4

I couldn't find any information via search to determine this via the Intel or HP website. Can you enable NUMA nodes on this? How can you tell the side of each node if so? Is it always best practice to enable and size VMs based on this if it's available or does that depend if you're running an application like SQL that makes use of this?

ewwhite
  • 194,921
  • 91
  • 434
  • 799
Untalented
  • 323
  • 1
  • 4
  • 14

3 Answers3

5

Since Intel launched their Nehalem (55xx-series) all =>2 socket Xeons are NUMA by default, including the E5-2630's. Some BIOS/UEFI's allow this to be disabled but the option is usually hidden away to discourage it.

You don't mention which hypervisor you're using but certainly any recent server-class products from VMWare, MS and Xen/KVM are NUMA-aware and will automatically attempt to keep any VMs vCPU on a single NUMA node - this becomes less easy the more congested your hosts become obviously but basically don't worry about it, leave it to the hypervisor.

Chopper3
  • 100,240
  • 9
  • 106
  • 238
4

Yes, the DL360p is NUMA-capable. It's the system default.

Assuming VMWare, the handling of VMs on NUMA architecture has changed slightly as the product evolved. See this excellent site for a good breakdown.

With the current edition of VMWare (5.x), you will end up with NUMA nodes inside the guest if you configure more vCPUs than exist on on physical NUMA node. So >8 in your particular server's case. From: http://vxpertise.net/?p=124

Assigning more than 8 vCPUs to a VM spread across multiple sockets create virtual NUMA nodes inside the guest allowing for better scheduling decisions in the guest.

Also: HP's NUMA Whitepaper.

ewwhite
  • 194,921
  • 91
  • 434
  • 799
0

You can not "disable" or "enable" NUMA. It is how hardware was designed to overcome physical limitations and costs.

When you "disable" it, you actually activate node interleaving on a NUMA architecture. This is just a memory allocator which will hide NUMA topology to OS. All modern hypervisors, OSes, .Net/Java platforms and big applications have support for NUMA and they will schedule and balance the load to utilize the architecture in an efficient way. Bug are still squashed as we are speaking :D

Best is to leave NUMA enabled and install recent versions of software that have proper support for NUMA. By software, include also UEFI firmware updates which can contain microcode updates. Microcode updates can be loaded into CPU by OS during each boot (similar to what firmware is doing).

Mircea Vutcovici
  • 16,706
  • 4
  • 52
  • 80