This repository contains NBBS, a lock-free back-end allocator that adheres to the buddy system specification.
It also keeps several benchmarks and modules to evaluate NBBS and the Linux Kernel Buddy System.
The code targets x86_64 architectures equipped with Linux.
Here, you can find several buddy system allocators. In particular:
- 1lvl-nb: this is the classical NBBS implementation discussed in several papers [Cluster'18, CCGrid'19];
- 4lvl-nb: this is the memory optimized version of our NBBS allocator (16x compress ratio);
- the spin-locked version of the above-mentioned allocators (1lvl-sl and 4lvl-sl).
You can configure the allocator by setting the following macro at compile time:
- MIN_ALLOCABLE_BYTES
- MAX_ALLOCABLE_BYTES
- NUM_LEVELS
We provide several benchmarks to evaluate NBBS and other back-end allocators. The benchmarks are built to run also on kernel side, so you can even evaluate the Linux-Kernel buddy system.
- Linux scalability: each thread makes a burst of allocations followed by a burst of memory release operations.
- Thread test: it operates like linux scalability, but the length of bursts is divided by the number of threads.
- Costant occupancy: each thread pre-allocates blocks of different order and makes a burst of allocations followed by a burst of memory release operations.
- Cached allocation: each thread repeatedly allocates and releases an individual memory buffer.
In order to run the benchmark to evaluate the Linux Buddy System (kernel-sl), you need to mount the kernel-bd-api module.
- Compile just typing
make - In each benchmark folder you will find one executable for each allocator. Run them by typing
./TB_<bench_name>-<allocator> <num_of_threads> <mem_size>
For further information about NBBS allocators, please send an email to:
marotta at diag dot uniroma1 dot it
Current:
- Romolo Marotta
- Mauro Ianni
- Alessandro Pellegrini
- Francesco Quaglia
Former:
- Andrea Scarselli