Enhanced Platform Awareness Capability

Owned by Gavin SC
Last updated: Nov 01, 2017 by Srinivasa Addepalli
2 min read

Priority

2

Summary

The capability orchestrate and configure VNFs with EPA requirements.

Use Cases

Performance, Security and resource isolation

R1 Status

Not Supported.

R2 Status



5G has challenging performance requirements which will require some VNFs to utilize hardware optimization.  For example, a VNF may utilize SR-IOV and DPDK to bypass the Linux IP stack and minimize latency.

In order to support this it will be necessary to model and orchestrate VNF requirements and compute host support for attributes such as:

  • Real-time kernel and hypervisor

  • Single Root I/O Virtualization (SR-IOV)

  • CPU Pinning

  • NUMA Awareness

  • Huge Pages

  • PCIe Passthrough (Example: GPUs, FPGA etc..)

  • Resource Director Technology (DDR bandwidth allocation,  Cache size allocation etc..)

  • Accelerated virtual switch using Data Plan Development Kit (DPDK)

  • Accelerated virtual switch using smartNIC

  • Cryptography accelerators such as AES-NI, SIMD/AVX, QAT

  • Random Data Generation using native instructions such as RDRAND

  • Compression Accelerators such as QAT

  • Data Plane VNF acceleration

  • Security features such as TPM, SGX

  • Hardened Kernel and Hypervisor


Project

Impact

Project

Impact

SDC

Support modeling EPA attributes.

Policy 

Support definition of homing policies based on EPA attributes.

APPC

Support configuration of EPA settings.

  • Inquiring OF to get the best of DC region to bring up the workload.

  • Calling Multi-Cloud "Workload instantiate API' with returned information (including ONAP formatted EPA values)

SO

Support additional HEAT features.

If SO calls MULTICLOUD for VNF/VNFC instantiation, then it needs to get hold of DC region and workload flavor information from OF and pass same to MULTICLOUD as part of instantiation API.

MULTICLOUD

  • Discovery of EPA (For every region of DCs configured)

    • Discovery of workload flavors (e.g flavors in case of Openstack, VM sizes in case of Azure, instnace types in case of AWS) supported by DC region (Periodically by inquiring VIM in the regions)

    • Reading attribute values of each workload flavor (such as memory size, disk size, CPU pinning, Network acceleration etc..)

    • Normalizing the attributes to ONAP EPA attribute format.

    • Storing the normalized attributes in A&AI on corresponding DC and region graph node.

  • During workload instantiation

    • SO/APPC provide ONAP EPA attribute values as part of workload instantiation API of MultiCloud.

    • Right plugin associated with the region is called.

    • Plugin converts EPA information (which is in ONAP format) to VIM specific information before it asks VIM to instantiate the workload.

  • Capacity/Quota information discovery

    • Periodic fetching of current capacity of VIM.

    • Setting up same in the A&AI on per DC region.

A&AI

Support EPA attributes.

  • Provides a way to store on per DC region basis, set of workload flavors and associated EPA attributes & values

OF

  • Selection of the best region and workload-flavor by matching the EPA properties with region specific EPA records.