Linux Memory Allocation

AI Summary

An AI-generated summary of this video lecture.

This lecture introduces Linux memory allocation, beginning with the Buddy Allocator and progressing through SLOB, SLAB, and SLUB allocators before covering Huge Pages and NUMA. The Buddy Allocator manages physical memory in contiguous page blocks whose sizes are powers of two, splitting larger blocks on demand and merging adjacent same-size buddy blocks on free to reduce external fragmentation. A worked example traces an 8-page block splitting into 4-page and then 2-page blocks, followed by merging two adjacent 2-page buddy blocks into a 4-page block. The SLOB Allocator is presented as a simple, low-overhead allocator for small-memory systems that maintains a list of free blocks and uses first-fit search; its limitations include external fragmentation and slow linear searching, and it was removed from the Linux kernel in version 6.8. A four-row worked example demonstrates free-list operations: initial state, allocating 20 KB, allocating 15 KB, and freeing the first 20 KB block, with a memory bar from address 0 to 99. The SLAB Allocator maintains caches of pre-allocated objects organized as Cache → Slab (one or more pages) → Objects, where slabs contain multiple objects of the same size created in advance for fast reuse. The SLUB (Unqueued Slab) Allocator is then introduced as the default slab allocator in modern Linux, with a simpler data structure and better performance; its five-step working flow covers kernel request through object return, including what happens when an object is freed. The lecture concludes by introducing Huge Pages as larger-than-normal pages to reduce memory-management overhead and NUMA (Non-Uniform Memory Access) architecture, with a simple example contrasting local versus remote memory access speeds.

Chapters

1. 0:00 – 2:00 00:00-02:00

   The slide titled 'Linux Memory Allocation' defines the Buddy Allocator as a Linux kernel mechanism for managing physical memory in blocks of contiguous pages. Bullets state it divides free memory into blocks whose sizes are 'powers of 2 pages,' splits a larger block when a smaller one is required, and merges adjacent same-size blocks with their buddy on free. It provides efficient allocation of 'physically contiguous pages' and helps reduce 'external fragmentation.' In red handwriting beneath the bullets, a powers-of-two sequence is written: '1, 2, 4, 8, 16' at first, then extended to '32, 64.'

2. 2:00 – 5:00 02:00-05:00

   A worked example slide shows two stacked diagrams. The blue-headed 'BUDDY ALLOCATOR – SPLITTING' panel traces an 8 Pages block split into two 4 Pages blocks, then a 4 Pages block split into 2 Pages and 2 Pages alongside a remaining 4 Pages block. The green-headed 'BUDDY ALLOCATOR – MERGING' panel shows two adjacent yellow '2 Pages' blocks under a 'Buddy Blocks' bracket combining via a 'Merge' arrow into one green '4 Pages' block. Side 'Key Points' boxes list rules such as blocks of size 2^n pages (1, 2, 4, 8...) and that merging reduces fragmentation. The instructor underlines and circles key terms on the Buddy Allocator slide before transitioning to SLOB Allocator details, pointing out an 'Important Current-Linux Note' regarding SLOB removal.

3. 5:00 – 10:00 05:00-10:00

   The SLOB Allocator slide describes it as a 'very simple, low-overhead kernel allocator intended for small-memory systems.' It maintained a 'simple list of free memory blocks' and used a 'first-fit style search to find a suitable free block.' A warning icon marks the note that SLOB is not present-day default and was removed from the Linux kernel in version 6.8, with SLUB becoming the default slab allocator. A 'Key Point' states SLOB focuses on simplicity and low memory overhead rather than maximum performance, with limitations including external fragmentation and slower linear searching. A four-row diagram then walks through memory operations: '1. Initial State,' '2. Allocate 20 KB,' '3. Allocate 15 KB,' and '4. Free First Block (20 KB),' each pairing a Free List table (Start Address, Size in KB) with a Memory bar marked 0 to 99, segments shifting between green 'Free' and red/yellow 'Allocated' blocks. Red handwritten checkmarks, underlines, and arrows link Free List entries to memory segments.

4. 10:00 – 15:00 10:00-15:00

   The SLAB Allocator slide presents a definition line, five bullet points including 'Maintains caches of pre-allocated objects,' and a vertical diagram: Cache → Slab (Page or Multiple Pages) → a row of Object boxes. Side captions read 'Collection of slabs for a specific object type,' 'A slab contains multiple objects of the same size,' and 'Objects are created in advance and reused for fast allocation.' A red circle highlights 'Objects of the same type/size,' and a large red curly brace groups the three levels. Sections for 'Working of SLAB Allocator,' 'Advantages of SLAB,' and 'Limitations of SLAB' are indicated. Red underlines progressively appear across the definition, bullets, and diagram captions.

5. 15:00 – 20:00 15:00-20:00

   The SLUB (Unqueued Slab) Allocator is introduced as the default slab allocator in modern Linux kernels, with a simpler data structure and better performance compared to SLAB. A visual flowchart illustrates the 5-step working flow of SLUB, from kernel request to return to kernel. The lecture discusses 'When an Object is Freed' and the 'Advantages of SLUB,' with red underlines highlighting key terms like 'simpler data structure' and 'performance and scalability.' The lesson then transitions to 'Huge Pages,' described as larger-than-normal memory pages to reduce overhead, and introduces 'NUMA (Non-Uniform Memory Access)' architecture and its impact on memory access time, with a 'Simple Example (NUMA)' diagram showing local versus remote memory access speeds.

6. 20:00 – 20:59 20:00-20:59

   The final sampled frames continue the NUMA discussion, with the 'Simple Example (NUMA)' diagram visible showing local versus remote memory access speeds. The instructor appears to be concluding the lecture on Linux Memory Allocation, having covered Buddy Allocator, SLOB, SLAB, SLUB, Huge Pages, and NUMA. No new slides or major transitions are evident in this final window beyond the continuation of the NUMA example.

The lecture follows a clear pedagogical progression through Linux memory allocation mechanisms. It begins with the Buddy Allocator, establishing the foundational concept of power-of-two block sizes and the split/merge operations that manage physical contiguous pages while reducing external fragmentation. The worked example with 8-page, 4-page, and 2-page blocks makes the abstract split/merge rules concrete. The SLOB Allocator is then presented as a simpler alternative for constrained systems, with its first-fit search and list-based free block management contrasted against the Buddy Allocator's structured approach; its removal in kernel 6.8 signals a shift toward more sophisticated allocators. The four-row free-list example provides a practical demonstration of allocation and deallocation tracking. The SLAB Allocator introduces object-level caching with the Cache-Slab-Object hierarchy, addressing the need for fast allocation of frequently used kernel objects. SLUB refines this with a simpler data structure and better scalability, becoming the modern default. The lecture concludes by broadening scope to hardware-level memory management with Huge Pages and NUMA, connecting software allocation strategies to physical memory architecture. Key transitions include the move from page-level (Buddy) to object-level (SLAB/SLUB) allocation, and from software allocators to hardware memory topology.