In computing, garbage collection is a system of automatic memory management which seeks to reclaim memory used by objects which will never be referenced in the future. It is commonly abbreviated as GC. The part of a system which performs garbage collection is called a garbage collector.
When a system has a garbage collector it is usually part of the language run-time system and integrated into the language. The language is said to be garbage collected. Garbage collection was invented by John McCarthy as part of the first Lisp system. When someone refers to a garbage collected language what is meant is that garbage collection is a feature specified by the language. There are also some languages (mostly formal languages) which would require a garbage collector in any practical implementation, but whose specifications do not require garbage collection, usually because they are not intended to be used as practical programming languages. The lambda calculus is an example of such a language.
The Symbolics Lisp machine was unique in having hardware support for garbage collection.
The basic principle of how a garbage collector works is:
- Determine what data objects in a program cannot be referenced in the future
- Reclaim the storage used by those objects
Although in general it's impossible to know the moment an object has been used for the last time, garbage collectors use conservative estimates that allow them to identify when an object could not possibly be referenced in the future. For example, if there are no references to an object in the system, then it can never be referenced again. This is the criterion used by most modern garbage-collection systems, but see Disadvantages of Tracing Garbage Collectors below.
While garbage collection assists the management of memory, the feature is also almost always necessary in order to make a programming language type safe, because it prevents several classes of runtime errors. For example, it prevents dangling pointer errors, where a reference to a deallocated object is used.
Tracing garbage collectors
Tracing garbage collectors are the most common type of garbage collector. They focus on locating reachable objects, which are objects that may be referenced in the future, and then discard all remaining objects.
Reachability of an object
More formally, objects can be reachable in only two ways:
- A distinguished set of objects are assumed to be reachable -- these are known as the roots. In a typical system these objects will be the machine registers, the stack, the instruction pointer, the global variables; in other words, everything that a program can reach directly.
- Anything referenced from a reachable object is itself reachable. This is referred to as "transitivity."
Informally, a reachable object is one that the program could get to by starting at an object it can reach directly, and then following a chain of pointer references.
Basic algorithm
Tracing garbage collectors perform garbage collection cycles. A cycle is started when the collector decides (or is notified) that it needs to reclaim storage, which in particular happens when the system is low on memory. A cycle consists of the following steps:
- Create initial white, grey, and black sets; these sets will be used to maintain progress during the cycle. Initially the black set is empty, the grey set consists of specially denoted objects known as the "roots" and possibly some additional objects chosen according to the particular algorithm used, and the white set is everything else. At any time in the algorithm a particular object will be in exactly one of the three sets. The set of white objects can be thought of as the set of objects that we are trying to reclaim the storage for; in the course of the cycle the algorithm will remove many objects from the white set, leaving behind a set of objects that it can reclaim the storage for.
- (This step is repeated until there are no objects in the grey set.) Pick an object from the grey set. Move all the white objects that are referenced (reachable in one step of following pointers) from the selected object into the grey set. Move the selected object from the grey set to the black set.
- When there are no more objects in the grey set, then all the objects remaining in the white set are not reachable and the storage occupied by them can be reclaimed.
The tricolour invariant is an important property of the objects and their colours. It is simply this:
- No black object points directly to a white object.
Notice that the algorithm above preserves the tricolour invariant.
The initial partition has no black objects, so the invariant trivially holds.
Subsequently whenever an object is made black any white objects that it references are made grey, ensuring that the invariant remains true.
When the last step of the algorithm is reached, because the tricolour invariant holds, none of the objects in the black set point to any of the objects in the white set (and there are no grey objects) so the white objects must be unreachable from the roots,
and the system calls their finalisers and frees their storage.
Some variations on the algorithm do not preserve the tricolour invariant but they use a modified form for which all the important properties hold.
Variants of the basic algorithm
The basic algorithm has several variants.
- There is the issue of whether the garbage collector moves objects in memory (that is, changes their address) or not: moving or non-moving GC.
- Some collectors can correctly identify all pointers (references) in an object; these are called "precise" (or "exact"; or "accurate") collectors, the opposite being a "conservative" or "partly conservative" collector. "Conservative" collectors have to assume that any bit pattern in memory is a pointer if (when interpreted as a pointer) it would point into any allocated object. Thus, conservative collectors may have some false negatives, where storage is not released because of accidental fake pointers. However, this is not a big drawback in practice.
- There is the issue of whether the garbage collector can run interleaved or in parallel with any of the rest of the system: the simplest garbage collectors stop the rest of the system while they perform a cycle; they are non-incremental, whereas incremental collectors interleave their work with units of activity from the rest of the system. Some incremental collectors can run fully parallel in a separate thread; these can theoretically run on a separate CPU, but cache issues make this less practical than it may initially appear.
Mark and sweep
Tracing collectors can also be divided by considering how the three sets (of white, grey, and black objects) are implemented. A mark-sweep GC maintains a bit (or two) with each object to record whether it is white or black; the grey set is either maintained as a separate list or using another bit. A Copying GC is a moving GC which, in its simplest form, moves all reachable objects to a single memory area, and then reuses all memory outside this area. Copying GC has the important benefit of resisting fragmentation.
Generational GC
There is the issue of when to perform a cycle and what objects to place in the initial grey set. A simple collector will always put only the roots in the initial grey set, and everything else will be initially white.
Statistically speaking, the objects most recently created in the runtime system are also those most likely to quickly become unreachable. A Generational GC divides objects into generations and will generally only place the objects of a subset of all the generations into the initial white set (the grey set being everything else). This can result in faster cycles.
Disadvantages of tracing garbage collectors
The primary problems with tracing garbage collection arise from the nature of how it is invoked. A collection cycle can occur at any time and can require an arbitrarily long amount of time to complete. While acceptable in many applications, in others for which timing and quick response is critical such as real-time applications, it may cause disaster. Incremental garbage collection helps deal with this problem.
A more fundamental issue is that garbage collectors violate locality of reference, since they deliberately go out of their way to find bits of memory that haven't been accessed recently. This is a particular issue because modern computer architectures get more and more of their performance from manifold intricate levels of caching, which depend crucially for their effectiveness on the assumption of locality of reference. In other words, garbage collection is an inherently cache-hostile activity.
(Actually, that isn't much of a problem with generational garbage collection, and a copying collector automatically defragments memory in a way that keeps related data together, so garbage collection can be more cache-friendly than explicit memory allocation.)
The other main problem is that a large amount of garbage can build up very quickly, particularly in functional programming languages, before the garbage collector has a chance to collect any of it, resulting in allocation inefficiencies, a temporary bloating in the image size, and a long collection cycle once it occurs; we say that it is not prompt.
To see how tracing garbage collectors are not prompt, consider this example:
loop
x := new object
x := 0
With a naïve tracing garbage collector, each time through the loop more and more memory is allocated, until all the memory in the system is consumed and the garbage collector is forcibly invoked. It's clear why this is unacceptable when we realize that this code never needs more than the memory required for one object!
(That's a bad example because it's a case where GC can easily be much faster than explicit memory management. With GC, "new" just increments a pointer in the young generation pool, which is about as fast as stack allocation. Collection time is proportional to number of live objects, not number of allocated objects, so if no new live objects are created by the loop, collection takes very little time. In contrast, explicit memory management requires one call to "new" and "free" per loop iteration, and at least one of those must be slower than the GC version of "new". Still, explicit memory management is probably better if you need hard realtime guarantees.)
Finally, a fault shared by all existing garbage collectors is the conservative assumption that memory is still in use if it is still reachable. In many systems, references are kept long after they cease to be used. A popular research area is to find less conservative ways of collecting objects that will never be used again; that is, to safely collect objects to which references still exist. In systems with standard garbage-collection, freeing of this memory can be accelerated by explicitly destroying pointers, but this forces some of the responsibility of memory management back on the programmer.
As one simple example, consider the command-line arguments passed to the startup function. Typically, the program which uses them will parse these arguments and then perform a command accordingly, never touching the arguments themselves again. But they are still kept in memory, because references to them still exist at the bottom of the call stack, in the entry function. Although these objects occupy minimal memory, similar situations often occur with very large or numerous objects.
(That's a bad example because in programs without GC, the command-line arguments are never freed either, there's no safe way to free them, but in a program with GC, the command-line arguments will sometimes get freed if they become inaccessible, so GC wins in that case. A better example is a symbol table that maps names to values. Nothing in the symbol table can ever be collected until the entire symbol table is inaccessible, so it may still be necessary to do some explicit memory management with reference counting or similar. In some cases, weak references can be used to avoid explicit memory management.)
All of these are often used as arguments against garbage collection and for explicit memory handling, particularly in time-critical applications. However, other forms of garbage collection do not necessarily share all of these faults, although the last is nearly universal, but they may have different faults of their own; all options should be considered when deciding whether to use garbage collection.
Reference counting
Reference counting is a form of garbage collection where each object has a count of the number of references to it, and the object's memory is reclaimed when the count reaches zero. Reference counting has some advantages: it is easy to implement and its execution time is predictable (there is no need to pause the program to look for unreference objects). However, the performance of systems using reference counting can be slower than other garbage collection techniques, because of the need to update reference counts whenever a reference is changed. Reference counting also requires some space in each object. Simple reference counting also fails to reclaim the memory used by data structures that have cycles (such as doubly linked lists). See reference counting for more information.
Implementations
Programming languages whose standard implementations include automatic garbage collection:
External links
Last updated: 05-15-2005 06:43:50
