Dr. Dobb's is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Channels ▼

Andrew Koenig

Dr. Dobb's Bloggers

Moving Data — Why It Matters

August 24, 2011

In a recent article , I talked about the idea of moving, rather than copying, data as an important new idea in C++0x. This post, and the next few, will go into more detail about the idea, along with some of the surrounding history and theory.

Brian Kernighan once laid down two rules for programmers who want to optimize their programs:

  1. Don't do it.
  2. (for experts only) Don't do it yet.

I agree. I also agree with his similar prescription: "Make it work before you make it faster." Computers exist to make things easier for people, not the other way around. The idea of moving, rather than copying data, is mostly an optimization. So why do I think it's important?

  • Unlike many other optimizations, this one often works invisibly. For example, if I use push_back to append an element to a vector<string>, that code is apt to run much more quickly on an implementation that supports data movement than on one that does not, and I did not have to change any of my code to take advantage of that speed improvement.
  • Moving data is a notion that, until now, was implemented — not quite correctly — as two separate operations. For example, if I write

    a = b;

    I am saying that I want to make a copy of b and put that copy in a. If I know that I am no longer going to use the value of a, then I have overspecified what I want the machine to do. I can make that overspecification less important by writing

    a = b;

    where clear is a (hypothetical) function that gets rid of b's contents without actually destroying its object, but between these two statements, there is still a moment during which two copies of b's value will exist. By writing

    a = std::move(b);

    I am making it clear that there is no need ever to make a second copy of b's contents.
  • Because moving an object does not create a second copy of the object, it becomes possible to define classes with objects that can be moved but not copied. Indeed, the iostream classes have this characteristic.

    It makes sense not to allow stream objects to be copied, because if it were possible to do so, the library would have to assign a meaning to program fragments such as

    istream in = std::cin;
     std::cin >> x;

    Questions would arise: Does reading from std::cin affect what the next read from in will see? Does using put_back on one of the streams affect the other? What about closing one of them and reopening it with another file? Or, for that matter, the same file?

Allowing a stream to be moved but not copied makes it possible, for example, to have a vector<ifstream> without worrying about how to answer these questions.

Moreover, the optimization has more of an effect than one might think:

  • Many implementations take surprisingly long to allocate and free dynamic memory, because in order to do so, the system must maintain a data structure that keeps track of all allocated memory in the entire machine. Such a structure can be large and complicated, hence slow to manipulate.
  • Modern processors usually run much faster than the memory to which they are attached. To compensate for this speed difference, they have caches that they pre-fill with data needed for future instructions. When an instruction dereferences a pointer, as is usually needed for programs that use dynamic memory, the processor has to wait until the memory can report its contents; this delay can take as long as executing many instructions.

Jon Bentley once conducted an informal test to see how many elements one had to put into a vector before it would take longer to insert and delete elements in a vector than in a list. He found the crossover point for vectors of built-in types (such as int) to be between 50 and 100 elements. In other words, deleting an element from a vector, and then copying all of the subsequent elements, took less time than the dynamic memory allocation associated with deleting an element from a list, until the vector grew to many tens of elements. That difference came both from the overhead of dynamic memory allocation and the ability to use the processor's built-in cache.

Kernighan's admonition about optimization applies differently to compiler writers and library designers and implementers than it does to application programmers, because many programmers and machines use a single compiler and library. As a result, an optimization in a compiler will make a lot of programs run faster — and some of those programs might well have been unacceptably slow were it not for the optimization. In effect, an optimization that is part of a compiler — if done correctly — is an optimization that a lot of that compiler's users no longer need to consider.

In short, moving data without copying it is not only an optimization, it's a big optimization. Moreover, it's a way to specify a common operation more accurately and succinctly. As a result, library authors can say more accurately how they intend their libraries to work, and a lot of expensive memory allocation and deallocation becomes completely unnecessary. Although I don't normally care much for optimizations, I'll make an exception for this one.

Related Reading

More Insights

Currently we allow the following HTML tags in comments:

Single tags

These tags can be used alone and don't need an ending tag.

<br> Defines a single line break

<hr> Defines a horizontal line

Matching tags

These require an ending tag - e.g. <i>italic text</i>

<a> Defines an anchor

<b> Defines bold text

<big> Defines big text

<blockquote> Defines a long quotation

<caption> Defines a table caption

<cite> Defines a citation

<code> Defines computer code text

<em> Defines emphasized text

<fieldset> Defines a border around elements in a form

<h1> This is heading 1

<h2> This is heading 2

<h3> This is heading 3

<h4> This is heading 4

<h5> This is heading 5

<h6> This is heading 6

<i> Defines italic text

<p> Defines a paragraph

<pre> Defines preformatted text

<q> Defines a short quotation

<samp> Defines sample computer code text

<small> Defines small text

<span> Defines a section in a document

<s> Defines strikethrough text

<strike> Defines strikethrough text

<strong> Defines strong text

<sub> Defines subscripted text

<sup> Defines superscripted text

<u> Defines underlined text

Dr. Dobb's encourages readers to engage in spirited, healthy debate, including taking us to task. However, Dr. Dobb's moderates all comments posted to our site, and reserves the right to modify or remove any content that it determines to be derogatory, offensive, inflammatory, vulgar, irrelevant/off-topic, racist or obvious marketing or spam. Dr. Dobb's further reserves the right to disable the profile of any commenter participating in said activities.

Disqus Tips To upload an avatar photo, first complete your Disqus profile. | View the list of supported HTML tags you can use to style comments. | Please read our commenting policy.