Low Latency Server Virtualization

Similar documents
OCP Engineering Workshop - Telco

CPU Agnostic Motherboard design with RapidIO Interconnect in Data Center

RapidIO.org Update. Mar RapidIO.org 1

RapidIO.org Update.

NVM PCIe Networked Flash Storage

Sub-microsecond interconnects for processor connectivity The opportunity

RoCE vs. iwarp Competitive Analysis

Mellanox Technologies Maximize Cluster Performance and Productivity. Gilad Shainer, October, 2007

ARISTA: Improving Application Performance While Reducing Complexity

SwitchX Virtual Protocol Interconnect (VPI) Switch Architecture

STORAGE CONSOLIDATION WITH IP STORAGE. David Dale, NetApp

NVMe over Universal RDMA Fabrics

Cisco Nexus 4000 Series Switches for IBM BladeCenter

The NE010 iwarp Adapter

Mellanox Technologies Delivers The First InfiniBand Server Blade Reference Design

EXTENDING AN ASYNCHRONOUS MESSAGING LIBRARY USING AN RDMA-ENABLED INTERCONNECT. Konstantinos Alexopoulos ECE NTUA CSLab

MELLANOX EDR UPDATE & GPUDIRECT MELLANOX SR. SE 정연구

STORAGE CONSOLIDATION WITH IP STORAGE. David Dale, NetApp

VIRTUALIZING SERVER CONNECTIVITY IN THE CLOUD

Maailman paras palvelinjärjestelmä. Tommi Salli Distinguished Engineer

Messaging Overview. Introduction. Gen-Z Messaging

IDT Bridging the PCI Express Interconnect Divide Application Note

All Roads Lead to Convergence

Implementing RapidIO. Travis Scheckel and Sandeep Kumar. Communications Infrastructure Group, Texas Instruments

Storage Protocol Offload for Virtualized Environments Session 301-F

PCI Express x8 Single Port SFP+ 10 Gigabit Server Adapter (Intel 82599ES Based) Single-Port 10 Gigabit SFP+ Ethernet Server Adapters Provide Ultimate

ALCATEL-LUCENT ENTERPRISE DATA CENTER SWITCHING SOLUTION Automation for the next-generation data center

In-chip and Inter-chip Interconnections and data transportations for Future MPAR Digital Receiving System

FRONT-HAUL COMPRESSION FOR EMERGING C- RAN AND SMALL CELL NETWORKS

Introduction to Ethernet Latency

Interconnect Your Future

IBM WebSphere MQ Low Latency Messaging Software Tested With Arista 10 Gigabit Ethernet Switch and Mellanox ConnectX

Voltaire. Fast I/O for XEN using RDMA Technologies. The Grid Interconnect Company. April 2005 Yaron Haviv, Voltaire, CTO

Data Path acceleration techniques in a NFV world

Communication has significant impact on application performance. Interconnection networks therefore have a vital role in cluster systems.

srio Hot Swap Application Note - 552

Industry Standards for the Exponential Growth of Data Center Bandwidth and Management. Craig W. Carlson

Cisco UCS Virtual Interface Card 1225

Ethernet: The High Bandwidth Low-Latency Data Center Switching Fabric

Performance Optimizations via Connect-IB and Dynamically Connected Transport Service for Maximum Performance on LS-DYNA

Survey of ETSI NFV standardization documents BY ABHISHEK GUPTA FRIDAY GROUP MEETING FEBRUARY 26, 2016

Key Measures of InfiniBand Performance in the Data Center. Driving Metrics for End User Benefits

NETWORK ARCHITECTURES AND CONVERGED CLOUD COMPUTING. Wim van Laarhoven September 2010

Dragon Slayer Consulting

Enyx soft-hardware design services and development framework for FPGA & SoC

OpenOnload. Dave Parry VP of Engineering Steve Pope CTO Dave Riddoch Chief Software Architect

Interconnect Your Future

Optimize New Intel Xeon E based Ser vers with Emulex OneConnect and OneCommand Manager

2017 Storage Developer Conference. Mellanox Technologies. All Rights Reserved.

How to Network Flash Storage Efficiently at Hyperscale. Flash Memory Summit 2017 Santa Clara, CA 1

SR-IOV Support for Virtualization on InfiniBand Clusters: Early Experience

Interconnect Your Future

Deploy a Next-Generation Messaging Platform with Microsoft Exchange Server 2010 on Cisco Unified Computing System Powered by Intel Xeon Processors

BlueGene/L. Computer Science, University of Warwick. Source: IBM

Configuring SR-IOV. Table of contents. with HP Virtual Connect and Microsoft Hyper-V. Technical white paper

Data Center & Cloud Computing DATASHEET. PCI Express x8 Single Port SFP+ 10 Gigabit Server Adapter. Flexibility and Scalability in Data Centers

Bridging from PCI Express to PCI in Small-Form Factor Embedded Computing Designs

TDT Appendix E Interconnection Networks

Chapter 6 Connecting Device

Jeff Dodson / Avago Technologies

Cisco UCS Virtual Interface Card 1227

Accelerating Real-Time Big Data. Breaking the limitations of captive NVMe storage

Learn Your Alphabet - SRIOV, NPIV, RoCE, iwarp to Pump Up Virtual Infrastructure Performance

Media-Ready Network Transcript

iscsi Technology: A Convergence of Networking and Storage

iscsi or iser? Asgeir Eiriksson CTO Chelsio Communications Inc

PERFORMANCE ACCELERATED Mellanox InfiniBand Adapters Provide Advanced Levels of Data Center IT Performance, Productivity and Efficiency

IDT for FPGAs CLOCKS AND TIMING INTERFACE AND CONNECTIVITY MEMORY AND LOGIC POWER MANAGEMENT RF PRODUCTS

Nexus DC Tec. Tomas Novak. BDM Sponsor. Sponsor. Sponsor Logo. Sponsor. Logo. Logo. Logo

Fault Tolerant and Secure Architectures for On Chip Networks With Emerging Interconnect Technologies. Mohsin Y Ahmed Conlan Wesson

Interconnect Your Future Enabling the Best Datacenter Return on Investment. TOP500 Supercomputers, November 2017

Network Adapters. FS Network adapter are designed for data center, and provides flexible and scalable I/O solutions. 10G/25G/40G Ethernet Adapters

Rack Disaggregation Using PCIe Networking

What a Long Strange Trip It s Been: Moving RDMA into Broad Data Center Deployments

24th Annual ROTH Conference

I/O Virtualization The Next Virtualization Frontier

Paving the Road to Exascale

ECE/CS 757: Advanced Computer Architecture II Interconnects

Company. Intellectual Property. Headquartered in the Silicon Valley

FC-NVMe. NVMe over Fabrics. Fibre Channel the most trusted fabric can transport NVMe natively. White Paper

OCP3. 0. ConnectX Ethernet Adapter Cards for OCP Spec 3.0

VPI / InfiniBand. Performance Accelerated Mellanox InfiniBand Adapters Provide Advanced Data Center Performance, Efficiency and Scalability

Performance Considerations of Network Functions Virtualization using Containers

Cisco Unified Computing System Delivering on Cisco's Unified Computing Vision

Fast packet processing in the cloud. Dániel Géhberger Ericsson Research

ก ก Information Technology II

High Performance Computing

EXPERIENCES WITH NVME OVER FABRICS

Cisco HyperFlex HX220c M4 and HX220c M4 All Flash Nodes

Creating an agile infrastructure with Virtualized I/O

Ethernet. High-Performance Ethernet Adapter Cards

The Myricom ARC Series of Network Adapters with DBL

Cisco HyperFlex HX220c M4 and HX220c M4 All Flash Nodes

SmartNICs: Giving Rise To Smarter Offload at The Edge and In The Data Center

Cisco HyperFlex HX220c M4 Node

Paul Hodge 25 Oct 2016 RUN FAST. NEVER STOP. The Next Generation Virtual Platform

Welcome to the IBTA Fall Webinar Series

Maximizing heterogeneous system performance with ARM interconnect and CCIX

Sun Dual Port 10GbE SFP+ PCIe 2.0 Networking Cards with Intel GbE Controller

JANUARY 28, 2014, SAN JOSE, CA. Microsoft Lead Partner Architect OS Vendors: What NVM Means to Them

Transcription:

Low Latency Server Virtualization Using RapidIO November 29-30, 2011 Server Design Summit Mohammad Akhter Devashish Paul Integrated Device Technology 2010 Integrated Device Technology, Inc.

The Analog and Digital Company The World Leader in Timing, Serial Switching and Interfaces in Silicon Timing Analog in Memory Interfaces System in DisplayPort Expertise in RapidIO Switching in PCIe Solutions in Business PC Audio in VME Bus Solutions in PowerPC Host Analog Bridges Application-Optimized Mixed-Signal Solutions Maximum system performance Lowest overall Bill-of-Materials Shortest time-to-market Communications Computing Consumer Re-thinking the System to Architect a better solution PAGE 2

RapidIO Success RapidIO Enables Disruptive Server Architecture Server IO Virtualization IO Virtualization Flows Servers with RapidIO IO Virtualization using RapidIO Summary Agenda PAGE 3

Lowest Latency Interconnect ~ 100 ns Shipped millions of devices Scalable System RapidIO Success Inherently scales to large system with thousands of nodes Scale s, blades, Aggregate Chassis to large systems Addresses Critical Interconnect Problems in multiple markets Military, Video, Computing, Wireless Industry Large Eco-System Silicon Partners and Suppliers PAGE 4

RapidIO Enables Disruptive Server Architectures Supports both x86 and ARM based Architectures Eliminates need for large ToR Switches Large Servers x86 CPU with RapidIO Micro-servers using ARM CPU with power efficient low cost RapidIO Interconnect Supports Thousands of s and flows Best-in-class end-to-end latency Switch latency around 100 ns Supports Secured Virtualization Supports any kind of topology High Performance Low Latency RapidIO based Server Start, Mesh, Dual-star, Hypercube, Torus etc.. RapidIO Switch Hub RapidIO Switch RapidIO CPS-1848 48 lane switch PCIe2 S-RIO 2 PCIe2 S-RIO 2 PCIe2 S-RIO 2 PCIe2 S-RIO 2 x86 x86 x86 x86 CPU CPU CPU CPU RapidIO Based RapidIO CPS-1848 48 lane switch PAGE 5

RapidIO in Server Lowest latency 100 ns Scales to 1000 s of nodes 8W per switch < 2 W per NIC 13x13 mm PCIe to S-RIO NIC RapidIO Switch Hub RapidIO CPS-1848 48 lane switch RapidIO Switch PCIe2 S-RIO 2 PCIe2 S-RIO 2 PCIe2 S-RIO 2 PCIe2 S-RIO 2 x86 CPU x86 CPU x86 CPU RapidIO Based x86 CPU RapidIO CPS-1848 48 lane switch 13x13 mm PCIe to S-RIO NIC For < 2W PAGE 6

Server IO Virtualization What is Server IO Virtualization Logical sharing of physical IO resources between server and data center resources Why care? On-demand efficient allocation of IO Match CPU and OS processing requirements to IO Allow better utilization of CPU Allow many nodes in the system to access IO resources PAGE 7

Server IO Virtualization Applications OS Kernel/Driver Driver/OS ory ory Memo CPU 0 Hardware I/O Device PAGE 8

Server IO Virtualization Applications (Kernel/Driver) Driver/OS Mem Mem mory CPU 0 Hardware I/O Device PAGE 9

IO Virtualization - Flows Memo mory (Kernel/Driver) (Kernel/Driver) CPU 0 Memo Memo ory CPU 0 I/O Device I/O Device SAN LAN PAGE 10

IO Virtualization - Bypass Memory (Kernel/Driver) CPU 0 I/O Device Communication through is expensive Higher CPU overhead Higher latency Early days IO virtualization was done entirely by software Virtualized enabled IO Card bypasses Data path goes directly from OS to IO Card IO device creates virtualized view of distinct resources and address space to OS Control/Management functions may still flow through PAGE 11

IO Virtualization using RDMA (Kernel/Driver) (Kernel/Driver) Memory Memory CPU 0 RDMAP DDP MPA TCP Memory CPU 0 I/O Device I/O Device Ethernet Switch Today s I/O NIC have too many steps for RDMA! Higher latency and Power Higher cost and CPU overhead PAGE 12

IO Virtualization using RapidIO (Kernel/Driver) (Kernel/Driver) Memory Memory CPU 0 Memory CPU 0 Tsi721 I/O Device Tsi721 I/O Device S-RIO Switch PCIe to S-RIO bridge and Switch Simplies IO Virtualization Improves RDMA without overhead built in DMA/Messaging Reduces latency Supports large number of nodes Improves IO and CPU utilization PAGE 13

RapidIO: Low power Servers (Kernel/Driver) (Kernel/Driver) Memory Memory CPU 0 Memory CPU 0 Tsi721 I/O Device Tsi721 I/O Device S-RIO Switch PCIe to S-RIO bridge replaces your NIC for <2W in a 13x13 package!! PAGE 14

RapidIO: Layered Protocol System Integration Implem mented in Ha ardware Logical Specification Transport Specification Physical Specification Logical I/O x8, x16 Parallel Message Passing Globally ll Shared Memory Transport x1, x2, x4, x8, x16 Serial Flow Control Data Streaming Multicast Future Physical Specs System Bring up Interop Error Mgmt PAGE 15

IO Virtualization using RapidIO Low latency Memory to Memory cut-through transactions Bypass for data transfer I/O virtualization in RapidIO 1000 s ofs with 16 bit deviceid Message Passing Logical Layer Map Mailbox, Letter, xmbox to IOV needs Direct IO Logical Layer Many nodes with 66-bit source/destination address and TID Data Streaming Logical Layer VM VNIC VM VNIC VMM VMM API Management Device Driver 1000 s of flows with streamid, xtype, xh CPU 0 Memory Memory RapidIO PAGE 16

x86 Server with RapidIO Gen2 (Today) PCIe to S-RIO bridge and Gen2 S-RIO Switch brings x86 CPU into RapidIO based Servers Multiple p links each 20G Multiple storage, networking, computing and chassis links Scalable Virtualized System PAGE 17

Native RapidIO Gen2 in ARM Server Native RapidIO with ARM RapidIO Gen2 Switch Multiple l links each 20G Multiple storage links Multiple compute node links Multiple networking links Lowest power Lowest latency Lowest footprint Secured I/O Virtualization ti with RDMA PAGE 18

x86 Server with RapidIO 10xN (Future) 10xN RapidIO-PCIe Bridge brings x86 into RapidIO based Servers Support Large Servers Live VM migration Lowest Switch Latency Modular Scalable Architecture Upto 160 Gbaud per port with 10.3125 G per lane PAGE 19

ARM Server with Native S-RIO 10xN Native on-chip RapidIO Interconnect Lowest Latency Lowest Power consumption Smallest Footprint Secured Virtualized Architecture PAGE 20

Example S-RIO based Server Cluster with RapidIO NIC Scalable Architecture t with RapidIO Switch and NIC Any kind of Topology PAGE 21

Summary RapidIOR Simplifies Virtualization ti Low overhead RDMA Live Virtual Machine migration 1000 s of nodes and flows 100 s of virtual NICs Lowest Latency Interconnect in the industry ~100 ns switch latency Green Solution 8W per switch and <2W per NIC Enables large Data Center for Cloud computing Low Power Micro servers with ARM and NATIVE RapidIO Macro servers with x86 and PCIe to RapidIO NIC RapidIO Simplifies IO Virtualization Lowest Latency, Power, and Cost PAGE 22

Visit Serial RapidIO Products at www.idt.com Product Data Sheets Reference Designs RapidIO Trade Association http://www.rapidio.org/home/news/membernews RapidIO Overview http://www.eetimes.com/design/signal-processing- g p g dsp/4017736/backplane-tutorial-rapidio-pcie-and-ethernet RapidIO versus Ethernet http://advancedtca-systems.com/comparing-ethernet-rapidio/ Contact Integrated Device Technology 6024 Silver Creek Valley Road San Jose, CA 95138 PAGE 23