Memory Allocation

[Aseliot]

Can someone tell in a simple manner WHY you would choose to implement either best fit, worst fit, power of two or buddy memory allocation?

I understand that some of these ways of allocating can lead to more fragmentation in certain situations and that sometimes a combination of them might be the best solution.

 

But I can't really find the the situations in which it would be best to use either Best Fit, Worst Fit, Power of Two or Buddy.

Also what kind of program would have the most benefit/reason to use a buddy system? (Specific kind of task?)

[DnFx91]

7 minutes ago, Aseliot said:

Can someone tell in a simple manner WHY you would choose to implement either best fit, worst fit, power of two or buddy memory allocation?

I understand that some of these ways of allocating can lead to more fragmentation in certain situations and that sometimes a combination of them might be the best solution.

 

But I can't really find the the situations in which it would be best to use either Best Fit, Worst Fit, Power of Two or Buddy.

Also what kind of program would have the most benefit/reason to use a buddy system? (Specific kind of task?)

intuitively i would assume that if a program's resources are all in one continuous block of RAM, that it will give the best performance in all cases. Can't you just fix the wastage issue by using more RAM ?>

[Aseliot]

18 minutes ago, DnFx91 said:

intuitively i would assume that if a program's resources are all in one continuous block of RAM, that it will give the best performance in all cases. Can't you just fix the wastage issue by using more RAM ?>

Eh, I understand you can reclaim the space by joining the fragmentend space together. But it is an expensive operation.

 

But my question is not really about a solution to a problem, it's more about WHY would you use one or the other way of allocating specifically over the other and in what situation.

[DnFx91]

16 minutes ago, Aseliot said:

Eh, I understand you can reclaim the space by joining the fragmentend space together. But it is an expensive operation.

 

But my question is not really about a solution to a problem, it's more about WHY would you use one or the other way of allocating specifically over the other and in what situation.

sorry i think this might be a bit beyond my skillset. seems more like a question for github or reddit to be honest, i've always seen LTT forum as more exclusively hardware based with software, programming and computer science as an afterthought, and this seems to be reflected in most of the users i have spoken to. But that may be simply because i don't usually browse these kind of topics

 

Hopefully someone has your answer

[vorticalbox]

4 hours ago, Aseliot said:

Eh, I understand you can reclaim the space by joining the fragmentend space together. But it is an expensive operation.

 

But my question is not really about a solution to a problem, it's more about WHY would you use one or the other way of allocating specifically over the other and in what situation.

well I covered this a little in uni but never really went into why to use one over the other.

 

@Nuluvius may have more in sight.

[Mira Yurizaki]

Like any other algorithm, it depends on what your program does and what your needs are.

 

If you need as little memory fragmentation as possible and you don't care about performance, use best fit.

If you want your applications to have plenty of room to grow because you don't want to call the memory allocation function often (because it's an expensive operation), use worst fit.

If you want a balance between the two, use buddy memory allocation.

 

Keep in mind for x86 systems, you're limited to 4KB pages of memory. Other processors may not have this limitation, but they may have others. ARM for instance suffers in performance if you make memory accesses in addresses not aligned to 16-bit addresses.

[Unimportant]

9 minutes ago, M.Yurizaki said:

Keep in mind for x86 systems, you're limited to 4KB pages of memory. Other processors may not have this limitation, but they may have others. ARM for instance suffers in performance if you make memory accesses in addresses not aligned to 16-bit addresses.

Also relevant for x86 is that these 4KB pages are mapped from physical memory to virtual memory. A normal program only sees this virtual memory. The operating system can and does a lot of memory defragmentation already by remapping freed pages so they can be used again for something else.