10 Aug 2012 |
Lockless MPI version 1.2 is Released
Lockless Inc. has changed the technique used to share memory between MPI ranks. Now any programming language can be used, rather than just C. We use a novel "meta-process" technology that combines the attributes of threads and processes. The result is superior performance for all language bindings. This version supports C, C++ and FORTRAN.
|
23 Jul 2012 |
Lockless Allocator Updated
A minor update to the Lockless Allocator is released. This provides support for the ISO C11 aligned_alloc() function. It also adds the ability to hook the underlying sbrk() calls.
|
22 May 2012 |
Paper Released
Rahul Manghwani and Tao He at NYU have released a paper called Scalable Memory Allocation (Local Mirror). They benchmark the Lockless Memory Allocator and TCmalloc with up to 64 threads. They have many graphs which show how the absolute performance of the allocators varies.
|
24 Feb 2012 |
Lockless Allocator Updated
A minor update to the Lockless Allocator is released. This fixes a problem with the Microsoft Windows version, where slab allocations may spuriously fail on machines with more than 4GiB of memory. (The underlying problem was due to the different size of an unsigned long on linux versus windows.)
|
14 Nov 2011 |
The Lockless Memory Allocator has been released under the GPL version 3 License
We are proud to announce that the Lockless Memory Allocator has now been open-sourced. Check out the downloads page to obtain it. Try benchmarking your important applications to see if it provides any speed-ups. Proprietory users can still contact us for a commercial license if you wish to use the allocator in closed-source software.
|
10 Nov 2011 |
Lockless MPI version 1.0 is released
Lockless Inc. is proud to present a highly efficient MPI implementation of the complete MPI 1.3 spec. It uses lock-free techniques for superiour performance. It is designed for modern 64bit multi-core systems, and uses threads instead of processes. It can be over an order of magnitude faster in some cases due to being limited by the speed of L1 cache, rather than main memory.
|
29 Apr 2011 |
New Version of the Lockless Memory Allocator, and new benchmark results
The memory allocator benchmarks have been redone with new versions of jemalloc and tcmalloc. Both allocators have improved greatly over the past few months. Version 1.3 of the Lockless memory allocator has been released. It has reduced fragmentation and memory usage. It is typically a few tens of percent faster than version 1.2.
|
19 Jan 2011 |
New Benchmark Numbers
The memory allocator benchmarks have been redone yet again. The latest version of jemalloc no longer segfaults with large number of threads, so has been added. (It still needs a work-around for non-compliant memalign() behaviour though.) Hoard has been updated to the latest version.
|
9 Dec 2010 |
New Benchmark Numbers
The memory allocator benchmarks have been redone. The machine now has 11GiB of ram, so the maximum allocation size has been raised to 1MiB. A more recent kernel and eglibc has been used, which alters the results somewhat. The biggest change is the replacing of jemalloc with nedmalloc due to the former segfaulting.
|
8 Dec 2010 |
New Version the the Lockless Memory Allocator Released
A new version with even more performance is released. This version is much faster on Microsoft Windows, and with larger allocation sizes. It has decreased contention within the slab subsystem. It uses a slightly different algorithm for returning blocks to the operating sytem that vastly reduces in-kernel contention.
|
8 Dec 2010 |
t-test1.c Updated
The t-test1 benchmark has been updated to have defaults that are close to those used in the graphs on this website. It also has had minor code changes to prevent symbol clashes with Microsoft Windows header files.
|