AMD athlon 64 User Manual

Page of 48
46
Appendix A
40555
Rev. 3.00
June 2006
Performance Guidelines for AMD Athlon™ 64 and AMD Opteron™ 
ccNUMA Multiprocessor Systems
The function to set memory affinity for a thread is VirtualAlloc( )[
9
]. This function gives the 
developer the choice to bind memory immediately on allocation or to defer binding until first touch. 
].
]. 
By default, Windows uses the first touch binding policy for all data. Once memory is bound to a node, 
it normally resides on that node for its lifetime.
A.8
Tools and APIs for Node Interleaving  in Various 
OSs for AMD64 ccNUMA Multiprocessor Systems
This section discusses tools and APIs available for performing node interleaving under various 
operating systems.
A.8.1
Support under Linux
®
5
].
numactl is a command line tool, which is used for node interleaving all memory accessed by a 
program across a set of chosen nodes.
For example, to interleave all memory accessed by program foobar on nodes 0 and 1, use: 
numactl --interleave=0x03  foobar
libnuma offers several functions a program can use to interleave a given memory region across a 
set of chosen nodes.
Linux only supports the round robin node interleaving policy.
A.8.2
Support under Solaris
Solaris offers an API called madvise, which can be usedwith the MADV_ACCESS_MANY flag to 
tell the OS to use a memory policy that causes the OS to bind memory randomly across the nodes. 
This offers behavior similar to the round robin node interleaving of memory offered by Linux. 
This random policy is the default memory placement policy used by Solaris for shared memory.
A.8.3
Support under Microsoft
®
 Windows
®
Microsoft Windows does not offer node interleaving.