Under the Hood of .NET Memory Management - Simple Talk

Chapter 7: The Windows Memory Model
The L3 cache is stored on the motherboard and is the largest cache store; for example, the
Intel Core I7 processor has an 8 MB L3 cache. It's the slowest of all three caches, but still
faster than direct memory access.
When a memory address is requested, the caches are checked first, before an attempt is
made to access physical RAM, and the OS only actually has to deal with the paging file at
the end of that checking sequence.
The access order is L1–>L2–>L3–>physical memory–>page file.
The intricacies of the data cache could fill a whole new chapter in themselves, but we'll
not go there in this book, as I think we've covered enough for now.
The Long and Winding Road
We've gone from an allocation and garbage collection in .NET, right through object
retention and promotion, and on to subsequent allocation of ephemeral segments.
Beyond segments, we looked at how VirtualAlloc is used to create ephemeral
segments, what VirtualAlloc actually does, and what it translates to. Finally, we've
broken memory allocation down into its raw representation within the OS, and analyzed
how huge Virtual Address Spaces are managed to give processes access to resources far in
excess of their machine's physical memory limitations.
Summary
Maybe you're wondering why you needed to know all this. Well, I suppose the short
answer is because what happens beyond the .NET boundary is no longer a mystery.
.NET is built on top of the Windows API, and so its impacts and limitations come from
there as well.
212