How Windows NT Provides 4 Gigabytes of Memory (99707)

SUMMARY

Windows NT has an addressable memory space of four gigabytes. This is considered a virtual memory space as most computers do not have the physical RAM to actually contain four gigabytes of data. Because of the services provided by the Windows NT virtual memory manager, programs can be written to exploit large amounts of memory without being concerned with the type of memory or the amount available. The Windows NT virtual memory manager provides this large, virtual memory space to applications via two memory management processes: moving data between physical RAM and the hard disk or paging, and translating physical memory addresses to virtual memory addresses or mapped file I/O.

MORE INFORMATION

PAGING

When a file is larger than the size of your available system RAM, the virtual memory manager uses paging to move data between RAM and the hard disk. It accomplishes this by dividing all physical memory and virtual memory (the memory used by applications) into equal sized blocks or pages (4K each). When a request is made to access data that is not in RAM, the virtual memory manager swaps a page from RAM with the desired page from the paging file (a file on the hard disk used solely by the virtual memory manager for extra data storage).

Page states are registered in a virtual page table. If a page is in RAM and is immediately available to its process (or program), it is marked as valid. Pages that have been moved to the paging file are marked as invalid. When a process tries to access an invalid page, the central processing unit (CPU) generates a page fault. The virtual memory manager traps page faults and retrieves pages from the paging file. If there is no room in RAM to store a new page, a valid page is moved out.

PAGING POLICIES

There are three policies used by the virtual memory manager to determine how and when paging is performed:

Fetch

A demand-based paging algorithm is used to retrieve (or fetch) memory pages. The virtual memory manager waits until a process thread attempts to access an invalid page before loading the page into RAM. Because page faults and disk access are both somewhat slow, additional pages are loaded along with the desired page in a process called clustering. The additional pages are chosen by their proximity to the desired page, the theory being that the process may need information that spans several consecutive pages. This reduces the number of page faults generated and the amount of disk operations.

Placement

Placement refers to the location a page is stored in RAM. The virtual memory manager chooses the first free page in RAM it finds.

Replacement

If the placement policy fails because there are no free pages in RAM, the replacement policy is used to determine which page will be moved from RAM to the paging file. Each process has a group of valid pages in RAM; these are called the working set. For simplicity, a first-in, first-out (FIFO) algorithm is used. When a page replacement is required, the oldest page in the working set is moved to the paging file to make room for the new page. Because the replacement scheme is only applied to the working set of the current process, other processes are guaranteed that their pages will not be replaced by other processes.

MAPPED FILE I/O

If an application attempts to load a file that is larger than both the system RAM and the paging file combined, the mapped file I/O services of the virtual memory manager are used. Mapped file I/O enables the virtual memory manager to map virtual memory addresses to a large file, inform the application that the file is available, and then load only the pieces of the file that the application actually intends to use. Because only portions of the large file are loaded into memory (RAM or page file), this greatly decreases file load time and system resource drainage. This is a very useful service for database applications that often require access to huge files.