For those of you running multi-socket Opteron servers (and eventually, multi-socket Nehalem servers), AMD published a whitepaper last week on their work adapting Google's tcmalloc to be NUMA-aware. The whitepaper includes links to their source code / diffs. It appears to be quite a performance boost in their (very artificial) benchmark. I'll be trying it out soon myself.
Howard Chu wrote:
For those of you running multi-socket Opteron servers (and eventually, multi-socket Nehalem servers), AMD published a whitepaper last week on their work adapting Google's tcmalloc to be NUMA-aware. The whitepaper includes links to their source code / diffs. It appears to be quite a performance boost in their (very artificial) benchmark. I'll be trying it out soon myself.
Doh. Forgot to paste the link:
http://developer.amd.com/Assets/NUMA_aware_heap_memory_manager_article_final...
Howard Chu wrote:
Howard Chu wrote:
For those of you running multi-socket Opteron servers (and eventually, multi-socket Nehalem servers), AMD published a whitepaper last week on their work adapting Google's tcmalloc to be NUMA-aware. The whitepaper includes links to their source code / diffs. It appears to be quite a performance boost in their (very artificial) benchmark. I'll be trying it out soon myself.
Doh. Forgot to paste the link:
http://developer.amd.com/Assets/NUMA_aware_heap_memory_manager_article_final...
http://people.cs.vt.edu/~scschnei/streamflow/
This paper from 2006 indicates that Streamflow is even faster than tcmalloc, with comparable memory footprint. Has anyone here used it? I don't recall having heard of it when I was comparing allocators in 2007.
Hi
On 23 May 2009, at 07:09, Howard Chu wrote:
Howard Chu wrote:
Howard Chu wrote:
For those of you running multi-socket Opteron servers (and eventually, multi-socket Nehalem servers), AMD published a whitepaper last week on their work adapting Google's tcmalloc to be NUMA-aware. The whitepaper includes links to their source code / diffs. It appears to be quite a performance boost in their (very artificial) benchmark. I'll be trying it out soon myself.
Doh. Forgot to paste the link:
http://developer.amd.com/Assets/NUMA_aware_heap_memory_manager_article_final...
http://people.cs.vt.edu/~scschnei/streamflow/
This paper from 2006 indicates that Streamflow is even faster than tcmalloc, with comparable memory footprint. Has anyone here used it? I don't recall having heard of it when I was comparing allocators in 2007.
See this quote by the streamflow author on http://apps.ycombinator.com/item?id=361498
Then there's my work, which is frankly not suitable for using in a real application since it's tested as heavily as the other two I mentioned: http://people.cs.vt.edu/~scschnei/streamflow
That quote is, however, 192 days old. Things may have changed.
Oskar
-
Howard Chu -
Oskar Pearson