DELL EMC TECHAL SOLUTION BRIEF ARCHITECTING A CLOUD FABRIC WHEN DEPLOING VIRTUALIZATION OVERLAS Version 2.0 Author: VICTOR LAMA Dell EMC Networking SE May 2017
Architecting a Data Center Cloud Fabric: Leveraging Virtualization Overlays Organizations interested in building a private cloud environment will typically have the following network-related requirements: Multitenancy (The ability of a network to provide isolated logical segments for use by different customers/entities aka Tenants) Centralized (programmatic) network management and control Integration with Cloud Management Platforms Support for cloud-native applications, virtualized compute and container orchestration engines, such as Docker and Kubernetes Pervasive multi-layer, inter-tenant/inter-element security to address horizontal traffic flows between workloads within the data center A. Possible Cloud Networking Solutions: There are generally two approaches to delivering multitenancy cloud fabrics. Both involve the instantiation of virtualized networks via one of the following methods: Physical network (underlay) virtualization, i.e. network slicing via the classic SDN approach. For the purpose of this solution brief, the term SDN refers to the Open Networking Forum s description of a centralized control plane that is abstracted from, and programmatically provisions, the data plane. Network Virtualization Overlays (NVO) are logical constructs that are abstracted from the underlying physical network via a data plane encapsulation protocol (i.e. VxLAN, NVGRE, etc.). In this scenario, the underlay becomes an engine for generic IP forwarding, while all the complexity is relegated to the network edge (hypervisor or hardware gateway). It is also possible to deploy a combination of both to create an intelligent underlying physical network that has visibility to the virtualized overlay environment, including all tenant networks and endpoints. This would facilitate troubleshooting and allow the physical and virtual network environments to be managed from a single graphical interface (See Big Cloud Fabric [BCF] from Big Switch Networks [BSN]).
Page 2 of 6 Figure 1 Understanding the difference between an Underlay and an Overlay SPINE SPINE SPINE SPINE Dell-OS9/OS10 Dell-Cumulus Dell-Big Switch PHSICAL NETWORK (UNDERLA) NSX/OpenStack Tenant Tunnels (OVERLAS) Victor Lama Dell
Page 3 of 6 This particular solution brief assumes that an NVO-based approach, such as ware s NSX or OpenStack Neutron, will be deployed to deliver a virtualized multitenancy network. Both NVO architectures can deliver the following: Overlay Network Virtualization abstractions (hypervisors act as tunnel endpoints, encapsulation/de-encapsulation engines) Centralized Controllers that manage virtual networking constructs Micro-segmentation for pervasive security (native support with NSX, support via 3 rd party vendors for OpenStack) Integration with Kubernetes Integration with an SDN-based intelligent underlay (See choice 4 in Section B below) B. Possible Accompanying Physical Network (Underlay) Solutions: Now the task is to build the physical network infrastructure to which the physical servers, the NSX/OpenStack controllers, and other network peripherals need to connect. Depending on the design requirements, the network fabric can be a leaf and spine (Clos) or a traditional hierarchical, three-tiered tree architecture (access, aggregation and core). All Dell EMC data center switches can support both architectures with switch interfaces that can simultaneously forward Ethernet traffic at line rate without any oversubscription in the backplane (non-blocking). All Dell EMC switches that have the ON designation (Open Networking) leverage the ONIE (Open Networking Install Environment) installer, which is a lightweight Linux operating system ( shim ) with a bootloader that enables the switch to load multiple, validated thirdparty operating systems. This is what is termed the disaggregated model, and it is similar to the manner in which an x86 compute platform can run different operating systems, such as UNIX, Linux or Windows (See figure 2). All Dell EMC data center S and Z-Series switches are ON-enabled and leverage advanced chipsets from Broadcom. As such, several choices of switch operating systems and concomitant physical network (underlay) architectures present themselves to the cloud architect.
Page 4 of 6 Figure 2 The Open Networking Software Disaggregation Model Vertically Integrated Traditional Switches SOFTWARE NX-OS JUNOS OS9 Cisco Juniper Dell HARDWARE Disaggregated Model Open Networking SOFTWARE Cumulus, Big Switch, OS10 ONIE ABSTRACTION Dell HARDWARE Disaggregation provides the cloud architect with the flexibility to choose a network switch operating system that best fits the desired architecture and operational paradigm for configuring, provisioning and managing the cloud infrastructure. Table 1 below provides some insight into the available tool sets and technology ecosystem integration that each Dell EMC network switch operating system, or partner operating system, offers. NOTE: There are other Dell EMC Networking software partner vendors that are part of the Open Networking ecosystem and will be showcased in other Solution Briefs. Victor Lama Dell The Dell EMC Networking physical underlay operating system choices are as follows: 1. [LEGAC] S/Z-Series ON Switches + OS 9.x 2. [ON-LINUX] S/Z-Series ON Switches + OS 10.x 3. [ON-LINUX] S/Z-Series ON Switches + Cumulus Linux 4. [ON-LINUX] S/Z-Series ON Switches + Pluribus Open Netvisor Linux (ONVL) 5. [ON-SDN] S/Z-Series ON Switches + BSN Switch Light OS and Switch Light Virtual (If OpenStack P+V)
Page 5 of 6 Underlay Fabric Supported Topology L2 L3 Automation and Orchestration NSX Integration OpenStack Integration Supported Dell Switch LEGAC OS9.x REST-API, Smart Scripting, Puppet, Ansible, others Hardware VTEP/OVSDB Hardware VTEP/OVSDB All Dell Switches ON-LINUX OS10.x Linux API, Existing Linux Tools (Chef, Puppet, Ansible, etc.) Hardware VTEP/OVSDB (OS10.4.1 Q1C18) Hardware VTEP/OVSDB (OS10.4.1 Q1C18) S4100-ON and Future S/Z-Series ON Switches ON-LINUX Cumulus Linux API, Existing Linux Tools (Chef, Ansible, etc.), Zero Touch- Provisioning (ZTP) Hardware VTEP/OVSDB Hardware VTEP/OVSDB, REST API + ML2 Mechanism Driver All S/Z-Series ON-based Switches ON-LINUX ONVL Linux API, Ansible, controller-less distributed control plane (vport) Q4 C17 Netvisor ML2 Plugin (Q4 2017) All S/Z-Series ON-based Switches ON- SDN Big Switch BCF (Clos) REST API, Controller-Based Automation and Orchestration, NSX-specific Automation Hardware VTEP, APIbased with vcenter (Deep Integration) Redhat and Mirantis integration. Canonical supported. OVS user Space agent (Switch Light VX), BSN L3 plugin for OpenStack Controller and compute nodes, ML2 drivers (Deep Integration) S/Z-Series 10/40/100G (multi-rate) ON Switches Table 1 Dell EMC Network Switch Operating System Characteristics
Page 6 of 6 SUMMAR Multitenant cloud network fabrics can be constructed by implementing network virtualization in one of two manners; either by programmatically configuring the physical network via an SDN controller, or as in the case of this solution brief by deploying Network Virtualization Overlays (NVO), aka tunnels, where the tenant tunnels are abstracted from the physical network and all the complexity is pushed to the compute edge. In that case, the physical network becomes a simplified IP forwarding engine. Dell EMC offers the choice to take either approach by exploiting Open Networking abstractions (hardware-software disaggregation), advanced merchant silicon from Broadcom (and other vendors in the future), and an ecosystem of innovative software partners whose switch operating systems can provide deep integration, automation and orchestration with cloud management platforms, self-service portals and other cloud elements. For further information and a deeper understanding of the technologies mentioned in this brief, please see the following links. http://en.community.dell.com/techcenter/extras/m/white_papers/20442752 http://en.community.dell.com/techcenter/networking/w/wiki http://en.community.dell.com/techcenter/networking/w/wiki/11619.dell-emc-networking-and-big-switch-networks http://en.community.dell.com/techcenter/networking/w/wiki/7481.dell-emc-networking-and-cumulus-linux-os Learn more at Dell.com/Networking Copyright 2017 Dell Inc. or its subsidiaries. All Rights Reserved. Dell, EMC and other trademarks are trademarks of Dell Inc. or its subsidiaries. Other trademarks may be the property of their respective owners. Published in the USA.