windows server scalability and virtualized i/o fabric for blade server chris pettey cto, nextio son...
TRANSCRIPT
Windows Server Scalability Windows Server Scalability And Virtualized I/O Fabric And Virtualized I/O Fabric For Blade ServerFor Blade ServerChris PetteyChris PetteyCTO, NextIOCTO, NextIO
Son VoBaSon VoBaProgram Manager, Microsoft CorporationProgram Manager, Microsoft CorporationWindows Server Platform ArchitectureWindows Server Platform Architecture
AgendaAgenda
What is Shared I/O?What is Shared I/O?
Value of Shared I/OValue of Shared I/O
Architecture for I/O VirtualizationArchitecture for I/O Virtualization
User experienceUser experience
What Is Shared I/O?What Is Shared I/O?
Dedicated I/ODedicated I/O
Independent Operating Independent Operating Systems (OS)Systems (OS)
Each OS owns Each OS owns a physical adaptera physical adapter
Each OS owns Each OS owns the system components the system components to communicate to communicate to the adapterto the adapter
Example: Chipset and PCI Example: Chipset and PCI Express busExpress bus
OS #2 OS #3
I/OController
I/OController
OS #1
I/OController
Shared I/OShared I/O
Independent Operating Independent Operating Systems (OS)Systems (OS)
Each OS owns a virtualized Each OS owns a virtualized I/O controllerI/O controller
Corresponds to a shared Corresponds to a shared physical I/O controllerphysical I/O controller
Virtualization Enablers Virtualization Enablers control access control access to shared devicesto shared devices
I/OController
I/O
Controller
OS #2 OS #3OS #1
Virtualization Enablers
Characteristics Of Shared I/OCharacteristics Of Shared I/O
Multiple, independent operating systemsMultiple, independent operating systemsOSs do not coordinate with each otherOSs do not coordinate with each other
Each OS has a virtual set of resources to controlEach OS has a virtual set of resources to controlVirtual resources behave and interact in manners resembling Virtual resources behave and interact in manners resembling physical resourcesphysical resources
Single resources accessed by one or more OSsSingle resources accessed by one or more OSsSystem resources, e.g., Chipset, act as access points to I/OSystem resources, e.g., Chipset, act as access points to I/O
PCI Express acts as the connectivity from CPU and Chipset to I/OPCI Express acts as the connectivity from CPU and Chipset to I/O
I/O devices are simultaneously accessed by each OSI/O devices are simultaneously accessed by each OS
Virtualization Enablers control shared componentsVirtualization Enablers control shared componentsIsolate each OS for protectionIsolate each OS for protection
Provide virtual view of physical shared resourceProvide virtual view of physical shared resource
Manage functions for physical devicesManage functions for physical devices
Value Of Shared I/OValue Of Shared I/O
Shared I/O = Lower CostShared I/O = Lower Cost
CPU
CPU
PC
IeCPU
CPU
PC
IeEthernet
Fibre Channel Ethernet Switch
Fibre Channel SwitchCPU
CPU
PC
IeCPU
CPU
PC
Ie
Ethernet
Fibre Channel
CPU
CPU
PC
IeCPU
CPU
PC
Ie
Ethernet
Fibre Channel
InfiniBand
InfiniBand Switch
InfiniBand
InfiniBand
Shared I/O = Lower CostShared I/O = Lower Cost
Lower Acquisition costLower Acquisition cost
Lower TCOLower TCO
Lower Maintenance CostLower Maintenance Cost
Lower Upgrade CostLower Upgrade Cost
CPUCPU
CPUCPU PCIe
PCIe
CPU
CPU
PC
Ie
CPUCPU
CPUCPU PCIe
PCIe
CPU
CPU
PC
Ie
CPUCPU
CPUCPU PCIe
PCIe
CPU
CPU
PC
Ie
Shared I/OPCI Express
Switch
Shared Ethernet Controller
Shared Fibre Channel Controller
Shared InfiniBand Controller
New Technology?
Dedicated I/O Blade ServersDedicated I/O Blade Servers
10 Ethernet LOM chips10 Ethernet LOM chips
10 Fibre Channel 10 Fibre Channel oror InfiniBand Daughter InfiniBand Daughter CardsCards
4 Switches: 2 Ethernet Switches and 2 Fibre Channel 4 Switches: 2 Ethernet Switches and 2 Fibre Channel oror 2 InfiniBand 2 InfiniBand
Shared I/O Blade ServersShared I/O Blade Servers
2 Shared Ethernet chips2 Shared Ethernet chips
2 Shared Fibre Channel chips2 Shared Fibre Channel chips
2 Shared InfiniBand chips2 Shared InfiniBand chips
2 Shared I/O PCI Express Switches2 Shared I/O PCI Express Switches
Flexibility = Future ProofFlexibility = Future Proof
CPUCPU
CPUCPU PCIe
PCIe
CPU
CPU
PC
Ie
CPUCPU
CPUCPU PCIe
PCIe
CPU
CPU
PC
Ie
CPUCPU
CPUCPU PCIe
PCIe
CPU
CPU
PC
Ie
Shared I/OPCI Express Switch
Shared 1G Ethernet Controller
Shared 4G Fibre Channel Controller
Shared InfiniBand Controller
New Technology?
Shared 10G Ethernet Controller
Shared SAS Controller
Architecture For Architecture For I/O VirtualizationI/O Virtualization
Dedicated I/O Blade Server DesignDedicated I/O Blade Server Design
CPUCPU
CPUCPU PC
IeP
CIeCPU
CPU
Ethernet NIC
Fibre Channel HBA
Ethernet SwitchPC
Ie
CPUCPU
CPUCPU PC
IeP
CIeCPU
CPUFibre Channel HBA
Ethernet NIC
CPUCPU
CPUCPU PC
IeP
CIeCPU
CPU
Ethernet NIC
Fibre Channel HBABlade Server ChassisBlade Server Chassis
EnterpriseNetwork
Multiple intra-chassis fabricsMultiple intra-chassis fabrics
Fixed I/O configuration at orderFixed I/O configuration at order
Internal versus external switch Internal versus external switch compatibility concernscompatibility concerns
Management tools mustManagement tools mustcontemplate multiple fabricscontemplate multiple fabrics
PC
IeP
CIe Fibre Channel
Switch
FCFCFabricFabric
Shared I/O Blade Server DesignShared I/O Blade Server Design
CPUCPU
CPUCPU
CPU
CPU
CPUCPU
CPUCPU
CPU
CPU
CPUCPU
CPUCPU
CPU
CPU
Increased FlexibilityIncreased Flexibility
Simplified ManagementSimplified Management
Higher PerformanceHigher Performance
Lower CostLower Cost
Standard Standard PCI Express PCI Express
Mid-planeMid-plane Shared Ethernet NIC
Shared Fibre Channel HBA
Shared I/OPCI Express
Switch
Enhanced Enhanced PCI Express PCI Express
ProtocolProtocol
PC
IeP
CIe
PC
Ie
EnterpriseNetwork
FCFCFabricFabric
Blade Server ChassisBlade Server Chassis
Shared I/O In BladesShared I/O In Blades
Blades run independentlyBlades run independentlySoftware for OS, applications, etc. is unique for each bladeSoftware for OS, applications, etc. is unique for each blade
Blades have independent PCIe hierarchiesBlades have independent PCIe hierarchiesPCIe hierarchy is virtualizedPCIe hierarchy is virtualized
Root Complex, BIOS, etc. utilize PCIe as if it was dedicatedRoot Complex, BIOS, etc. utilize PCIe as if it was dedicated
Shared I/O components present virtualized componentsShared I/O components present virtualized componentsSwitches present multiple, virtual switchesSwitches present multiple, virtual switches
Controllers present multiple, virtual controllersControllers present multiple, virtual controllers
I/O Sharing through hardware componentsI/O Sharing through hardware componentsSwitches and Controllers combine to enable sharingSwitches and Controllers combine to enable sharing
Shared I/O In Virtual MachineShared I/O In Virtual Machine
Each Each OSOS runs independently runs independently
Virtual I/OVirtual I/O adapters appear adapters appear as physical componentsas physical components
Virtualization EnablerVirtualization Enabler is a combination of software is a combination of software and hardwareand hardware
Hypervisor softwareHypervisor software
PCIe Root Complex enablement, PCIe Root Complex enablement, e.g., Address Protection and e.g., Address Protection and Translation Table (ATPT) technologyTranslation Table (ATPT) technology
Physical SystemPhysical System is PCIe is PCIe I/O solutionI/O solution
Physical I/OPhysical I/O comprised of comprised ofSingle PCIe fabricSingle PCIe fabric
Dedicated I/O adaptersDedicated I/O adapters
Virtual Machine #1
Virtualization Enabler
Physical System
Virtual Machine #2
OS #2OS #1
Virtual I/O Virtual I/O
Physical I/O
Blade & Virtual Machine RelationshipBlade & Virtual Machine Relationship
SimilaritiesSimilaritiesEach OS is independentEach OS is independent
Legacy OSs Legacy OSs are fully supportedare fully supported
PCIe I/O systemPCIe I/O system
Single PCIe adapter Single PCIe adapter used by multiple OSsused by multiple OSs
Virtual adapters appear Virtual adapters appear as PCIe devicesas PCIe devices
DifferencesDifferencesVirtual Machine utilizes Virtual Machine utilizes software Hypervisor software Hypervisor for virtualizationfor virtualization
Blades utilize PCIe Switch Blades utilize PCIe Switch for virtualizationfor virtualization
Virtual Machine may use Virtual Machine may use traditional or Shared I/O traditional or Shared I/O enabled adaptersenabled adapters
Blades use Shared I/O Blades use Shared I/O enabled adaptersenabled adapters
Blade & Virtual Machine CombinationBlade & Virtual Machine Combination
Blade and Virtual Machine I/O Sharing are complementaryBlade and Virtual Machine I/O Sharing are complementary
Component solutions for Blades benefit Virtual MachineComponent solutions for Blades benefit Virtual Machine
Management model can service both Blades and Virtual MachinesManagement model can service both Blades and Virtual Machines
CPUCPU
CPUCPU
CPU
CPU
CPUCPU
CPUCPU
CPU
CPU
Shared Ethernet NIC
Shared FibreChannel HBA
Virt
ualiz
atio
n E
nabl
ers
OS
#2
OS
#3
OS
#1
Virt
ualiz
atio
n E
nabl
ers
OS
#2
OS
#3
OS
#1
PC
IeP
CIe
Shared I/OPCI Express
Switch
EnterpriseNetwork
FCFCFabricFabric
Blade Server ChassisBlade Server Chassis
Component Impact for Shared I/OComponent Impact for Shared I/O
BladeBlade Virtual MachineVirtual Machine
ChipsetChipset No impactNo impact ATPT functionsATPT functions
PCIe SwitchPCIe Switch Virtualized switchVirtualized switch No impactNo impact
I/O AdapterI/O Adapter Virtualized adapterVirtualized adapterNo impactNo impact
*Virtualized adapters provide *Virtualized adapters provide increased performanceincreased performance
OSOS No impactNo impact Hypervisor enabled OSHypervisor enabled OS
Guest OSGuest OS No impactNo impact No impactNo impact
User ExperienceUser Experience
Legacy OS and ApplicationsLegacy OS and Applications
Shared I/O fully supports legacy OS Shared I/O fully supports legacy OS and application softwareand application software
Applies equally to Blades and Virtual MachinesApplies equally to Blades and Virtual Machines
Blade solution fully supports all Microsoft OSBlade solution fully supports all Microsoft OS
No guest OS software introducedNo guest OS software introducedLegacy software is fully containedLegacy software is fully contained
No new “intermediate” drivers or 3No new “intermediate” drivers or 3rdrd party software party software
Guest OS and applications function normally Guest OS and applications function normally as in stand-alone serveras in stand-alone server
Migration from single server to Shared I/O is seamlessMigration from single server to Shared I/O is seamless
Devices And DriversDevices And Drivers
Blades support legacy devices dedicated to a single bladeBlades support legacy devices dedicated to a single blade
Virtual Machine enables software virtualization of legacy devicesVirtual Machine enables software virtualization of legacy devices
Blades and Virtual Machine enable Shared I/O enabled devices Blades and Virtual Machine enable Shared I/O enabled devices to be shared by multiple OSsto be shared by multiple OSs
Shared FibreChannel HBA
Shared Ethernet NIC
Dedicated Ethernet NIC
Dedicated Fibre Channel HBA
EnterpriseNetwork
FCFCFabricFabric
CPUCPU
CPUCPU
CPU
CPU
CPUCPU
CPUCPU
CPU
CPU
Virt
ualiz
atio
n E
nabl
ers
OS
#2
OS
#3
OS
#1
Virt
ualiz
atio
n E
nabl
ers
OS
#2
OS
#3
OS
#1
PC
IeP
CIe
Shared I/OPCI Express
Switch
Blade Server ChassisBlade Server Chassis
Management ModelManagement Model
Each OS manages its own virtual environmentEach OS manages its own virtual environmentBlades = Chipset, Virtual PCIe switch, and Virtual AdaptersBlades = Chipset, Virtual PCIe switch, and Virtual Adapters
Virtual Machine = Virtual Chipset and Virtual AdaptersVirtual Machine = Virtual Chipset and Virtual Adapters
Single control point manages common componentsSingle control point manages common componentsBlades = Switch Firmware for Switch and Shared I/O AdaptersBlades = Switch Firmware for Switch and Shared I/O Adapters
Virtual Machine = Hypervisor for all hardware componentsVirtual Machine = Hypervisor for all hardware components
Blade and Virtual Machine management can stageBlade and Virtual Machine management can stageBlade management partitions Switch and Adapters into virtual Blade management partitions Switch and Adapters into virtual PCIe systemsPCIe systems
Hypervisor partitions virtual PCIe system and chipset Hypervisor partitions virtual PCIe system and chipset into Virtual Machines into Virtual Machines
DeviceManagement
Tool
DeviceManagement
Tool
SwitchManagement
Tool
Common ManagementCommon Management
CPUCPU
CPUCPU
CPU
CPU
CPUCPU
CPUCPU
CPU
CPU
Shared I/OAdapter
Shared I/OAdapter
PC
IeP
CIe
Shared I/OPCI Express
Switch
Virtual Machine #1
Virtualization Enabler
Physical System
Virtual Machine #2
OS #2OS #1
Virtual I/O Virtual I/O
Physical I/O
Common Management CapabilitiesCommon Management Capabilities
Discovery of devices and capabilitiesDiscovery of devices and capabilitiesDiscovery, enumeration, etc. of PCIe devicesDiscovery, enumeration, etc. of PCIe devices
Partitioning of resourcesPartitioning of resourcesAssigning virtual resources to OSAssigning virtual resources to OS
Programming of shared functionsProgramming of shared functionsSetting operational parameters for physically Setting operational parameters for physically shared functionsshared functions
E.g., 10/100/1000 Ethernet link speedE.g., 10/100/1000 Ethernet link speed
Device specific functionDevice specific functionSupport for vendor specific device Support for vendor specific device management functionsmanagement functions
Common Management InterfaceCommon Management Interface
WS-ManagementWS-ManagementA DMTF Preliminary A DMTF Preliminary Standard defining Standard defining a web services a web services based protocolbased protocol
Is data model neutralIs data model neutral
A suitable management A suitable management protocol for both virtual protocol for both virtual devices and switchdevices and switch
DMTF CIMDMTF CIMProvides a consistent Provides a consistent abstraction of devices abstraction of devices and their virtual and their virtual environmentsenvironments
Independent of the Independent of the I/O Virtualization I/O Virtualization techniques and techniques and implementationsimplementations
Shared I/O and Industry StandardsShared I/O and Industry Standards
PCI SIG is standardizing elements PCI SIG is standardizing elements of I/O Virtualizationof I/O Virtualization
Virtual Machines and BladesVirtual Machines and Blades
Focused on I/O devices for PCIeFocused on I/O devices for PCIe
Microsoft and NextIO are active participants Microsoft and NextIO are active participants
Shared I/O leverages existing standardsShared I/O leverages existing standardsProtocols for Ethernet, FC, SAS, Protocols for Ethernet, FC, SAS, etc. unchangedetc. unchanged
Management models for DMTF, Management models for DMTF, etc. leveragedetc. leveraged
WinHEC Microsoft PavilionWinHEC Microsoft PavilionWinHEC Technical SessionWinHEC Technical Session
PCIe Switch In Blade PCIe Switch In Blade Server DesignsServer Designs
Technology DemoTechnology Demo
Shared I/O solution for Blade ServersShared I/O solution for Blade ServersPCI Express Switch with virtualization supportPCI Express Switch with virtualization support
Virtualized Fibre Channel HBAVirtualized Fibre Channel HBA
Unmodified Blade ChassisUnmodified Blade ChassisDell PowerEdge 1855 (Intel Xeon)Dell PowerEdge 1855 (Intel Xeon)
FSC PRIMERGY BX630 (AMD Opteron)FSC PRIMERGY BX630 (AMD Opteron)
Legacy supportLegacy supportNo change to OS or legacy driver No change to OS or legacy driver
No change to Chipset, Blades or ChassisNo change to Chipset, Blades or Chassis
Fibre ChannelSwitch
Technology DemoTechnology Demo
CPU
CPU
CPU
CPU
CPU
CPU
Fibre ChannelController
Fibre ChannelController
Fibre ChannelController
…Insert PCIe pass-thru cardsinstead of Fibre Channel Cards
PCI ExpressPass-thru
PCI ExpressPass-thru
PCI ExpressPass-thru
Note: The Fibre Channel midplanebecomes a PCI Express midplane
Hit EnterHit Enter
PC
IeP
CIe
PC
Ie
EthernetSwitch
EthernetController
EthernetController
EthernetController
NextIO PCIe Switch
Shared 4G Fibre Channel
PowerEdge 1855 or PRIMERGY BX630PowerEdge 1855 or PRIMERGY BX630
Shared I/O ValueShared I/O Value
Reduced component costReduced component cost
Reduced TCOReduced TCO
Increased flexibility Increased flexibility
Increase performanceIncrease performance
Zero impact to Legacy SoftwareZero impact to Legacy Software
Evolutionary solution starting in 2007Evolutionary solution starting in 2007
Call To ActionCall To Action
Plan for shared I/O in your next Blade Server designPlan for shared I/O in your next Blade Server design
Visit NextIO demo at the Microsoft PavilionVisit NextIO demo at the Microsoft Pavilion
Attend virtualization and management sessionsAttend virtualization and management sessionsDevice Virtualization Architecture (VIR040)Device Virtualization Architecture (VIR040)
How to Use the WMI Interfaces with Windows Virtualization (VIR043)How to Use the WMI Interfaces with Windows Virtualization (VIR043)
Hypervisor, Virtualization Stack, and Device Virtualization Architectures (VIR047)Hypervisor, Virtualization Stack, and Device Virtualization Architectures (VIR047)
PCIe Address Translation Services and I/O Virtualization (VIR071)PCIe Address Translation Services and I/O Virtualization (VIR071)
Windows Virtualization Best Practices and Future Hardware Directions (VIR124)Windows Virtualization Best Practices and Future Hardware Directions (VIR124)
Storage Management Directions (STO085)Storage Management Directions (STO085)
Windows Server Manageability Directions and Updates (SER120)Windows Server Manageability Directions and Updates (SER120)
Additional ResourcesAdditional Resources
Web ResourcesWeb ResourcesPCI-SIG (PCI-SIG (www.pcisig.orgwww.pcisig.org))
Distributed Management Task Force (Distributed Management Task Force (www.dmtf.orgwww.dmtf.org))
© 2006 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.
The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation.
MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.