43 lines
1.5 KiB
Plaintext
43 lines
1.5 KiB
Plaintext
On some larger systems, memory has non-uniform access times. It is organized
|
|
in so-called NUMA-nodes. A NUMA node is a part of memory which is accessible
|
|
faster for some particular subset of processors.
|
|
|
|
# lscpu
|
|
Architecture: x86_64
|
|
CPU op-mode(s): 32-bit, 64-bit
|
|
Byte Order: Little Endian
|
|
CPU(s): 24
|
|
On-line CPU(s) list: 0-23
|
|
Thread(s) per core: 2
|
|
Core(s) per socket: 6
|
|
Socket(s): 2
|
|
NUMA node(s): 2
|
|
Vendor ID: GenuineIntel
|
|
CPU family: 6
|
|
Model: 45
|
|
Stepping: 7
|
|
CPU MHz: 1200.000
|
|
BogoMIPS: 4588.34
|
|
Virtualization: VT-x
|
|
L1d cache: 32K
|
|
L1i cache: 32K
|
|
L2 cache: 256K
|
|
L3 cache: 15360K
|
|
NUMA node0 CPU(s): 0-5,12-17
|
|
NUMA node1 CPU(s): 6-11,18-23
|
|
|
|
This machine has two NUMA nodes and 24 processors (they are called CPUs, but
|
|
that is not very precise). Half of them belongs to NUMA node0 and the second
|
|
half to node1.
|
|
|
|
When a processor accesses data that lies in its NUMA node, it is faster than
|
|
accessing data from some other node. The local access is called LMA and the
|
|
remote one is called RMA.
|
|
|
|
The NUMATOP tool watches what happens around the NUMA nodes and it sees both
|
|
LMAs and RMAs. If the machine has only one NUMA node, only LMAs can be seen.
|
|
|
|
The test tries to trigger LMAs and check whether NUMATOP registers it. If we
|
|
have more than one NUMA node, we test RMAs as well.
|
|
|