TECHNICAL WHITE PAPER
Ubuntu OpenStack on VMware vSphere:
A reference architecture for deploying OpenStack while limiting changes
to existing infrastructure
A collaboration between Canonical and VMware
Summary
Canonical, the Ubuntu and OpenStack experts, and VMware, the virtualization experts, have combined their collective experience in customer facing
deployments. The team has built a series of reference architectures to help customers explore the benefits of OpenStack within their data centers.
The reference architecture in this paper is intended for organizations with
existing VMware data center deployments or expertise who want to limit
changes to their underlying VMware infrastructure, but see benefits in
a common abstraction and orchestration layer via OpenStack open APIs
and Dashboard to control compute workloads.
UBUNTU OPENSTACK PLUS VMWARE – A PERFECT MATCH IN THE EVOLUTION OF THE DATA CENTER
Organizations strive for increased operational efficiency and agility within their data centers which is why the old methodology of a single server per application evolved into server virtualization. This technology helped reduce data center hardware sprawl by consolidating multiple workloads per server resulting in higher hardware utilization. Prior to virtualization, the procurement, racking, stacking, provisioning, and networking of hardware generated overhead and took time. VMware quickly established itself as the leader in this space with a range of solutions to suit different organizations’ needs.
As the server virtualization footprint grew, organizations began to segregate and tailor deployments to various hypervisor stacks depending on cost, test and development workloads versus production deployments, compliance, etc.
This led to virtual machine (VM) sprawl and it became difficult to manage and scale applications running across the different hypervisors, servers, brands, etc.
The advent of “cloud computing”, especially private clouds, promised to alleviate this problem by creating a coherent environment that is easier to manage and scale. The OpenStack project is one of the prime examples in this area.
OpenStack is an open source computing platform for public and private clouds.
It is one of the largest and fastest growing open source projects to date.
OpenStack takes a set of heterogenous and isolated hypervisors (i.e. KVM, ESXi, Xen, LXC), storage and networks across a data center or multiple data centers and turns them into pools of resources. All managed and consumed via open APIs and a web-based dashboard. Ubuntu quickly established itself as the reference platform to develop and deploy OpenStack. And Canonical, the commercial sponsor of Ubuntu and platinum member of the OpenStack project, became the leader in helping organizations adopt and deploy
Ubuntu OpenStack as their public or private cloud technology. VMware joined the OpenStack project as a gold member in 2012 and announced a collaborative partnership with Canonical in 2013. The goal of this partnership was to aid organizations in their adoption of OpenStack, especially in combining it with their existing VMware infrastructure.
OpenStack as a control layer above pools of resources in the data center has benefits; however, organizations have heavy investments both financially and in staff technical competency with established VMware technologies, so how can they reap the benefits of a next generation cloud platform in OpenStack, while still getting the best out of their existing VMware hypervisor base?
What’s the best approach to educate their staff on OpenStack? OpenStack APIs
allow users to customize and configure down to the network level and VMware
NSX is one of the most advanced and feature rich SDN solutions available today
working seamlessly with OpenStack, ESXi and KVM, but how can this be done
without major disruptions? What changes are needed to applications to achieve
an “open cloud” using multiple hypervisors i.e. KVM for web tier apps and
VMware ESXi for more heavyweight backend applications?
Given the above pressures and scenarios organizations face in their adoption of OpenStack, VMware and Canonical created a collection of OpenStack migration best practices based on our experiences together in the field. A high-level overview of OpenStack migration options is given below, from the least to most “invasive”:
1. Maintain the existing VMware vCenter technology stack and deploy OpenStack services as VMs running on top of VMware’s ESXi hypervisor.
To minimize changes to the established VMware infrastructure even further, deploy OpenStack nova-network rather than OpenStack Neutron with an SDN. This allows organizations to familiarize and educate themselves on OpenStack (APIs) while maintaining a consistent and known infrastructure.
This environment is for proof of concept only.
2. Run OpenStack control services as hosts within VMware vCenter, but offer OpenStack compute options on multiple hypervisors, e.g. KVM and ESXi.
Implement VMware NSX as the SDN for a richer network topology. Use OpenStack regions or host aggregates to allow users the choose which compute hypervisor to deploy their workload on. In this approach, developers learn to make their workloads/applications hypervisor-agnostic by moving from failover to fault resistant “cloud oriented” designs. The data center infrastructure changes are minimally invasive.
3. Deploy OpenStack control services on bare-metal hardware or on an open source hypervisor such as KVM. Allow for multiple hypervisors (KVM, VMware ESXi, Xen, etc.) for OpenStack compute services and run VMware NSX as the SDN solution. This design encourages vendor diversity within the data center and turns a heterogeneous set of hypervisors, storage and network options into pools of resources available and configured on-demand.
In the next sections, we will outline the reference architecture specific to migration option number one. This migration option contains our recommended configuration, design, and implementation path matching real-world deployments
KEY ELEMENTS OF REFERENCE ARCHITECTURE
OpenStack Havana: The open source software for building private and public clouds Ubuntu 12.04 Long-term Support: The reference operating system for
OpenStack deployments and development
VMware vCenter version 5.1 or greater: The platform for managing VMware vSphere environments
INTENDED AUDIENCE
This paper assumes the reader is experienced with VMware vCenter and Ubuntu.
The reader should be familiar with OpenStack services (Compute, Keystone, etc.)
along with techniques to scale and segregate an OpenStack deployment.
VMware vSphere Design
OVERVIEW
The OpenStack components are installed as Virtual Machines in a vSphere Cluster.
This approach provides the following benefits:
• High availability via vSphere HA
• Better use of the hardware
• Flexibility to scale up and scale out easily as required
• Flexibility to adjust the specifications of each component ( RAM, Disk, vCPU, etc. )
• Faster deployment times
OPENSTACK DESIGN
Logical Ubuntu Openstack Cloud Design
Auth & API
OpenStack Cloud
CLI Horizon
Dashboard
Region One
AZ1
AZ2
Logical Ubuntu Cloud on vSphere Design
Networks
VM Cluster N (vSphere)
Instances (vSphere VMs)
VM Cluster 1 (vSphere)
Instances (vSphere VMs)
Management Cluster (vSphere)
MAAS
JuJu
Nagios
MySQL RabbitMQ
Nova Compute
VM Cluster 1
Nova Compute
VM Cluster 1
Ceph Nodes (x3)
Virtual Networks
Management Network VM Network
Virtual Networks Nova Cloud Controller
Keystone Cinder API Glance API OpenStack Dashboard
Ceph Rados Gateway
Design Notes:
• A floating network (not shown) is optional
• Each vSphere cluster is associated with a nova-compute. One cannot map multiple clusters to the same nova-compute, otherwise the clusters would get merged to look like a single hypervisor thereby removing the option of having clusters in different OpenStack availability zones
• This setup allows for one nova service and one nova.conf for both clusters and each is represented as a separate nova-compute hypervisor instance to the OpenStack Nova scheduler
• As of this writing, using one nova.conf for both clusters is not recommended since there is no established method to define clusters into individual OpenStack availability zones.
• OpenStack component HA is achieved via Juju and Metal-as-a-Service (MAAS)
• OpenStack services shown in the Management Cluster can be distributed
to other clusters depending on resource availability (not shown)
VMWARE ESXI HYPERVISORS
Network
OVERVIEW
Virtual networks exist to attach the VMs vNICs to the right physical networks.
These are the vSphere networks for the environment:
In this design, OpenStack Havana is implemented with nova-network.
OpenStack Neutron plus VMware NSX would be a recommended next step, but was not selected in this design.
DHCP AND DNS FOR THE OPENSTACK COMPONENTS
MAAS dynamically manages DHCP and DNS for all the OpenStack nodes using the Management Network.
The MAAS node will also provide the Ubuntu Precise 12.04 LTS base images to the VMs in the Ubuntu Cloud via PXE boot through the same network.
VM Attribute
2
Specification
4 GB Number of CPUs
Memory
1 (Management network) 20 GB
20 GB Number of vNIC ports Disk 1
Disk 2
vSphere Network
Network for the Ubuntu Cloud components:
• SSH traffic to access Ubuntu Cloud Components
• Internal Traffic between Ubuntu Cloud Components
• PXE booting
• iSCSI
• Ceph Storage
• Ceph Object Storage Description
Management
Only for vSphere, not related to the OpenStack infrastructure Flat network for the OpenStack instances traffic
VM Network
VMWare Management
MANAGEMENT NETWORK ISOLATION
This design consists of one main network called the Management Network.
Depending on your network configuration, you can connect a cloud portal or clients to this network to access the OpenStack APIs from other networks via routing.
For security reasons this network should be isolated and only accessible from trusted services like a portal or a management client machine.
Because this design is entirely on top of VMware vSphere running nova- network, OpenStack security groups are not available. As of this writing, OpenStack compute security group functionality is only achievable on vSphere when used in combination with VMware NSX SDN solution.
Storage
Each availability zone should have a Tier 2 SAN with sufficient resources for the planned workload available to be distributed via vSphere datastores to each vSphere cluster.
Notes:
• The vSphere datastores used for the instances should not be used for any other purpose
• Disconnect any other datastore from the ESXi hosts not to be used for the instances: http://docs.openstack.org/havana/config-reference/content/
vmware.html
OPENSTACK INSTANCES STORAGE
The OpenStack Instances are stored in a dedicated vSphere datastore.
BLOCK STORAGE WITH CINDER USING THE VMWARE DRIVER
OpenStack Cinder is handled using the VMware driver released with OpenStack Havana. Note: The current Cinder Juju Charm needs manual configuration after deployment to set up the VMware driver.
OBJECT STORAGE WITH CEPH RADOS GATEWAY
A minimal configuration of Object Storage is needed to deploy OpenStack instances via Juju. For that purpose Ceph RADOS Gateway will be deployed with a default configuration in 3 VMs.
Ceph RADOS Gateway will frontend the stored images and OpenStack Glance
will point to it.
VM Specification
The recommended specs for the Ceph VMs:
VM IMAGE STORAGE
The storage of the VM templates (images) is handled by the OpenStack Glance.
Glance provides multi-tenant image storage services for an OpenStack deployment.
In this design, to maximise availability of the images, Object Storage with Ceph RADOS Gateway will be used.
VM Attribute
2
Specification
4 GB Number of CPUs
Memory
1 (Management network) 20 GB
20 GB Number of vNIC ports Disk 1
Disk 2
