Also known as
random access memory.
RAM (random access memory) is a type of physical memory(2) that can be read from and written to.
In Java, the reference objects together with ordinary references and finalization generate a hierarchy of reachability that guides the collector(1) on what to do when an object is about to die. There are six strengths:
Basically, an object is only as reachable as the weakest link in the strongest path from the roots. Note that the Java specification’s description of the reachabilities is a bit patchy, but that’s what it intends. It is unspecified where Java Native Interface’s weak global references fit into this.
Relevance to memory management
An exception which occurs when reading from an address in virtual memory.
A system with no virtual memory capability can be said to have real memory.
On older architectures, programs could only directly access data in real memory. Where this was inefficient, they had to store data on disk, and sometimes had alternate portions of program image called overlays.
This word is usually used only in connection with automatic memory management.
In memory management, a reference is the general term for a link from one object to another. Some programming languages have more specific meanings for the term.
The terms “pointer” and “reference” are often interchangeable, but some programming languages differentiate the two in subtle ways.
In the MPS
A reference is represented in the C interface by a value of type mps_addr_t (an alias for void *) which points to a memory location within the object (typically the base of the object, but for objects with headers this may not be the case). The pointer returned by mps_alloc() and mps_reserve() is a reference to the object allocated.
The client program is free to represent references as it chooses (for example, with tags), provided that during scanning it is able to decode a reference from its representation into the MPS interface representation and encode a reference from the MPS into its representation.
Reference counting systems perform automatic memory management by keeping a count in each object, usually in a header, of how many references there are to the object. Objects to which there are no references cannot be accessed by the mutator; they are therefore dead and may be reclaimed.
The reference count is incremented for each new reference, and is decremented if a reference is overwritten, or if the referring object is recycled. If a reference count falls to zero, then the object is no longer required and can be recycled.
There are four main problems with simple reference counting:
Reference counting has the advantage that it can reclaim objects promptly, and for this reason it is often used to reclaim non-cyclic data structures in file systems, databases and operating system kernels. When there is a possibility of cyclic data structures, reference counting is sometimes used together with a tracing garbage collector that runs infrequently. Such combinations are generally less efficient than using a tracing collector by itself, but the promptness of reference counting may be important.
Pauses due to reference counting are typically fairly short, and it may be appropriate as a form of incremental garbage collection. But removing a single reference may cause the recycling of a large number of objects at once, so it is not suited to real-time systems where minimum pause times must be guaranteed. There are more complex variations of the technique that address this problem.
Reference counting is often used because it can be implemented without any support from the language or compiler. In C++ this can be encapsulated in a class, using a smart pointer. However, it would normally be more efficient to use a tracing garbage collector instead. The performance of reference counting can be improved substantially with compiler support, using refinements such as deferred reference counting, which has been successfully used in Smalltalk and other languages.
Despite the problems, reference counting is often used for distributed garbage collection. This is because refinements such as weighted reference counting require less inter-process communication than tracing.
In Java, a reference object (java.lang.ref.Reference) encapsulates a reference to some other object, in order to make the garbage collector handle it specially. In particular, a Java program can use this to detect when the referent becomes unreachable.
Basically, the encapsulated reference is a weak reference(1); it will be cleared by the collector(1) when all other references to the referent have disappeared. However, in order to better control what happens at the end of an object’s lifetime, Java 1.2 provides three classes of reference objects, each with its own peculiarities: SoftReference, WeakReference, and PhantomReference. Each of these classes has its uses in managing memory. The reference objects together with ordinary references and finalization generate a hierarchy of reachability that guides the collector on what to do when an object is about to die.
A reference object can be registered with a queue, and it will be enqueued when the collector determines that the referent is softly, weakly or phantom reachable, as the case may be. A program can use these queues to perform some action when an object is dying. This allows finer control than the older finalization mechanism alone.
This feature was introduced in Java 1.2 (confusingly, part of the Java 2 Platform).
Region inference infers a region for each object. When a region dies, all the objects in it are known to be dead, whether reachable or not. Regions obey a strict stack discipline; that is, when a region dies, all younger regions also die. In this way, region inference occupies a middle ground between stack allocation and heap allocation.
Relevance to memory management
Run-time systems for garbage-collected languages sometimes partition the set of machine registers a priori into two categories: those always traced and updated by the garbage collector and those ignored by it.
This partitioning provides a separation of concerns between the compiler and the garbage collector. The compiler can generate code that produces values the garbage collector would not be able to handle (say, because they have no tags), as long as those values are kept in the ignored registers. The garbage collector can trust that the registers it looks at always contain valid data, and can perform exact garbage collection.
Register set partitioning increases the demand for registers (register pressure), but may reduce the amount of boxing needed.
Relocating means moving data from one location to another and updating all references.
Relocation is often performed to avoid external fragmentation.
Program loading sometimes relocates code and static data.
Many memory management algorithms depend on partitioning the objects and require special handling for references between partitions. Keeping track of such references in a remembered set eliminates the need to scan the originating partition to find them.
In the MPS
A reference that logically belongs to a formatted object and so must be fixed when the object is scanned, but which is not stored within the block containing the object. (For example, in an auxiliary table of some sort.)
The MPS does not generally support remote references because those references may be protected and so if scan method attempts to fix them this will hit a barrier(1) and cause a re-entrant call to the MPS.
This is useful in an incremental or concurrent collector(1), as no read barrier is required; the mutator can continue to use old objects. The collector uses a write barrier to replicate the writes to the new copies.
Reserving addresses prevents other components of the program using the same addresses, without consuming swap space. This technique is often used in BIBOP schemes, where one might want to reserve a large amount of address space but only sparsely map it.
An object is said to have been resurrected if it was determined to be finalizable by the garbage collector (that is, the only thing keeping it alive was the fact that it required finalization), but then a new strong reference was created to it.
This can happen via a weak reference(1) or by the finalization procedure storing a permanent copy of its reference to the object.
In the MPS
Also known as
ROM (read-only memory) is a type of physical memory(2) that can be read from, but not written to. The contents of ROM are usually set in the factory.
In the MPS