《SNIA-SDC23-White-Open-Programmable-Infrastructure-Project-Intro.pdf》由会员分享,可在线阅读,更多相关《SNIA-SDC23-White-Open-Programmable-Infrastructure-Project-Intro.pdf(27页珍藏版)》请在三个皮匠报告上搜索。
1、1|2023 SNIA.All Rights Reserved.Virtual ConferenceSeptember 28-29,2021Open Programmable Infrastructure Project IntroductionHow we can together implement DPU/IPU Infrastructure across all VendorsDr Joseph L White OPI TSC Chair&Dell Fellow/VP2|2023 SNIA.All Rights Reserved.AbstractThe Open Programmabl
2、e Infrastructure(or OPI)is an open-source effort within the Linux Foundation to develop a standard API for utilizing SmartNICs,DPUs and IPUs,and other coprocessors or processing elements.It will allow users to provision and orchestrate all devices in the same way,thus allowing them to handle many di
3、fferent devices,implement new devices,and change or replace devices without learning a new command structure.It will also allow manufacturers to create a standard API,deliver new or upgraded devices faster,and benefit from a large ecosystem.It makes learning curves for new devices shorter and implem
4、entation or software errors easier to find.It opens new markets for devices and eliminates concerns over one-of-a-kind implementations.This session will explore the goals and progress that the OPI project has undertaken.DPUs have many different use cases implemented by many different vendors.Our goa
5、l is to define a common framework for all of these devices to meet those use cases:This includes Infrastructure/workload isolation,Security,Network offload and acceleration,and Storage offload and acceleration.Over the last year since OPI joined the Linux foundation we have welcomed 14 member compan
6、ies to our project spanning the landscape from vendors,to integrators,to test infrastructure vendors,end users,as well as operating system and ISV vendors.We have sought to create common provisioning and lifecycle management frameworks,defined APIs for the management of these devices to meet the mos
7、t common use cases we have researched from end users and developed a developer platform and lab to test and explore these common frameworks.Come listen to industry experts as we explore the DPU/IPU ecosystem and the OPI projects progress toward a common set of frameworks,and how these assist end use
8、rs with ease of deployment,lowers the total cost of development and ownership,and thus provides for broader adoption of this new class of devices.opiproject.org34|2023 SNIA.All Rights Reserved.With Moores law slowing down,ever increasing demands for compute,and exponential growth in data trafficWe n
9、eed heterogeneous computeWe need composability.Workload specific resources per hostHyperscalers deploy DPU/IPUs w/non-standard frameworksWe want Standard APIs for Edge,Telco,EnterpriseHardware needs to be abstractedsolution providers can focus on deploying servicesEase of development&deploymentNeed
10、to drive efficiency in large computing environments TCO savingsStandards and common APIs needed to drive broader adoption of DPU/IPUs Flywheel effectStandardsEase of deployment/useLower costBroader adoption4Why should you care about OPI for DPU/IPU?5|2023 SNIA.All Rights Reserved.OPI Premier Members
11、OPI General Members56|2023 SNIA.All Rights Reserved.Project GoalsCreate community-driven standards-based open ecosystem for DPU/IPU-like technologiesCreate vendor agnostic framework and architecture for DPU/IPU-based software stacksReuse existing or define a set of new common APIs for DPU/IPU-like t
12、echnologies when requiredProvide implementation examples to validate the architectures/APIsCPUNICUSER APPLICATION/WORKLOADNETWORKSTORAGEMgmt&Security.Baremetal OS/Container Platform/(Partitioning)HypervisorDedicated ProcessorHW Accelerators/Offload EnginesPCIePhysical serverDPU/IPU-like device Barem
13、etal OS/Container Platform/HypervisorSW SUBSYSTEMOOBM/BMCTRUSTLINE67|2023 SNIA.All Rights Reserved.New chapter in modern system architectureCPUNICUSER APPLICATION/WORKLOADBaremetal OS/Container Platform/(Partitioning)HypervisorHW Accelerators/Offload EnginesPCIePhysical serverSmartNICNETWORK FUNCTIO
14、NSTORAGE FUNCTIONMgmt&Security Fn.Network Control PlaneStorage Control Plane Mgmt&Security CP.CPUNICUSER APPLICATION/WORKLOADNETWORKSTORAGEMgmt&Security.Baremetal OS/Container Platform/(Partitioning)HypervisorDedicated ProcessorHW Accelerators/Offload EnginesPCIePhysical serverDPU/IPU-like device Ba
15、remetal OS/Container Platform/HypervisorSW SUBSYSTEMTraditional SmartNIC modelComputer is CPU+SmartNIC as peripheral that is fully controlled by the CPUCPU+domain-specific HW accelerationStatic device functionDPU/IPU modelNIC&HW accelerators move to DPU/IPU-like device with its own CPUSoftware defin
16、ed device functionComputer is an aggregation of independently intelligent subsystemsOOBM/BMCTRUSTLINE78|2023 SNIA.All Rights Reserved.DPU Example ExpandedSwitchPCIeMgtBMCNICAcceleratorsGeneral CoresGPUP4 PipelinesBMCCPUMemoryStorageDPUCompressionEncryptionGPU,AI Processing4-32 or more Cores plusDRAM
17、&HBMPCIemgtProtocol AccelerationNVMe-oF,TCP,RoCEP4 PipelineInline micro-processorsStorage ControllerRegular ExpressionFPGA2-8 PortsEmbedded SwitchingNetwork Packet ProcessingNIC FunctionsEncryptionSecure KeysRegExHashingHosting System89|2023 SNIA.All Rights Reserved.DPU/IPU Use casesInfrastructure w
18、orkload isolationControl Plane offloadHost Lifecycle and ProvisioningHost OffloadsSecuritySecurity domains(Host and DPU/IPU)FW,intrusion detection and preventionNetworking offload and accelerationVirtual switch offload,IPSEC,TLSStorage offload and accelerationnvme/tcp offload,compression,and dedupe
19、acceleration9EdgeEnterpriseCloudTelco Core&EdgeApplies Across:10|2023 SNIA.All Rights Reserved.OPI Organizational StructureBoard of DirectorsOutreach CommitteeTechnical Steering CommitteeDeveloper PlatformUse CaseAPI&Behavioral ModelProvisioning&Lifecycle10OPI Technical DeliverablesOpen-Source Proje
20、ctsSpecifications/StandardsReference PlatformsTest Suites&CasesPOC/Prototypes11|2023 SNIA.All Rights Reserved.Scope and Goals of Working GroupsAPI&Behavioral ModelProvisioning&LifecycleDeveloper PlatformUse CaseIndependent testing LabVirtual&Hardware POCsSimulation EnvironmentCI/CDDiscovery&Provisio
21、ningInventoryBoot sequencingLifecycle&UpdatesMonitoring&TelemetryAreas of high interestStorage,Security,Networking,AI/MLUse cases gathered from end usersOVS/OVNNVMe/PCIe to NVMe/TCP bridgeBasic Firewall.Object modelsHost&Management facing APIsTaxonomy for Services(Networking,Storage,Security)Re-use
22、industry standard APIs(OpenConfig,VPP,FRR,etc)Reference Orchestration Client1112|2023 SNIA.All Rights Reserved.12opiproject.orgOPI Hugo WebsiteartworkOPI related logos and artwork.opiOPI Main RepositorySPDKopi-spdk-bridgeOPI Storage gRPC to SPDK json-rcp bridge POCopi-nvidia-bridgeOPI gRPC to Nvidia
23、 bridge third party repoopi-marvell-bridgeOPI gRPC to Marvell bridge third party repoopi-spdk-bridgeOPI storage gRPC to SPDK json-rpc bridgespdk-csi(Forked from spdk/spdk-csi)&spdkCSI driver to bring SPDK to Kubernetes storage through NVMe-oF or iSCSI.Supports dynamic volume provisioning and enables
24、 Pods to use SPDK storage transparently.sztp&sztpdSecure Zero Touch Provisioning(sZTP)in OPIopi-smbios-bridgeOPI gRPC to SMBIOS bridge for inventorysmbios-validation-tool(Forked from google/smbios-validation-tool)sessionOffload(Forked from att/sessionOffload)Open API for IP Applications to Offload T
25、CP/UDP Session Packet Processing to Hardwareopi-strongswan-bridgeOPI IPSEC gRPC to strongSwan vici API bridgeopi-prov-lifeProvisioning,Lifecycle and Platform ManagementDiscovery&ProvisioningInventoryBoot sequencingLifecycle&UpdatesMonitoring&Telemetry(OTEL)godpuA Container Storage Interface(CSI)libr
26、ary,client,and other helpful utilities created with Go for OPIpydpuPython library and cli to communicate with DPUs and IPUsopi-pocDeveloper Platform and PoC WorkIntegration Platform DefinitionSoftware Networking PoC via p4-ebpfspdk based storage device PoCopi-apiOpen Programmable Infrastructure API
27、and Behavioral ModelCreate a Taxonomy for services:NetworkingSecurityStorageGatewayTelemetryAI/MLotelCommon DPU Telemetry definitionOPI Repositories 13|2023 SNIA.All Rights Reserved.Momentum and Progress1314|2023 SNIA.All Rights Reserved.Create a Multi-Vendor Open API definition forStorage ServicesN
28、etwork ServicesSecurity ServicesAI/MLTelemetrySystem and Lifecycle Management14API ObjectivesStorageSecurityNetworkAI/MLAPIConsistent,standardized,protobuf based GRPC APIs for network and storage objectsVarious orchestration systems could integrate with the APIs once for various DPU/IPU vendorsDPU/I
29、PU vendors would write the shim/translation to underlying vendor specific SDKProvide Industry standard interface for DPU/IPU15|2023 SNIA.All Rights Reserved.Target use casesData center:On-prem cloud,public-cloud,5G cloudK8s clustersPhysical/Virtual appliance acceleration15API GoalsMulti-Tenant Share
30、d CloudEVPN GatewayIPSec(strongSwan)Storage Initiator/TargetInitially focus on Networking and Storage Interfaces for DPU/IPUProvide open source prototype/reference implementation to the APIs16|2023 SNIA.All Rights Reserved.16gRPC for configuration/control interface through API GatewayDirect delivery
31、 of gRPC messages to appropriate shim layergRPC to REST translationSupport gNMI and gNOIAPI MechanismExpose VF/PF for data consumptionNetworkingMulti-tenant public cloudEVPN based telco cloudK8s based on-prem deploymentNVMe Initiator/Target based storage virtualizationStorageSecurityStrongswan based
32、 IPSec implementation17|2023 SNIA.All Rights Reserved.17v1alpha API definition for storage solutionfrontend(host facing),middle-end(services for volumes),and backend(target specific)APIsReference implementation readyTo map the OPI APIs to SPDK code running on the DPU/IPUOPI SPDK Bridge https:/ suppo
33、rtOpen source bridge code for AMD,Intel,Marvell,and Nvidia DPU/IPUsCI/CD IntegrationContinuous testing/validation against the reference implementationProgress:Storage APIgRPC ClientOPI-SPDK gRPC Bridge/Server reference implementationStandard SPDKVendor HW Accelerationjson-rpc APIsDPDKOPI-APIsjson-rp
34、c API18|2023 SNIA.All Rights Reserved.18Create a common API framework and extensions forCloudTelcoK8sSupport network service capabilitiesOVS,VPP,SONiC,Leverage existing API modelsOVS,OpenConfig,CNI,Networking APIAPI gatewayload balancergRPCRESTNB API Layer(OpenConfig,CNI)Vendor SDKLinux KernelHW Acc
35、eleratorsNetwork ServiceClient19|2023 SNIA.All Rights Reserved.19v1alpha API definition for cloudMulti-tenant public cloud:https:/ based telco-cloud use cases:https:/ IPSec on DPU/IPUs:https:/ Reference implementationsProposals to leverage open source to build reference implementationsIPSec with str
36、ongSwan https:/ supportCloud APIs:AMDTelco Cloud APIs:IntelIPsec Security APIs:Intel,NvidiaCI/CD IntegrationContinuous testing/validation on the API definitionProgress:Network API20|2023 SNIA.All Rights Reserved.20APIs for implementing multi-tenant VPC(virtual private cloud)Tenant(VPC)Subnets/Networ
37、ksMapping(Endpoints)RoutesNetwork Security PoliciesVPC PeeringLooking to making a reference implementationOPI Networking API Multi-tenant Public Cloud 21|2023 SNIA.All Rights Reserved.21OPI Networking API EVPN Gateway22|2023 SNIA.All Rights Reserved.Demos22Networking Cloud DemoVideo recordingAPI pro
38、tocol buffersclients https:/ and https:/ DemoVideo recordingAPI protocol buffersclients https:/ and https:/ DemoVideo recording TBDAPI protocol buffersclients https:/ and https:/ SNIA.All Rights Reserved.NGINX POC Example F5 OPI TeamApplication PodApplication PodDPUTraffic Source PodTraffic Target P
39、odApplication PodTraffic Source PodTraffic Target PodTraffic Source PodTraffic Target PodHostHostDPUAcceptance Test for HW accelerationAbility to measure and compare HW acceleration featuresInbound DPU,handle traffic inbound to hostFirewall and TLS offloadOutput DPU,handle traffic from hostSwitching
40、 and Loadbalancing,ECMPWorks well if Host is a proxy of some sortHairpin Host-DPU-Host pure HW acceleration testNo expensive optics or switching required!Source-DPU-Host-DPU-Target is valid customer use case,but test is intended to be simpler24|2023 SNIA.All Rights Reserved.Traffic Generation.T-REX
41、Open Source?Apache Bench?Maybe we work with Keysight?Traffic Source MethodX86 or ARM bare metal host.DPUVM or server.RH OpenShift installed and clustered with RH SNO on DPUNGINX WebserverDPU Interface 1DPU Interface 2Host Interface 1Host Interface 1VLAN 1 ExternalVLAN 2InternalHost Interface 1PCIe b
42、usDPU MGMT Interface sZTP,TFTP,DHCP,DNS,etc servicesVLAN 3MgmtRH SNO mmmmNGINX ProxyVM or server.RH OpenShift installed and clustered with RH SNO on DPUNGINX WebserverVM or server.RH OpenShift installed and clustered with RH SNO on DPUNGINX WebserverVM or server.RH OpenShift installed and clustered
43、with RH SNO on DPUNGINX WebserverHost Interface 1Enough NGINX webservers to Max out DPU throughputRH OpenShiftNGINX WebserverRH OpenShiftNGINX WebserverNGINX POC Example F5 OPI Team25|2023 SNIA.All Rights Reserved.The objective of the Open Programmable Infrastructure Project is to foster a community-driven standards-based open ecosystem for next generation architectures and frameworks based on DPU/IPU-like SNIA.All Rights Reserved.Please take a moment to rate this session.Your feedback is important to us.