Matthew Faithfull wrote:
If you're performing entirely linear operations, add 10,000 items, then iterate over them a few times in order, then free them, it will give you very little or nothing.

regards,
George

Http Request

San246815-Nov-07 4:28

San2468

15-Nov-07 4:28

Hi All,
I would like to post a xml string to a http server & in turn the http server is supposed to return another xml as response. In dot net I found WebRequest & WebResponse classes works as I expect, but I need to do it in VC++. I have tried using CHttpConnection of MFC(wininet) & IXmlHttpRequest, can somebody suggest the appropriate option.

Thanks.

Re: Http Request

led mike15-Nov-07 4:54

15-Nov-07 4:54

Who am I, I don't Know wrote:
I have tried using CHttpConnection of MFC(wininet)

That should work.

Re: Http Request

San246815-Nov-07 5:10

San2468

15-Nov-07 5:10

Following is the code lines I am trying

CString cstrServer = L"SomeServer";
CString cstrObject = L"SomePage.aspx";

CInternetSession HttpSession;
CHttpConnection *pHttpConnection = HttpSession.GetHttpConnection(cstrServer, nPort, NULL, NULL);

if(pHttpConnection)
{
CHttpFile* pHttpFile = pHttpConnection->OpenRequest(L"POST", cstrObject);

//I supposed to do something here, but I receive CInternetFile* in CHttpFile NULL
}
Unable to get the problem.

Re: Http Request

led mike15-Nov-07 5:25