Home

Awesome

scalloc Build Status

Note

We completely rewrote the allocator from scratch. Please use the current version to obtain benchmarking numbers for scalloc. See Benchmarking.

A Fast, Multicore-Scalable, Low-Fragmentation Memory Allocator

The problem of concurrent memory allocation is to find the right balance between temporal and spatial performance and scalability across a large range of workloads. Our contributions to address this problem are: uniform treatment of small and big objects through the idea of virtual spans, efficiently and effectively reclaiming unused memory through fast and scalable global data structures, and constant-time (modulo synchronization) allocation and deallocation operations that trade off memory reuse and spatial locality without being subject to false sharing. We have implemented an allocator, scalloc, based on these ideas that generally performs and scales in our experiments better than other allocators while using less memory and is still competitive otherwise.

Contents

Disclaimer

While the code provided in this repository is complete in the sence that it supports the full API, it may still require tuning for a specific production environment (read: your environment).

Binaries

We do not explicitly provide pre-built binaries. However, binaries for Linux (x86_64) can be found in the release section of this repository.

Building from source

We support building on OS X (>10.6) and Linux-based systems using gyp. The only requirement to actually get started is a working git installation.

Setting up a build environment

Checkout the latest version

git clone https://github.com/cksystemsgroup/scalloc.git
cd scalloc

If you don't have a global gyp installation, you can get a local one using

tools/make_deps.sh

Generate a build environment (using the gyp installation from the previous step)

build/gyp/gyp --depth=. scalloc.gyp

Additionally, scalloc provides some compile-time configuration flags:

Flags may be set when creating the build files using gyp by passing them as flags, i.e., -Dflag=value. For example, -Dreuse_threshold=20.

We support the following build configurations:

... on Linux

After setting up the build environment building scalloc is as easy as

BUILDTYPE=Debug make # default
BUILDTYPE=Release make

... on OSX

Open scalloc.xcodeproj and build the project using Xcode, or build it from the command line using

build/gyp/gyp --depth=. scalloc.gyp --build=Release

Using scalloc

... on Linux

In order to make use of scalloc, preload it using LD_PRELOAD.

LD_PRELOAD=/path/to/libscalloc.so ./foo

scalloc heavily makes use of 64bit address space. If you run into mmap limits you need to disable overcommit accounting. Additionally, make sure that transparent huge pages are disabled. On recent versions of Linux you can do this by

sudo sh -c "echo 1 > /proc/sys/vm/overcommit_memory"
sudo sh -c "echo never > /sys/kernel/mm/transparent_hugepage/enabled"

See the kernel docs on overcommit accounting and transparent hugepages for more information.

... on OSX

Similar to preloading on Linux, one can preload scalloc using

DYLD_INSERT_LIBRARIES=/path/to/libscalloc.dylib DYLD_FORCE_FLAT_NAMESPACE=1 ./foo

Benchmarking

See cksystemsgroup/scalloc-artifact for setting up a benchmarking environment to compare scalloc against other allocators.