virtio-balloon: prioritize most recently used memory when inflating the balloon #2085
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This struct member is unused and can be removed.
This will allow spinlock code to be compiled for both kernel and non-kernel usage without having to use `#ifdef KERNEL` directives.
This change introduces a new heap type that implements a buddy memory allocator and is now used as physical memory heap; this brings the following improvements: - lower memory fragmentation: the buddy allocator keeps track of free contiguous memory areas of each power-of-2 size, and when allocating a requested size, uses a minimum-sized free area to satisfy the request - higher memory re-use (which maximizes TLB hit rate and minimizes host memory usage): when there are multiple minimum-sized memory areas available to satisfy an allocation request, the buddy allocator selects the most recently used area - higher scalability: the buddy allocator does not rely on linear searches on a given memory range to select a free area; therefore, the average CPU use to execute allocation requests does not increase with the size of physical memory The typical overhead of the buddy allocator is around 0.1%, i.e. 99.9% of the physical memory available for a VM can be used for kernel and user program allocations. The size of the bootstrap heap does no longer depend on the size of physical memory, therefore calculation of the memory size is no longer done when unneeded, and the bootstrap heap is set up in common code instead of platform-specific code. The PAGEHEAP_LOWMEM_PAGESIZE constant now sets the upper limit for physical memory allocation requests on low-memory guests, and has been changed to 1 MB to be able to allocate the async queue used by the scheduler. The `pages` kernel heap is now a simple wrapper (whose purpose is to avoid complete exhaustion of physical memory) around the page-backed heap, and is used by both the pagecache and the mmap code.
When inflating the memory balloon to decrease host memory usage, allocating large-sized pages from the physical memory heap is usually not effective in finding previously-used memory areas (which usually are memory ranges that have been unmapped by the user program, or pages that have been evicted from the page cache, and correspond to memory ranges that have actually been populated in the host machine); as a result, balloon inflation does not allow the host hypervisor to significantly decrease its memory usage. This change fixes the above issue by modifying the balloon driver so that it allocates minimum-sized pages when inflating the balloon; this, thanks to the MRU allocation policy of the buddy memory allocator, allows finding previously-used memory areas, which the hypervisor can then release to the host OS.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When inflating the memory balloon to decrease host memory usage, the existing code is not effective in finding previously-used memory areas (which usually are memory ranges that have been unmapped by the user program, or pages that have been evicted from the page cache, and correspond to memory ranges that have actually been populated in the host machine); as a result, balloon inflation does not allow the host hypervisor to significantly decrease its memory usage.
This change fixes the above issue by:
Using a buddy memory allocator as physical memory heap brings the following improvements:
The typical overhead of the buddy allocator is around 0.1%, i.e. 99.9% of the physical memory available for a VM can be used for kernel and user program allocations.