SlideShare a Scribd company logo
Data Path acceleration
techniques in a NFV world
©Happiest Minds Technologies Pvt. Ltd. All RightsReserved
Mohanraj Venkatachalam, Purnendu Ghosh
Abstract
NFV is a revolutionary approach offering greater flexibility and scalability in the deployment of
virtual network functions. It provides significant advantages over the current vendor-locking
infrastructure using physical network functions. The NFV architecturehas three buildingblocks
a) Infrastructure b)VirtualNetwork Functions c)Orchestrator.
Telcoshave stringentSLA towards latency,throughput, packets-per-sec,
bits-per-seconds to meet both QOE, QOS and guaranteed bandwidth per user subscription.
Therefore, an NFV system has to match the data path capabilitiesat parwith actual Physical
Network Functions. Alot ofresearch is ongoinginthe development ofdata path acceleration
techniques for a virtualized infrastructure by using hypervisors (e.g.:Xen, KVM, VMware,
Hyper-V) inthe following possible ways: (a) Device Emulation(b)
Para-Virtualization (c) PCIPass-through.
Let us walkthroughwith some of the standard HW andSW techniques availableto achievea
comparable performance of VNF (Virtual Network Function) runningin a virtualized environ-
ment as compared to PNF (Physical Network Function) used in legacy network architecture.
The criticalaspect inall ofthese approachesis to:
•Increase packetthroughput fromphysical NICto virtual NICs
•Reduce the number of copiesdonebefore it is processed bythe application
residing in theVM
•Reduce the numberof CPU interruptsbefore a packetis delivered to a VM
©Happiest Minds TechnologiesPvt. Ltd. All Rights Reserved
VMDq (Virtual Machine DeviceQueues)
These are specializednetwork controllers capableof sorting data packets into multiple queues.
Adistinct queue->vNIC->CPU_Core mappingensures independent parallel I/Ostreams for
each VMworkload.
This solution offloads the overhead of data packet sorting from hypervisor switch to network
siliconand improves the network I/Othroughput over 100 as compared with a traditional
solution. AVMDq solution consists (this is Intel specific) improvements to Hypervisor Switch
(e.g.:VMware ESX). This arrangement ensures that Rxinterruptsare directlyaddressed to CPU
coreofthe VM thereby reducing the number ofcopiesrequired
However, with VMDq there is a performance penalty due to the number of interrupts for VM
directed data-planetraffic. AnSR-IOV based solution brings considerable improvement inthis
aspect as described in thenext section.
SR-IOV
SR-IOV or Single-Root IO Virtualization is a specificationwhich allows a single PCIe device
configured as multiple virtual devicesandthey appearas independent network functions to
the VMs runningin a hypervisor. These devices consist of VFs (named as Virtual Functions)
acting as light-weightPCIe unitsthat are capableofbeingconfigured for data traffic move-
ment. There are also PFs (Physical Functions) which contain full PCIe functions and can be
used to configure SR-IOV enableddevices.
Figure 1:Single Root I/OVirtualizationand Sharing(SR-IOV)
Seen as PCI Express *
Device Ethernet Controllerin
OS “Virtual Function Driver”
Seen as PCI Express *
Device Ethernet Controllerin
OS“ Physical FunctionDriver”
Hardware resources
areavailableforSR-IOV
defined VirtualFunctions
utilizing a HardwareVEB
VMVM
Operati
ngSyste
m
VirtualSwitch
VM
Intel’VT-d
PCIExpress
PFDriver
PF Driver
Name Name
Space Space
Hardware Virtual Ethernet Bridge(VEB)
Intel”Ethernet
withSR-IOV
unctionF
Virtual
1 unctionF
Virtual
2 unctionF
Virtual
n
Physical
Function
LAN
Traffic
©Happiest Minds TechnologiesPvt. Ltd. All Rights Reserved
To read the complete success story visit
https://www.happiestminds.com/whitepa
pers/Data-Path-acceleration-techniques-
in-a-NFV-world.pdf
For any Queries, Write to us at:
business@happiestminds.com

More Related Content

PDF
HPC Cloud: Clouds on supercomputers for HPC
PDF
XPDDS17: How to Abstract Hardware Acceleration Device in Cloud Environment - ...
PDF
From Rack scale computers to Warehouse scale computers
PDF
Решения WANDL и NorthStar для операторов
PDF
POWER9 for AI & HPC
PDF
NFV в сетях операторов связи
PDF
Решения NFV в контексте операторов связи
PDF
Nap extras
HPC Cloud: Clouds on supercomputers for HPC
XPDDS17: How to Abstract Hardware Acceleration Device in Cloud Environment - ...
From Rack scale computers to Warehouse scale computers
Решения WANDL и NorthStar для операторов
POWER9 for AI & HPC
NFV в сетях операторов связи
Решения NFV в контексте операторов связи
Nap extras

What's hot (20)

PPTX
Sharing High-Performance Interconnects Across Multiple Virtual Machines
PDF
Inside the Volta GPU Architecture and CUDA 9
PDF
Exploring the Performance Impact of Virtualization on an HPC Cloud
PDF
Different approaches to performance enhancements in network virtualization fo...
PDF
POWER10 innovations for HPC
PPSX
From virtual to high end HW routing for the adult
PDF
OSインストーラーの自作方法
DOCX
Moving CCAP To The Cloud
PDF
Развитие MX маршрутизаторов
PDF
Unifying Network Filtering Rules for the Linux Kernel with eBPF
PDF
A Framework with Cloud Integration for CNN Acceleration on FPGA Devices
PDF
Quantum Computing in China: Progress on Superconducting Multi-Qubits System
PDF
PLNOG 13: Piotr Szolkowski: 100G Ethernet – Case Study
PDF
@IBM Power roadmap 8
PPT
PPTX
Re-Think Storage – PernixData. Meet & greet with Frank Denneman
PDF
CONDOR: An automated framework to accelerate convolutional neural networks on...
PDF
RISC-V and OpenPOWER open-ISA and open-HW - a swiss army knife for HPC
PDF
Introduction of Fujitsu's HPC Processor for the Post-K Computer
PPT
Title Subtitle
Sharing High-Performance Interconnects Across Multiple Virtual Machines
Inside the Volta GPU Architecture and CUDA 9
Exploring the Performance Impact of Virtualization on an HPC Cloud
Different approaches to performance enhancements in network virtualization fo...
POWER10 innovations for HPC
From virtual to high end HW routing for the adult
OSインストーラーの自作方法
Moving CCAP To The Cloud
Развитие MX маршрутизаторов
Unifying Network Filtering Rules for the Linux Kernel with eBPF
A Framework with Cloud Integration for CNN Acceleration on FPGA Devices
Quantum Computing in China: Progress on Superconducting Multi-Qubits System
PLNOG 13: Piotr Szolkowski: 100G Ethernet – Case Study
@IBM Power roadmap 8
Re-Think Storage – PernixData. Meet & greet with Frank Denneman
CONDOR: An automated framework to accelerate convolutional neural networks on...
RISC-V and OpenPOWER open-ISA and open-HW - a swiss army knife for HPC
Introduction of Fujitsu's HPC Processor for the Post-K Computer
Title Subtitle
Ad

Similar to Data path-acceleration-techniques-in-a-nfv-world (20)

PDF
Summit 16: Optimizing OPNFV for Distributed NFV Applications
PDF
XPDS14 - Xen as High-Performance NFV Platform - Jun Nakajima, Intel
PDF
Introduction to nfv movilforum
PDF
Summit 16: ARM Mini-Summit - NXP QorIQ NFV Solutions - NXP Semiconductors
PPTX
Netsft2017 day in_life_of_nfv
PDF
Meaningful and Necessary Operations on Behalf of NFV
PDF
High performance and flexible networking
PDF
NFV & Openstack
PDF
Understanding NFV and the MANO stack
PDF
NFV Tutorial
PDF
NFV Tutorial
PPTX
Nfv compute domain
PDF
Evolving Virtual Networking with IO Visor [OpenStack Summit Austin | April 2016]
PDF
Evolving Virtual Networking with IO Visor
PPTX
NFV foundation/NFV For Dummies
PDF
Platforms for Accelerating the Software Defined and Virtual Infrastructure
PDF
Presentación Laboratorio NFV de Telefónica de Antonio Elizondo
PPTX
Hyper-V Networking
PPTX
Modern Networking Unit 3 Network Function virtualization
PPTX
Supporting Virtualized Telco Applications with OpenStack
Summit 16: Optimizing OPNFV for Distributed NFV Applications
XPDS14 - Xen as High-Performance NFV Platform - Jun Nakajima, Intel
Introduction to nfv movilforum
Summit 16: ARM Mini-Summit - NXP QorIQ NFV Solutions - NXP Semiconductors
Netsft2017 day in_life_of_nfv
Meaningful and Necessary Operations on Behalf of NFV
High performance and flexible networking
NFV & Openstack
Understanding NFV and the MANO stack
NFV Tutorial
NFV Tutorial
Nfv compute domain
Evolving Virtual Networking with IO Visor [OpenStack Summit Austin | April 2016]
Evolving Virtual Networking with IO Visor
NFV foundation/NFV For Dummies
Platforms for Accelerating the Software Defined and Virtual Infrastructure
Presentación Laboratorio NFV de Telefónica de Antonio Elizondo
Hyper-V Networking
Modern Networking Unit 3 Network Function virtualization
Supporting Virtualized Telco Applications with OpenStack
Ad

More from Happiest Minds Technologies (20)

PDF
Happiest MInds - Pimcore PIM Expertise.pdf
PDF
A Quick Guide to Product Information Management (PIM)
PDF
Largest Electricity provider in the US- Case Study
PDF
BFSI GLOBAL TRENDS FY 24
PDF
ARTIFICIAL INTELLIGENCE IN DIGITAL BANKING
PDF
DIGITAL MANUFACTURING
PDF
Exploring the Potential of ChatGPT in Banking, Financial SERVICES & Insurance
PDF
AN OVERVIEW OF THE METAVERSE
PDF
VMware to AWS Cloud Migration
PDF
Digital-Content-Monetization-DCM-Platform-2.pdf
PDF
AUTOMATING CYBER RISK DETECTION AND PROTECTION WITH SOC 2.0
PDF
Cloud Reshaping Banking
PDF
Automating SOC1/2 Compliance- For a leading Software solution company in UK
PDF
PAMaaS- Powered by CyberArk
PDF
GUIDE TO KEEP YOUR END-USERS CONNECTED TO THE DIGITAL WORKPLACE DURING DISRUP...
PDF
SECURING THE CLOUD DATA LAKES
PDF
Complete Guide to General Data Protection Regulation (GDPR)
PDF
Azure bastion- Remote desktop RDP/SSH in Azure using Bastion Service as (PaaS)
PDF
REDUCING TRANSPORTATION COSTS IN RETAIL THROUGH INTELLIGENT FREIGHT AUDIT
PDF
REDUCING TRANSPORTATION COSTS IN CPG THROUGH INTELLIGENT FREIGHT AUDIT
Happiest MInds - Pimcore PIM Expertise.pdf
A Quick Guide to Product Information Management (PIM)
Largest Electricity provider in the US- Case Study
BFSI GLOBAL TRENDS FY 24
ARTIFICIAL INTELLIGENCE IN DIGITAL BANKING
DIGITAL MANUFACTURING
Exploring the Potential of ChatGPT in Banking, Financial SERVICES & Insurance
AN OVERVIEW OF THE METAVERSE
VMware to AWS Cloud Migration
Digital-Content-Monetization-DCM-Platform-2.pdf
AUTOMATING CYBER RISK DETECTION AND PROTECTION WITH SOC 2.0
Cloud Reshaping Banking
Automating SOC1/2 Compliance- For a leading Software solution company in UK
PAMaaS- Powered by CyberArk
GUIDE TO KEEP YOUR END-USERS CONNECTED TO THE DIGITAL WORKPLACE DURING DISRUP...
SECURING THE CLOUD DATA LAKES
Complete Guide to General Data Protection Regulation (GDPR)
Azure bastion- Remote desktop RDP/SSH in Azure using Bastion Service as (PaaS)
REDUCING TRANSPORTATION COSTS IN RETAIL THROUGH INTELLIGENT FREIGHT AUDIT
REDUCING TRANSPORTATION COSTS IN CPG THROUGH INTELLIGENT FREIGHT AUDIT

Recently uploaded (20)

PDF
A contest of sentiment analysis: k-nearest neighbor versus neural network
PDF
WOOl fibre morphology and structure.pdf for textiles
PDF
Getting started with AI Agents and Multi-Agent Systems
PPT
Module 1.ppt Iot fundamentals and Architecture
PDF
DP Operators-handbook-extract for the Mautical Institute
PDF
Video forgery: An extensive analysis of inter-and intra-frame manipulation al...
PPTX
cloud_computing_Infrastucture_as_cloud_p
PDF
2021 HotChips TSMC Packaging Technologies for Chiplets and 3D_0819 publish_pu...
PDF
1 - Historical Antecedents, Social Consideration.pdf
PPTX
Tartificialntelligence_presentation.pptx
PDF
NewMind AI Weekly Chronicles - August'25-Week II
PDF
Enhancing emotion recognition model for a student engagement use case through...
PPTX
Final SEM Unit 1 for mit wpu at pune .pptx
PDF
Hybrid model detection and classification of lung cancer
PDF
Univ-Connecticut-ChatGPT-Presentaion.pdf
PDF
DASA ADMISSION 2024_FirstRound_FirstRank_LastRank.pdf
PDF
Zenith AI: Advanced Artificial Intelligence
PDF
Transform Your ITIL® 4 & ITSM Strategy with AI in 2025.pdf
PPTX
TechTalks-8-2019-Service-Management-ITIL-Refresh-ITIL-4-Framework-Supports-Ou...
PDF
Getting Started with Data Integration: FME Form 101
A contest of sentiment analysis: k-nearest neighbor versus neural network
WOOl fibre morphology and structure.pdf for textiles
Getting started with AI Agents and Multi-Agent Systems
Module 1.ppt Iot fundamentals and Architecture
DP Operators-handbook-extract for the Mautical Institute
Video forgery: An extensive analysis of inter-and intra-frame manipulation al...
cloud_computing_Infrastucture_as_cloud_p
2021 HotChips TSMC Packaging Technologies for Chiplets and 3D_0819 publish_pu...
1 - Historical Antecedents, Social Consideration.pdf
Tartificialntelligence_presentation.pptx
NewMind AI Weekly Chronicles - August'25-Week II
Enhancing emotion recognition model for a student engagement use case through...
Final SEM Unit 1 for mit wpu at pune .pptx
Hybrid model detection and classification of lung cancer
Univ-Connecticut-ChatGPT-Presentaion.pdf
DASA ADMISSION 2024_FirstRound_FirstRank_LastRank.pdf
Zenith AI: Advanced Artificial Intelligence
Transform Your ITIL® 4 & ITSM Strategy with AI in 2025.pdf
TechTalks-8-2019-Service-Management-ITIL-Refresh-ITIL-4-Framework-Supports-Ou...
Getting Started with Data Integration: FME Form 101

Data path-acceleration-techniques-in-a-nfv-world

  • 1. Data Path acceleration techniques in a NFV world ©Happiest Minds Technologies Pvt. Ltd. All RightsReserved Mohanraj Venkatachalam, Purnendu Ghosh
  • 2. Abstract NFV is a revolutionary approach offering greater flexibility and scalability in the deployment of virtual network functions. It provides significant advantages over the current vendor-locking infrastructure using physical network functions. The NFV architecturehas three buildingblocks a) Infrastructure b)VirtualNetwork Functions c)Orchestrator. Telcoshave stringentSLA towards latency,throughput, packets-per-sec, bits-per-seconds to meet both QOE, QOS and guaranteed bandwidth per user subscription. Therefore, an NFV system has to match the data path capabilitiesat parwith actual Physical Network Functions. Alot ofresearch is ongoinginthe development ofdata path acceleration techniques for a virtualized infrastructure by using hypervisors (e.g.:Xen, KVM, VMware, Hyper-V) inthe following possible ways: (a) Device Emulation(b) Para-Virtualization (c) PCIPass-through. Let us walkthroughwith some of the standard HW andSW techniques availableto achievea comparable performance of VNF (Virtual Network Function) runningin a virtualized environ- ment as compared to PNF (Physical Network Function) used in legacy network architecture. The criticalaspect inall ofthese approachesis to: •Increase packetthroughput fromphysical NICto virtual NICs •Reduce the number of copiesdonebefore it is processed bythe application residing in theVM •Reduce the numberof CPU interruptsbefore a packetis delivered to a VM ©Happiest Minds TechnologiesPvt. Ltd. All Rights Reserved
  • 3. VMDq (Virtual Machine DeviceQueues) These are specializednetwork controllers capableof sorting data packets into multiple queues. Adistinct queue->vNIC->CPU_Core mappingensures independent parallel I/Ostreams for each VMworkload. This solution offloads the overhead of data packet sorting from hypervisor switch to network siliconand improves the network I/Othroughput over 100 as compared with a traditional solution. AVMDq solution consists (this is Intel specific) improvements to Hypervisor Switch (e.g.:VMware ESX). This arrangement ensures that Rxinterruptsare directlyaddressed to CPU coreofthe VM thereby reducing the number ofcopiesrequired However, with VMDq there is a performance penalty due to the number of interrupts for VM directed data-planetraffic. AnSR-IOV based solution brings considerable improvement inthis aspect as described in thenext section. SR-IOV SR-IOV or Single-Root IO Virtualization is a specificationwhich allows a single PCIe device configured as multiple virtual devicesandthey appearas independent network functions to the VMs runningin a hypervisor. These devices consist of VFs (named as Virtual Functions) acting as light-weightPCIe unitsthat are capableofbeingconfigured for data traffic move- ment. There are also PFs (Physical Functions) which contain full PCIe functions and can be used to configure SR-IOV enableddevices. Figure 1:Single Root I/OVirtualizationand Sharing(SR-IOV) Seen as PCI Express * Device Ethernet Controllerin OS “Virtual Function Driver” Seen as PCI Express * Device Ethernet Controllerin OS“ Physical FunctionDriver” Hardware resources areavailableforSR-IOV defined VirtualFunctions utilizing a HardwareVEB VMVM Operati ngSyste m VirtualSwitch VM Intel’VT-d PCIExpress PFDriver PF Driver Name Name Space Space Hardware Virtual Ethernet Bridge(VEB) Intel”Ethernet withSR-IOV unctionF Virtual 1 unctionF Virtual 2 unctionF Virtual n Physical Function LAN Traffic ©Happiest Minds TechnologiesPvt. Ltd. All Rights Reserved
  • 4. To read the complete success story visit https://www.happiestminds.com/whitepa pers/Data-Path-acceleration-techniques- in-a-NFV-world.pdf For any Queries, Write to us at: business@happiestminds.com

Editor's Notes

  • #5: https://www.happiestminds.com/casestudies/Self-Knowledge-Assembly-Platform-Transforms-Customer-Service.pdf