“Out Of Memory” Does Not Refer to Physical Memory
Is Virtual memory infinite?
I have been asked in an interview if virtual memory is infinite? I answered saying that it is not infinite. Then the interviewer asked the explanation and what I suggested was that in windows we do have a manual way to configure virtual memory to a certain limit.
I would like to know if Virtual memory is really Infinite?
A
First of all, forget the idea that virtual memory is limited by the size of pointers on your machine.
Virtual memory limits are not the same as addressing space. You can address more virtual memory than is available in your pointer-based address space using paging.
- Virtual memory upper limits are set by the OS: for example, on 32-bit Windows the limit is 16TB, and on 64-bit Windows the limit is 256TB.
- Virtual memory is also physically limited by the available disc space.
For an excellent overview, which addresses various misconceptions, see the following:
COMMENTS
-
Yes @stusmith i think you are right about You can address more virtual memory than is available in your pointer-based ,and it is whole idea of using virtual memory. – Amit Singh Tomar Jul 7 '11 at 10:42
-
1
Virtual memory is not exactly limited by disk space, it can be allocated but not mapped, so-called "reserved virtual memory". See msdn.microsoft.com/en-us/library/windows/desktop/… and MEM_RESERVE for explanation. – Sergey Alaev Apr 6 '17 at 9:50
A
At the very least, the size of virtual memory is limited by the size of pointers on given platform (unless it has near/far pointers and non-flat memory model). For example, you cannot address more than about 2^32 (4GB) of memory using single 32-bit pointer.
In practice, the virtual memory must be backed up with something eventually -- like a pagefile on disk -- so the size of storage enforces a more practical limit.
-
Sorry, that's just wrong. Look up "PAE" (Physical Address Extension) for example. – stusmith Jul 7 '11 at 10:18
-
1
@stusmith : A PAE-enabled Linux-kernel requires that the CPU also support PAE. So, it is limited by the computer architecture, right? – Priyank Bhatnagar Jul 7 '11 at 10:28
- 2
@logic_max: Yes, but a 32-bit Intel chip is capable of supporting PAE. Maybe a better way of putting it is: it is the lowest value of [chip support, OS limit, disk space]. Usually that lowest value is disk space. – stusmith Jul 7 '11 at 10:30
@stusmith I don't think that ARM, PowerPC, MIPS, or many other CPUs have PAE. Neither do older x86 CPUs. There is no right answer to this poor question. In fact, depending on how you count (for the system, for a process), the answer is different, but still not infinite. Down-vote anyone who says it is infinite. – artless noiseMay 30 '14 at 17:47
- 1
It is also not true that everything must be backed by disk. A large zero initialized array can all be backed by a single page of zeros (typically 4-8k). Any read from any address will show zero. On the first write to a page, a fault happens and then the page is allocated (needs disk or memory) and remapped. This can allow large sparse arrays. – artless noise Jun 2 '14 at 20:52
The sum of the virtual address spaces of all the processes running at once can be larger than what the 'user pointer' register can address - since the 'process id' is effectively part of the virtual address. In some cases these "process-unique" addresses actually point to the same PM (e.g. shared libraries, data shared after fork() ) - but even allowing for that, it's possible for the total amount of truly distinct VM pages to exceed what one process can address. – greggo Aug 13 '14 at 16:51