We learned a lot about mmap in p6!
How does our version of mmap differ from linux?
The goal of this exercise is to
1. Become familiar with using mmap on a real system
2. Run some experiments!
Task 1:
Let's start with populate.cpp.
In class we implemented demand paging. We never read in any data unless we page
fault at that address. This was a design decision, we could have read in the
entire file when mmap was called. This has a number of benefits, especially if
some of the data is never read. It turns out this might not always be the
behavior you want.
Implement populate.cpp
./populate with no arguments should mmap a file using demand paging
./populate with any additional arguments should not demand page, but instead
read everything into memory when mmap is called.
After calling mmap, write a benchmark which performs worse with demand paging
than with all-at-once paging.
Task 2:
Moving on to huge.cpp
On the previous quiz we considered some tradeoffs between small and large page
sizes. Now let's see how these tradeoffs play out in the real world!
implement huge.cpp
./populate with no arguments should mmap a region using 4k pages
./populate with any additional arguments should mmap a region using 2MB pages.
After calling mmap, write a benchmark which performs better with huge pages
than with 4k pages.