OVNC 2015-Open Ethernet과 SDN을 통한 Mellanox의 차세대 네트워크 혁신 방안

정연구 Sr.SE 5.Feb 2015
오픈 이더넷과 SDN을 통한
Mellanox의 차세대 네트워크 혁신 방안

© 2014 Mellanox Technologies 2- Mellanox Confidential -
Agenda
 DataCenter Network 트랜드?
 Next Generation Software Defined Networks
 Mellanox Open Ethernet & OCP 소개

데이타센타 Network Trends

The Evolution of SDN
Stanford guys
wanted to change
the networking
world using
OpenFlow
Switches are a
build of closed
software being
sold as a package
from switch
vendors
“SDN” became
a popular term
– everything is
“SDN”
Switch vendors
made OpenFlow
“an important
protocol”
The networking
industry finally
delivers on the
promise of “SDN”
2010 2011 2012 2013 2014

기술 트랜드 : from Closed to Open
Protocol
Stack
OS
System
Silicon
ONIE
Open SDK

기술 트랜드 : Interconnect 네트워크의 중요성 증대
 Traditional role – Connecting users
• 80% of traffic is North-South
 New role – Connecting devices
• 75% of network traffic is now going East-West
 New challenges
• Storage bottlenecks
• Latency and slow I/O
Application Running

기술 트랜드 : from Modular to Scalable
Source: Scale and Efficiency in Data Center Networks, Google / UC San Diego
Virtual Modular SwitchModular Switch
Cost Effective, FlexibleExpensive, Complex

Next Generation Software Defined Networks

EVN: More than SDN - Efficient Clouds Need an Efficient Virtualized Network
차세대 Software Defined Networks
Efficient Virtualized Network (EVN)
Fully Integrated, World Class SDN Solution
EVN: Efficient Virtualized Network
Fully Integrated Solution Combines
RDMA, Convergence, & SDN/Virtualization
CONVERGENCEACCELERATIONVIRTUALIZATION

SDN 네트워크를 가속화 하기위한 세가지 요소
OpenFlowSoftware Defined Networks Overlay Network TunnelsVirtual Network Management
1. Centralized Software Based Control Plane
• Enables network virtualization
2. Overlay Networks – NVGRE/VXLAN/ Geneve
• Isolation, Scalability, Simplicity
• Mellanox accelerates overlay networks to offer bare metal speed
3. Industry Standard API – OpenFlow
• Enables an industry ecosystem and innovation
= + +

최초의 End to End SDN Interconnect solution
– Interconnect Offload Engines
 Challenge: Implementing overlay networks in software dramatically increases overheads and costs
 Solution: Use interconnect offload engines to handle all networking operations up to the VM
 Benefit: Reduce application cost, cloud CAPEX and OPEX
The Foundation of Cloud 2.0
The World’s First NVGRE / VXLAN Offloaded NIC

The Benefits of VM Provision & Fabric Policy in Hardware
Isolation, Performance & Offload, Simpler SDN
10/40GbE or
InfiniBand Ports
 Deliver the best application performance
 Simpler to manage - Real-time NIC provisioning via OpenFlow
 Provide tenant & application security/isolation
Quantum
Agent
최초의 End to End SDN Interconnect solution
- Mellanox 네트워크 가상화 (Neutron) 플러그 인
Compare eSwitch vs OVS
Qperf (TCP) Latency

Server
VM1 VM2 VM3 VM4
ConnectX-3 Pro 오버레이 네트워크를 기속화
Overlay Network Virtualization: Isolation, Simplicity, Scalability
Virtual Domain 3
Virtual Domain 2
Virtual Domain 1
Physical
View
Server
VM5 VM6 VM7 VM8
Mellanox SDN Switches & Routers
Virtual
View
NVGRE/VXLAN Overlay Networks
Virtual Overlay Networks Simplifies
Management and VM Migration
ConnectX-3 Pro
Overlay Accelerators Enable
Bare Metal Performance
OpenFlow
Virtual Network
Management API

Higher Throughput for Less CPU Transport Overhead
ConnectX-3 Pro 10GbE 어댑터를 통한 클라우드 오버레이 가속화 결과
Higher
Is Better
65%
Improvement
Lower
Is Better
79%
Improvement
NVGRE Initial Results

Higher
Is Better
600%
Improvement
Lower
Is Better
360%
Improvement
VXLAN Initial Results
Higher Throughput for Less CPU Transport Overhead
ConnectX-3 Pro 10GbE – VXLAN 오버레이 네트워크 가속화 결과

EVN: More than SDN - Efficient Clouds Need an Efficient Virtualized Network
Next Generation Software Defined Networks
Efficient Virtualized Network (EVN)
Fully Integrated, World Class SDN Solution
EVN: Efficient Virtualized Network
Fully Integrated Solution Combines
RDMA, Convergence, & SDN/Virtualization
CONVERGENCEACCELERATIONVIRTUALIZATION

RDMA: 데이타센타 내 자원 활용율을 높이는 가장 중요한
인터컨넥트 기술
ZERO Copy Remote Data Transfer
Low Latency, High Performance Data Transfers
InfiniBand - 56Gb/s RoCE* – 40Gb/s
Kernel Bypass Protocol Offload
* RDMA over Converged Ethernet
Application ApplicationUSER
KERNEL
HARDWARE
Buffer Buffer

RDMA – 동작 방식
RDMA over InfiniBand or
Ethernet
KERNELHARDWAREUSER
RACK 1
OS
NIC Buffer 1
Application
1
Application
2
OS
Buffer 1
NICBuffer 1
TCP/IP
RACK 2
HCA HCA
Buffer 1Buffer 1
Buffer 1
Buffer 1
Buffer 1

인터커넥트 기술 적용에 따른 CPU 효율 차이
~88% CPU
Utilization
UserSpaceSystemSpace
~53% CPU
Utilization
~47% CPU
Overhead/Idle
~12% CPU
Overhead/Idle
RDMA 미사용시 RDMA 사용시
UserSpaceSystemSpace

iSCSI over RDMA (iSER): 4배 이상의 Throughput
RDMA Accelerates iSCSI Storage
Hypervisor (KVM)
OS
VM
OS
VM
OS
VM
Adapter
Open-iSCSI w iSER
Compute Servers
Switching Fabric
iSCSI/iSER Target (tgt)
Adapter Local Disks
RDMA Cache
Storage Servers
OpenStack (Cinder)
1.3
5.5
0
1
2
3
4
5
6
iSCSI over TCP iSER
GBytes/s
OpenStack Storage Performance *
* iSER patches are available on OpenStack
branch: https://github.com/mellanox/openstack
Built-in OpenStack Components/Management & Cinder/iSER to Accelerate Storage Access

컨버전스: 스토리지 만을 위한 전용 네트워크의 필요성 제거
Storage – prio1
Management – prio2
vMotion – prio3
Networking – prio4
Web 2.0, Public & Private Clouds Converging on Fast RDMA Interconnects
Single Interconnect for Compute, Networking, Storage
RDMA: InfiniBand & Ethernet (RoCE*)
There is no Fibre Channel in the Cloud!
Converged Fabrics
56Gb/s InfiniBand
10/40Gb/s Ethernet
Compute
Networking
Storage
* RoCE: RDMA over Converged Ethernet

Mellanox 기반 클라우드 랙 솔루션 building Block
Industry Standard Servers
CloudX Rack
• Each equipped with Mellanox ConnectX-3
10GbE/40GbE/InfiniBand Adapter
• Running Cloud Stack / Hypervisor
VMware, Hyper-V, KVM, or OpenStack
• Potentially using local disks/SSD for cloud storage or cache
Mellanox High Performance & Density Switches
• Connected via 10GbE/40GbE/InfiniBand
• Delivering worlds’ best cost/performance
High Performance RDMA Attached Storage
• From variety of partners
• And/or software based storage appliances (SDS)
Mellanox
Cables

스위치와 어댑터에 대한 전체적인 OpenStack 통합
Integrated with
Major OpenStack
Distributions
In-Box With Havana and
Ice House
Neturon-ML2
support for
mixed
environment
(VXLAN, PV,
SRIOV)
Ethernet
Neutron :
Hardware
support for
security and
isolation
Accelerating
storage
access by up
to 5X
OpenStack Plugins Create Seamless Integration , Control, & Management

Mellanox Open Ethernet & OCP 소개

Open
Source
OCP
Automation
Silicon and
SDK
Ecosystem
Community
Partnerships
오픈 이더넷 이란?
 Provide better control over network
resources for the various, changing
data center needs by
 Separation of hardware and
software
 Free choice of
• The most adequate platform
• With the most suitable software
• Run it on the most efficient hardware
 Free choice of
• Switch Silicon
• Hardware
• Operating System
• Protocol Stack

하드웨어와 소프트웨어의 분리
 Open SDK + ONIE (Open Network Install Environment)
 A switch is a server with many ports: ASIC, Hardware, OS, Applications
Protocol
Stack
OS
System
Silicon
ONIE
Open SDK API

 Mellanox is first to market
• First OCP-compliant 10GbE adapter
• First OCP-compliant 40GbE adapter
 Shipping in high volume
• Single and dual port options
• Servers sold by multiple ODMs
 Support Microsoft OCP Cloud Server Specification
OCP 어댑터

SX1410 ToR 스위치
 1.92Tb/s throughput
 200/300ns L2/L3 latency
 x86 dual core
 Low power
 Unified management interfaces and FRUs
 Non blocking 10GbE ToR
 ONIE memory compliant
 SwitchX-2 based
 Port Configurations
• 48 SFP+ 10GbE – SR/LR/DAC
• 12 QSFP+ 40GbE – SR4/LR4/DAC
• Break-out capabilities:
40GbE ports 12 10 8 6 4 2 0
10GbE ports 48 54 56 58 60 62 64

Open Platform System Software Blocks
Network OS
Switch SDK Platform
Drivers
Open Ethernet Switch API Platform Interfaces
Provided by system vendor
ONIE Boot Loader
ONIE Any OS

오픈 스위치 SDK API 제공
https://github.com/Mellanox/SwitchX-interfaces/tree/master/SwitchX-2

 Highest Capacity in 1RU
• From 12 QSFP to 36 QSFP
 Unique Value Proposition
• 56GbE
• VPI
• End to end solution
• Lowest power
오픈 이더넷 하드웨어 제품 군
1U, 36 x QSFP Ethernet Switch
The Ideal 40GbE ToR/Aggregation
1U, 48 x SFP+ and 12 x QSFP Ethernet Switch
Non-blocking 10GbE 40GbE ToR
1U, 64 x SFP+ Ethernet Switch
Highest density 10GbE ToR
 Low latency
• 220ns L2 latency
• 330ns L3 latency
 Low power
• Under 1W per 10GbE interface
• 2.3W per 40GbE interface
• 0.6W per 10GbE of throughput
1U, Half Width, 12 x QSFP Ethernet Switch
Ideal storage/Database 10/40GbE Switch

ONIE를 위한 통합과 기여
 Open Network Install Environment
• Boot loader + Linux kernel + BusyBox
 Freedom of choice
• Load any Net-OS on any hardware
 ONIE for SwitchX-2 available on
• PPC
• x86
Source: http://onie.github.io/onie

최적의 데이타 센타를 위한 Mellanox 인터컨넥트
Lowest Latency Highest Throughput
Unlimited Scalability
Best Offload Engines
InfiniBand and Ethernet
Compute and Storage
RDMA
Best Return on Investment
Highest Utilization
Most Cost-Effective Compute and Storage Interconnect Solution
Virtualization
Overlay NetworksCongestion Control

Thank YouThank You

OVNC 2015-Open Ethernet과 SDN을 통한 Mellanox의 차세대 네트워크 혁신 방안

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to OVNC 2015-Open Ethernet과 SDN을 통한 Mellanox의 차세대 네트워크 혁신 방안 (20)

More from NAIM Networks, Inc. (20)

Recently uploaded (20)

OVNC 2015-Open Ethernet과 SDN을 통한 Mellanox의 차세대 네트워크 혁신 방안