Acceleration Management (BoF)

1.Research Motivation

1.1  What is acceleration?

NFV Acceleration deals with both hardware and software accelerations. It is provided by NFVI to meet some specific performance requirement of VNFs, which can be done by hardware, software or any combination thereof. At present, the core of accelerating technology is to find the appropriate design way of AAL( acceleration abstraction layer).

1.2  Why do we need acceleration in NFV?

  We can consider this problem in three aspects:

1)     We can consider this problem in three aspects;

2)     We concern about the best performance per processor core/cost/watt/square foot, whatever the absolute performance metric is;

3)     We want to reach maximum theoretical performance level.

1.3  What is acceleration management?

  Because the type of NFV acceleration deployment varies from an integrated deployment to a disaggregated deployment.

  In order to accommodate the possibility of a disaggregated deployment, we use acceleration management to ensure higher-level orchestration capabilities,utilization monitoring, and customer SLAs are supported.

1.4  Acceleration use case

1)     Compute Acceleration?IFA 001: Page 14-24?

Page

Title

NFV Components

MANO

Accelerator

14-15

IPSec tunnels termination VNFC

VNFC,VNFD,VIM,Orchestrator

Nf-Vi -Instantiation and VNFD,VIM

look-aside, in-line, fast path, software

15-16

Next Generation Fire Wall (NGFW) Acceleration

NFVI, Software Architecture, Performance and Security

Local or remote API

NFVI Fast Path, In-Line, and Look-Aside Accelerators

16-19

Virtual Base Station (VBS) L1 Acceleration

VNF/VNFC,VIM,Compute Nodes,Networking Nodes,HWA Nodes

VBS VNF ,Operating configurations,Orchestrator and VIM ,VNF Manager and or VNF EMS

Two options:1.In-Line;2. Look Aside accelerator

19-21

Virtual Acceleration Interface for VNFs

NFVI,MANO,Software Architecture, Performance and Security

Match VNF's accelerator's requirements with NFVI's accelerator capabilities

Crypto -Public key and Symmetric Key, IPsec Protocol Accelerator, etc.

22-23

Transcoding

VNF/VNFC,VIM,NFVO,VNFM

Requirements of VNFM and NFVO,VNFD,VIM

Look-aside,In-line,Fast Path or Optimized SW Path

24

Deep Packet Inspection

VNF/VNFC,VIM,NFVO,VNFM

Some functions listed in the table

Look-Aside, Fast-Path, etc.



2)     Network Acceleration?IFA 001: Page 24-34?

Page

Title

NFV Components

MANO

Accelerator

25

Load Balancing and NAT

Compute Node, including an intelligent NIC with offload capability

Managing the vSwitch through a local API is required. This acceleration is transparent to the VNFs.

NFVI Fast Path, In-Line, and Look-Aside Accelerators

26-29

NFVI Virtual Networking Offload

VNF, VIM, Compute Node, Network Node, Storage Node, HWA Node

Items of MANO related aspects

Fast Path; based on OF based packet processor

30-33

NFVI Secure Overlay Offload

VNF, VIM, Compute Node, Network Node, Storage Node, HWA Node

Items of MANO related aspects

Fast Path; based on OF based packet processor

33-34

Dynamic Optimization of Packet Flow Routing

VNF, VIM, Infrastructure Network including Network Controller

The VIM requests the Network Controller to provide a logical switch dedicated to the VNF

Physical switches, virtual switches and other accelerator devices



3)     Storage Acceleration?IFA 001: Page 35-36?

Page

Title

NFV Components

MANO

Accelerator

35

NVMe™ Over Fabric Enabled Acceleration

VNF/VNFC/VNFD ,VIM,Compute Node,Storage Node/Virtual Storage

VNFD,VIM

In-Line Accelerator

36

High Performance Persistent Memory on Compute Node

Compute Node ,VNF Manager ,VIM ,Orchestrator

VIM and Orchestrator need to know about available capacity, location and mode of access of persistent memory. VIM has to maintain usage count.

Optimized Store



1.5  Acceleration Management

In order to accommodate the possibility of a disaggregated deployment, we use acceleration management to ensure higher-level orchestration capabilities,utilization monitoring, and customer SLAs are supported.

2. Problem Statement

2.1  The description and issuance of acceleration requirement

1)     Two kinds of technical route:

Technical Route

Pros

Cons

Maturity level

Specify the model and release requirements for specific hardware/software,such as EPA/HPA ,etc.

Simple, easy to operate

It binds specific hardware/software to form chimneys, and is not conducive to platform generalization and resource pooling in consequence

It has been applied in practice

Specify acceleration management function as well as performance index requirements, such as items described in IFA004.

Evolve the decoupling ability of appliance and platform independently

It relies on virtualised acceleration technology, and requires a third party or additional mechanisms to establish correlation matching knowledge of requirements description and platform characteristics

It has not been actually applied



2)     A hybrid version of two technical routes is currently included in IFA011

3)     Related ONAP module include?VNF Modeling?VNF SDK?SDC?Mcloud

2.2  The discovery and monitoring of resource capability

1)     Two kinds of solutions:

Solution Name

Pros

Cons

Openstack Nova

It is applicable to instruction set expansion, DPDK and other computing nodes' internal hardware and software platform acceleration managementability, there are mature application cases

It is suitable for all types of acceleration capability discovery and state monitoring;

It can be combined with Nova or complete the discovery and monitoring of computing nodes' internal hardware and software platform acceleration management ability independently.

External resource scheduling–OpenstackCyborg

It cannot be applied to external acceleration capability management, and can not be applied to programmable acceleration capability management

Newcomer of OpenStack main version, the features /performance/stability need validation;

Supported accelerationchip types are limited and end-to-end application case validation is not implemented.

 

2)     Related ONAP module include?Mcloud?A&AI?DCAE

2.3  Matching & scheduling of demand and capability

1)     Concrete description way

Accurate matching? Optimal matching?

2)     Abstract description way

Establishment, maintenance and application of knowledge database

3)     Related ONAP module include: OOF?Policy

3. Proposals

3.1  Form an interest group

  • Anybody interested in joining the thread? Please contact Lei Huang, email address: 18350830036@163.com

  • We can form an interest group under usecase/architecture teamsidentify a target usecaseand develop a joint plan for C+ release.

3.2  Determine service scenarios

3.3  Confirm version plan