NVAlloc: Rethinking Heap Metadata Management in Persistent Memory Allocators

NVAlloc is a heap memory allocator for persistent memory (e.g., Intel Optane DCPMM); it emphasizes efficiently eliminating cache line reflushes and small random writes and alleviating slab-induced memory fragmentation in heap metadata management. NVAlloc has three new techniques:

• Interleaved Mapping
• Log-structured Bookkeeping
• Slab Morphing

Please read the following paper for more details:

Zheng Dang, Shuibing He, Peiyi Hong, Zhenxin Li, Xuechen Zhang, Xian-He Sun, and Gang Chen. NVAlloc: rethinking heap metadata management in persistent memory allocators. ASPLOS 2022.

Warning

The code can only be used for academic research.

We use part of jemalloc's code in our source code, please check jemalloc's COPYRIGHT.

Directories

• bin : output executable files
• lib : output library files
• include : header files
• include/internal/nvalloc_internal.h : It includes all header files, and should be included by all source files.
• src : source files
• test/src: test source files

System Requirements

1. 2nd Generation Intel Xeon Scalable Processors
2. Intel® Optane™ Persistent Memory 100 Series
3. cmake (version > 3.10)
4. libpmem
5. libjemalloc

Build

autogen.sh [d/r]
autogen.sh [d/r] [testname]

autogen.sh encapsulates CMakeLists.txt:

• [d/r] : debug/release
• [testname] : build test/src/testname.cpp

Other

Default pmem directory: /mnt/pmem/nvalloc_files/

Example

./autogen.sh r example    # test/src/example.cpp
./bin/example