Hyper-V vNUMA enable

https://blueprints.launchpad.net/nova/+spec/hyper-v-vnuma-enable

Windows Hyper-V / Server 2012 introduces support for vNUMA topology into Hyper-V virtual machines. This feature improves the performance for VMs configured with large amounts of memory.

Problem description

Currently, there is no support for Hyper-V instances with vNUMA enabled. This blueprint addresses this issue.

Use Cases

NUMA can improve the performance of workloads running on virtual machines that are configured with large amounts of memory. This feature is useful for high-performance NUMA-aware applications, such as database or web servers.

Hyper-V presents a virtual NUMA topology to VMs. By default, this virtual NUMA topology is optimized to match the NUMA topology of the underlying host. Exposing a virtual NUMA topology into a virtual machine allows the guest OS and any NUMA-aware applications running within it to take advantage of the NUMA performance optimizations, just as they would when running on a physical computer. [1]

Hyper-V related restrictions:

  • Hyper-V cannot create instances with asymmetric NUMA topology.

  • Hyper-V cannot guarantee CPU pinning.

Proposed change

If VM vNUMA is enabled, Hyper-V will attempt to allocate all of the memory for that VM from a single physical NUMA node. If the memory requirement cannot be satisfied by a single node, Hyper-V allocates memory from another physical NUMA node. This is called NUMA spanning.

If vNUMA is enabled, the VM can have assigned up to 64 vCPUs and 1 TB memory. If vNUMA is enabled, the VM cannot have Dynamic Memory enabled.

The Host NUMA topology can be queried, yielding an object for each of the host’s NUMA nodes. If the result is only a single object, the host is not NUMA based. Resulting NUMA node object looks like this:

NodeId : 0 ProcessorsAvailability : {94, 99, 100, 100} MemoryAvailable : 3196 MemoryTotal : 4093 ComputerName : ServerName_01

The Host NUMA topology will have to be reported by HyperVDriver when the method get_available_resource is called. The returned dictionary will contain the numa_topology field and it will contain an array with NumaTopology objects, converted to json.

The scheduler has already been enhanced to consider the availability of NUMA resources when choosing the host to schedule the instance on. [2]

Virtual NUMA topology can be configured for each individual VM. The maximum amount of memory and the maximum number of virtual processors in each virtual NUMA node can be configured.

Instances with vNUMA enabled are requested via flavor extra specs [2]:

  • hw:numa_nodes=NN - number of NUMA nodes to expose to the guest.

HyperVDriver must check if the instances require CPU pinning or asymmetric NUMA topology. As they are not supported, it should raise an Exception.

Equivalent image properties can be defined, with an ‘_’ instead of ‘:’. (example: hw_numa_nodes=NN). Flavor extra specs will override the equivalent image properties.

More details about the flavor extra specs and image properties can be found in the References section [2]. The implementation will be done in as similar fashion as libvirt.

Alternatives

None, there is no alternative to enable vNUMA with the current HyperVDriver.

Data model impact

None

REST API impact

None

Security impact

None

Notifications impact

None

Other end user impact

None

Performance Impact

This capability can help improve the performance of workloads running on virtual machines that are configured with large amounts of memory.

Other deployer impact

If the Host NUMA spanning is enabled, virtual machines can use whatever memory is available on the method, regardless of its distribution across the physical NUMA nodes. This can cause varying VM performances between VM restarts. NUMA spanning is enabled by default.

Checking the available host NUMA nodes can easily be done by running the following Powershell command:

Get-VMHostNumaNode

If only one NUMA node is revealed, it means that the system is not NUMA-based. Disabling NUMA spanning will not bring any advantage.

There are advantages and disadvantages to having NUMA spanning enabled and advantages and disadvantages to having it disabled. For more information about this, check the References section [1].

vNUMA will be requested via image properties or flavor extra specs. Flavor extra specs will override the image properties. For more information on how to request certain NUMA topologies and different use cases, check the References section [2].

There are a few considerations to take into account when creating instances with NUMA topology in Hyper-V:

  • Hyper-V cannot guarantee CPU pinning. Thus, the nova HyperVDriver will not create an instance having the hw:cpu_policy flavor extra-spec or hw_cpu_policy image property set to dedicated.

  • Hyper-V cannot guarantee asymmetric instance NUMA topologies and the nova HyperVDriver will not create them. For example, if the instance requires 2GB memory in NUMA Node 0 and 6GB in NUMA Node 1, the instance will not spawn. Same rule applies for the number of vCPUs.

Developer impact

None

Implementation

Assignee(s)

Primary assignee:

Claudiu Belu <cbelu@cloudbasesolutions.com>

Work Items

As described in the Proposed Change section.

Dependencies

None

Testing

  • Unit tests.

  • New feature will be tested by Hyper-V CI.

Documentation Impact

None

History

  • Approved in Liberty.

  • Added Hyper-V related restrictions.