NFVI Platform Capability Registry

From NFVwiki
Revision as of 07:43, 4 October 2018 by Vul (talk | contribs) (Initial content)

This is the ETSI NVS SOL registry for key/value pairs used to represent EPA requirements and settings, as part of NFV VNF descriptors.

WORK IN PROGRESS

Specifying EPA Capability Requirements using TOSCA-based VNF Descriptors

Some VNF suppliers make use of the underlying hardware platform capabilities in order to accelerate performance and optimize throughput of their VNF products. In order to ensure proper instantiation and operation of such VNFs, the VNF descriptor (VNFD) is used to describe VNF-specific EPA capability requirements that will be matched against the capabilities of the underlying hardware infrastructure resources, in order to ensure that appropriate resources are used for instantiation.

Specifying EPA requirements

There are five different types of EPA capability requirements that can be specified in the VNFdescriptor - CPU, memory, storage, network and NUMA. Each requirement type is specified using a TOSCA hash map containing a list of capabilities and their corresponding desired configuration values. The following table describes where within the VNFD the EPA requirement hash maps are defined and where within the VNFD they are used.

EPA Requirements Attribute Name Defined in... Used in...
CPU vdu_cpu_requirements tosca.datatypes.nfv.VirtualCpu tosca.nodes.nfv.VDU.Compute
Memory vdu_memory_requirements tosca.datatypes.nfv.VirtualMemory tosca.nodes.nfv.VDU.Compute
NIC network_interface_requirements tosca.datatypes.nfv.VirtualNetworkInterfaceRequirements tosca.nodes.nfv.VduCp

tosca.nodes.nfv.VnfExtCp

Storage vdu_storage requirements map tosca.nodes.nfv.VDU.VirtualStorage
NUMA logical_node_requirements tosca.datatypes.nfv.LogicalNodeData tosca.nodes.nfv.VDU.Compute

EPA Requirement Key/Value Pair Entry Format

EPA capability requirements are specified using a set of TOSCA hash maps within the VNFD. Each entry in the hash map fully defines a single EPA capability requirement. Within a given hash map entry, the "key" portion of the entry contains the EPA capability name and the "value" portion contains structured data representing the the attributes of the corresponding configuration value.

The "key" portion of each array entry contains a string specifying a particular EPA capability name. The "value" portion of each array entry is a structured field containing information used to interpret the requested capability configuration. The capability values are defined as TOSCA strings, and their content can vary from a single attribute to complex permutations of simple attributes, lists and hash maps. Content and structure of capability values are defined using JSON schemas that can be used to parse the capability value strings in order to determine the desired state of EPA capabilities.

All capability value strings contain a set of pre-defined attributes that are used refine the how the requirements should be treated and what configuration value schema should be used for its interpretation. The following table describes that pre-defined attributes and their purpose:

Array Entry Values
<key> A string specifying the name of a given EPA capability.
<value> A map of key-value pairs describing the configuration value of a given capability, where the key-value pairs are defined as follows
key type value
schemaVersion string A string describing the version the schema to be used for validation and interpretation of the <configuration-value>.
schemaSelector string An Identifier for the particular schema to use.
hardwarePlatform string A string describing what hardware platform associated with a given EPA capability. For vendor-neutral EPA capabilities, a value of "generic" is used.
mandatory string A boolean value expressed in string type specifying whether the requested capability is mandatory for for proper operation of the VNF. The value of "true" indicated that a given EPA capability must be present. The value of "false" indicates that the VNF can function with or without the capability.
configurationValue string A string describing the configuration value of a given capability, as defined by the schema.
... Structured data representing the EPA capability configuration value-specific attributes. Can vary from a single attribute to complex permutations of simple attributes, lists and hash maps.

Supported EPA Capability Requirements

VDU CPU Requirements

Capability Name Capability Value Generic Capability Descriptiopn
cpuModelSpecificationBinding strictBinding

equalOrBetterBinding

VDUs may be developed, compiled, optimized or validated on particular CPU models. Some deployments may wish to permit the VDU to be deployed on a platform with the specified CPU only, or with an alternative CPU with the same architecture, instruction set, and if specified, instruction set extensions, or with a CPU of equivalent or greater capability.
instructionSetRequirements aes, sse, avx, cat, cmt, mbm, ddio, smt, rdrand, etc etc Long list of instruction set extensions.
simultaneousMultiThreading Enabled

disabled

The use of Simultaneous Multi-Threading HW is an efficient way to increase the compute capacity of a platform. SMT HW threads share some CPU core resources. In some VDU implementations, it may be necessary to very explicitly control the HW thread allocation on a platform. This could be to help ensure locality in data caches or as a mechanism to enhance determinism
hypervisorConfiguration HPET

memoryCompaction

kernelSamepageMerging

Long list: High Precision Event Timer configuration, memory compaction, kernel samepage merging, etc.
computeRas pciDetectedAndCorrectedErrors pciDetectedAndUncorrectedErrors Reliability, Availability, Serviceability (RAS)

Long list of values: pciDetectedAndCorrectedErrors, pciDetectedAndUncorrectedErrors

cpuModel List of model identifiers The CPU model for which the VDU has been developed, compiled for, optimized on,  validated on or preferred for some reason.
directIoAccessToCache Values – TBD Descriptions related to cache functions – TBD
accelerator Values – TBD Descriptions related to accelerator functions – TBD
measuredLaunchEnvironment Values – TBD Descriptions related to boot environment functions – TBD
secureEnclave Values – TBD Descriptions related to secure region functions – TBD
numVirtualCpu 1-N Number of virtual CPUs
virtualCpuClock 0-N Minimum virtual CPU clock rate (e.g. in

MHz). The cardinality can be 0 during the allocation request, if no particular value is requested.

logicalCpuPinningPolicy dedicated

shared

Determines if CPUs from the host platform should be committed to the VDU or shared between VDUs.
logicalCpuThreadPinningPolicy require

isolate

prefer

Determines the manner in which CPU (HW) threads are allocated to VDUs. Require means CPU (HW) thread siblings should be allocated

Isolate means allocate CPU (HW) threads from different execution units.

Prefer means ideally allocate CPU HW threads from the same physical execution units but if not available, continue with allocation.

VDU Memory Requirements

Capability Name Capability Value Descriptiopn
memoryPageSize ANY, 4KB, 2MB, 1GB Memory page size
numberOfPages 0..N Number of pages of this specific page size.

Note, The size of memory requested in all instances of the vduMemRequirements must be less than or equal to the virtualMemSize attribute of the virtualMemoryData information element.

memoryAllocationPolicy strictLocalAffinity

preferredLocalAffinity

Strict Local (to node) Affinity or Preferred local (to node) affinity
memoryType Type of memory
memorySpeed Agreed unit of memory speed
memoryRas ECC, SDDC, thermalThrottling, demandAndPatrolScrubbing Long list of memory technologies
memoryBandwidth 0..N Agreed unit of memory bandwidth where 0 is unspecified.
processorCacheAllocationType Values – TBD Agreed type of processor cache allocation
processorCacheAllocationSize 0..N Agreed unit of processor cach

VDU Storage Requirements

Capability Name Capability Value Descriptiopn
storageIops 0..N Required storage characteristics (e.g. speed), including Key Quality Indicators (KQIs) for performance and reliability/availability
storageResilencyMechanism Erasure

tripleReplication

Erasure code based back-end, triple replication based back-end for ensuring data resiliency.

Logical Node Compute Requirements

Capability Name Capability Value Descriptiopn
numberCpu 0..N Number of CPU cores for this logical node. The cumulative number of CPU requests per node must equal the VDU level numVirtualCpu requirement.

Logical Node Memory Requirements

Capability Name Capability Value Descriptiopn
localNumaMemorySize 0..N The amount of memory that needs to be collocated with this specific logical (NUMA) node.

Logical Node i/O Requirements

Capability Name Capability Value Descriptiopn
pciVendorId PCI-SIG vendor ID for the device
pciDeviceId PCI-SIG device ID for the device
pciNumDevices Number of PCI devices required.
pciAddress Geographic location of the PCI device via the standard PCI-SIG addressing model of Domain:Bus:device:function
pciDeviceLocalToNumaNode required

notRequired

Determines if I/O device affinity is required.

Network Interface Requirements

Capability Name Capability Value Description
nicFeature LSO, LRO, RSS, RDMA Long list of NIC related items such as LSO, LRO, RSS, RDMA, etc.
dataProcessingAccelerationLibrary Dpdk Name of the data processing acceleration library required. Orchestration can match any NIC that is known to be compatible with the specified library.
dataProcessingAccelerationLibraryVersion Version Version of the data processing acceleration library required. Orchestration can match any NIC that is known to be compatible with the specified library.
interfaceType Virtio, PCI-Passthrough, SR-IOV, E1000, RTL8139, PCNET Network interface type
vendorSpecificNicFeature TBA List of vendor specific NIC related items.



WORK IN PROGRESS


To register VIM connection information, the following template must be filled and submitted to NFVsupport@etsi.org: template