How to determine if NUMA configuration is enabled or disabled?

环境

  • Red Hat Enterprise Linux 4
  • Red Hat Enterprise Linux 5
  • Red Hat Enterprise Linux 6
  • Red Hat Enterprise Linux 7

问题

  • How to determine if NUMA configuration is enabled or disabled?
  • ​numactl --show​​ does not show multiple nodes

​Raw​

# numactl --show
policy: default
preferred node: current
physcpubind: 0 1 2 3 4 5 6 7 8 9 10 11
cpubind: 0
nodebind: 0
membind: 0

  • ​numactl --hardware​​ does not list multiple nodes

​Raw​

available: 1 nodes (0)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11
node 0 size: 65525 MB
node 0 free: 17419 MB
node distances:
node 0
0: 10

  • grep -i numa /var/log/dmesg returns ​​No NUMA configuration found​

​Raw​

No NUMA configuration found
Faking a node at 0000000000000000-0000001027fff000
Bootmem setup node 0 0000000000000000-0000001027fff000
NODE_DATA [000000000000c000 - 000000000003ffff]
bootmap [0000000000100000 - 0000000000304fff] pages 205

  • Even with NUMA enabled, dmesg does not show any information of NUMA initialization.

决议

  • NUMA should be enabled in the BIOS
  • If NUMA is enabled on BIOS, then execute the command ​​numactl --hardware​​ to list inventory of available nodes on the system
    Example output of ​​numactl --hardware​​ on a system which has NUMA

​Raw​

# numactl --hardware
available: 2 nodes (0-1)
node 0 cpus: 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23
node 0 size: 8157 MB
node 0 free: 88 MB
node 1 cpus: 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31
node 1 size: 8191 MB
node 1 free: 5176 MB
node distances:
node 0 1
0: 10 20
1: 20 10

  • If dmesg does not show any information about NUMA, then increase the Ring Buffer size:
    Boot with 'log_buf_len=16M' (or some other big value). Refer the following kbase article ​​How do I increase the kernel log ring buffer size?​​ for steps on how to increase the ring buffer
  • If the server does not have NUMA support or if the BIOS option is not enabled, then the following messages will be seen in dmesg

​Raw​

No NUMA configuration found
Faking a node at 0000000000000000-0000001027fff000

  • If ACPI is disabled, that will also disable NUMA; verify that ACPI is not disabled by a grub.conf kernel parameter and remove it if found:

​Raw​

$ grep acpi=off /proc/cmdline
$

根源

NUMA has to enabled in the BIOS. If dmesg does not have records of numa initialization during bootup, then it is possible that NUMA related messages in the kernel ring buffer might have been overwritten. Increase the ring buffer so that more messages can be stored. The default kernel ring buffer size is 512 kilobytes.

诊断步骤

1. Check /var/log/dmesg for NUMA related messages

2. Use ​​numactl --hardware​​ to list the numa inventory of the system