Memory Management -Memory allocation -Garbage collection

Memory Allocation Memory pool: large block of contiguous memory Memory manager allocates memory by returning a handle to the user Use the term heap to refer to free memory accessed by a dynamic memory management scheme

Dynamic Allocation Blocks of any size may be requested in any order from the free-list For a request of m words, and a block size k, between m and k space is used for the request This can result in fragmentation if m != k

Fragmentation in Dynamic Allocation External fragmentation: lots of small free blocks Internal fragmentation: when all of block of size k is allocated for m words. This type of allocation is easier, if less efficient

Sequential Fit Method Attempt to find a “good” block The free-list is organized as a doubly- linked list Tag bit and block size fields The memory manager searches the free- list for a block of “suitable” size

Three sequential fit methods First fit –Start from the beginning (or middle) –May waste larger blocks by breaking them up Best fit –Examines entire list –Maximizes external fragmentation but will be more likely to be able to service large requests

Three sequential fit methods Worst fit –Allocates largest block through a sequential search –Minimizes external fragmentation Which is best? Depends on the expected types of memory requests

Sequential Fit A search of the free-list is in Ө(n) in the worst case Want to merge adjacent free blocks Need additional space to support the memory manager operations/linked list Is there anything that can be improved?

The Buddy Method Assume that memory is of size 2ⁿ for some n Both free and reserved blocks will be of size 2 k for k ≤ n The buddy system keeps a separate list of free blocks for each size

The Buddy Method For a request of size m, find the smallest k where 2 k ≥ m If such a k exists, allocate a block of size 2 k from the list If such a k does not exist, allocate the next larger block on the list, and split it in half until a block of size 2 k is created

The Buddy Method Advantages –Less external fragmentation –Cheaper search than a linked list –Merging adjacent blocks is easy (the buddy for any block of size 2 k is another block of the same size with the same address except the k th bit is reversed) Disadvantage –Allows internal fragmentation

Other memory allocation methods Segregated storage method: break available memory into several memory zones, each with its own management method Impose a standard size (cluster scheme) –Example: disk file management –Leads to internal fragmentation –Does not need to be contiguous

Failure Policies Happens when a memory request for a certain size cannot be serviced –Due to external fragmentation → compact memory, which physically moves data –Use a handle if the application relies on absolute positions of the data –Can defer the memory request (for example, when several processes are running at once)

Failure policies: Garbage Collection When no program variable points to a block of space, it is considered garbage (or a memory leak) Garbage collection involves determining which memory is garbage and recovering it Two common methods: –Reference count –Mark/sweep strategy

Garbage Collection Reference Count: each dynamically allocated memory block has a count field that is incremented and decremented for each pointer pointing to it (or away from it) –When the count reaches zero, the memory becomes garbage and is immediately placed in free store –Used by the UNIX file system, where the memory objects are linked together without cycles –Useful when the objects are large, such as a file

Garbage Collection Mark/sweep strategy –Uses a single bit marker instead of a count field –Works for cycles, but DFS is recursive Garbage collection phase occurs when free store is exhausted: –Clear all mark bits –Perform DFS from each pointer on the variable list, turning on the bits –Sweep through the memory pool for unmarked elements (which are garbage and placed in free store)