2 Considerations and Requirements #

The default networks set up in SUSE OpenStack Cloud are class C networks with 256 IP addresses each. This limits the maximum number of instances that can be started simultaneously. Addresses within the networks are allocated as outlined in the following table. Use the YaST Crowbar module to make customizations (see Chapter 7, Crowbar Setup). The last address in the IP address range of each network is always reserved as the broadcast address. This assignment cannot be changed.

For an overview of the minimum number of IP addresses needed for each of the ranges in the network settings, see Table 2.1, “Minimum Number of IP Addresses for Network Ranges”.

Note: Limitations of the Default Network Proposal

The default network proposal as described below limits the maximum number of Compute Nodes to 80, the maximum number of floating IP addresses to 61 and the maximum number of addresses in the nova_fixed network to 204.

To overcome these limitations you need to reconfigure the network setup by using appropriate address ranges. Do this by either using the YaST Crowbar module as described in Chapter 7, Crowbar Setup, or by manually editing the network template file as described in Section 7.5, “Custom Network Configuration”.

Note: Addresses for Additional Servers

Addresses not used in the ranges mentioned above can be used to add additional servers with static addresses to SUSE OpenStack Cloud. Such servers can be used to provide additional services. A SUSE Manager server inside SUSE OpenStack Cloud, for example, must be configured using one of these addresses.

SUSE OpenStack Cloud Crowbar supports different network modes defined in Crowbar: single, dual, and team. As of SUSE OpenStack Cloud Crowbar 8, the networking mode is applied to all nodes and the Administration Server. That means that all machines need to meet the hardware requirements for the chosen mode. The network mode can be configured using the YaST Crowbar module (Chapter 7, Crowbar Setup). The network mode cannot be changed after the cloud is deployed.

Other, more flexible network mode setups can be configured by manually editing the Crowbar network configuration files. See Section 7.5, “Custom Network Configuration” for more information. SUSE or a partner can assist you in creating a custom setup within the scope of a consulting services agreement (see http://www.suse.com/consulting/ for more information on SUSE consulting).

Important: Network Device Bonding is Required for HA

Network device bonding is required for an HA setup of SUSE OpenStack Cloud Crowbar. If you are planning to move your cloud to an HA setup at a later point in time, make sure to use a network mode in the YaST Crowbar that supports network device bonding.

Otherwise a migration to an HA setup is not supported.

2.1.2.1 Single Network Mode #

 

In single mode you use one Ethernet card for all the traffic:

2.1.2.2 Dual Network Mode #

 

Dual mode needs two Ethernet cards (on all nodes but Administration Server) to completely separate traffic between the Admin Network and the public network:

2.1.2.3 Team Network Mode #

 

Team mode is similar to single mode, except that you combine several Ethernet cards to a “bond” (network device bonding). Team mode needs two or more Ethernet cards.

When using team mode, you must choose a “bonding policy” that defines how to use the combined Ethernet cards. You can either set them up for fault tolerance, performance (load balancing), or a combination of both.

Enabling access to the Administration Server from another network requires an external gateway. This option offers maximum flexibility, but requires additional hardware and may be less secure than you require. Therefore SUSE OpenStack Cloud offers a second option for accessing the Administration Server: the bastion network. You only need a dedicated Ethernet card and a static IP address from the external network to set it up.

The bastion network setup (see Section 7.3.1, “Setting Up a Bastion Network” for setup instructions) enables logging in to the Administration Server via SSH from the company network. A direct login to other nodes in the cloud is not possible. However, the Administration Server can act as a “jump host”: First log in to the Administration Server via SSH, then log in via SSH to other nodes.

The Administration Server acts as a name server for all nodes in the cloud. If the Administration Server has access to the outside, then you can add additional name servers that are automatically used to forward requests. If additional name servers are found on your cloud deployment, the name server on the Administration Server is automatically configured to forward requests for non-local records to these servers.

The Administration Server must have a fully qualified host name. The domain name you specify is used for the DNS zone. It is required to use a sub-domain such as cloud.example.com. The Administration Server must have authority over the domain it is on so that it can create records for discovered nodes. As a result, it will not forward requests for names it cannot resolve in this domain, and thus cannot resolve names for the second-level domain, .e.g. example.com, other than for nodes in the cloud.

This host name must not be changed after SUSE OpenStack Cloud has been deployed. The OpenStack nodes are named after their MAC address by default, but you can provide aliases, which are easier to remember when allocating the nodes. The aliases for the OpenStack nodes can be changed at any time. It is useful to have a list of MAC addresses and the intended use of the corresponding host at hand when deploying the OpenStack nodes.

SUSE OpenStack Cloud offers two different types of services for persistent storage: object and block storage. Object storage lets you upload and download files (similar to an FTP server), whereas a block storage provides mountable devices (similar to a hard disk partition). SUSE OpenStack Cloud provides a repository to store the virtual disk images used to start instances.

Each node in SUSE OpenStack Cloud needs sufficient disk space to store both the operating system and additional data. Requirements and recommendations for the various node types are listed below.

Important: Choose a Hard Disk for the Operating System Installation

The operating system will always be installed on the first hard disk. This is the disk that is listed first in the BIOS, the one from which the machine will boot. Make sure that the hard disk the operating system is installed on will be recognized as the first disk.

2.2.2.3 Compute Nodes #

 

Unless an instance is started via “Boot from Volume”, it is started with at least one disk, which is a copy of the image from which it has been started. Depending on the flavor you start, the instance may also have a second, so-called “ephemeral” disk. The size of the root disk depends on the image itself. Ephemeral disks are always created as sparse image files that grow up to a defined size as they are “filled”. By default ephemeral disks have a size of 10 GB.

Both disks, root images and ephemeral disk, are directly bound to the instance and are deleted when the instance is terminated. These disks are bound to the Compute Node on which the instance has been started. The disks are created under /var/lib/nova on the Compute Node. Your Compute Nodes should be equipped with enough disk space to store the root images and ephemeral disks.

Note: Ephemeral Disks vs. Block Storage

Do not confuse ephemeral disks with persistent block storage. In addition to an ephemeral disk, which is automatically provided with most instance flavors, you can optionally add a persistent storage device provided by cinder. Ephemeral disks are deleted when the instance terminates, while persistent storage devices can be reused in another instance.

The maximum disk space required on a compute node depends on the available flavors. A flavor specifies the number of CPUs, RAM, and disk size of an instance. Several flavors ranging from tiny (1 CPU, 512 MB RAM, no ephemeral disk) to xlarge (8 CPUs, 8 GB RAM, 10 GB ephemeral disk) are available by default. Adding custom flavors, and editing and deleting existing flavors is also supported.

To calculate the minimum disk space needed on a compute node, you need to determine the highest disk-space-to-RAM ratio from your flavors. For example:

Flavor small: 2 GB RAM, 100 GB ephemeral disk => 50 GB disk /1 GB RAM

Flavor large: 8 GB RAM, 200 GB ephemeral disk => 25 GB disk /1 GB RAM

So, 50 GB disk /1 GB RAM is the ratio that matters. If you multiply that value by the amount of RAM in GB available on your compute node, you have the minimum disk space required by ephemeral disks. Pad that value with sufficient space for the root disks plus a buffer to leave room for flavors with a higher disk-space-to-RAM ratio in the future.

Warning: Overcommitting Disk Space

The scheduler that decides in which node an instance is started does not check for available disk space. If there is no disk space left on a compute node, this will not only cause data loss on the instances, but the compute node itself will also stop operating. Therefore you must make sure all compute nodes are equipped with enough hard disk space.

The Compute Nodes need to be equipped with a sufficient amount of RAM and CPUs, matching the numbers required by the maximum number of instances running concurrently. An instance started in SUSE OpenStack Cloud cannot share resources from several physical nodes. It uses the resources of the node on which it was started. So if you offer a flavor (see Flavor for a definition) with 8 CPUs and 12 GB RAM, at least one of your nodes should be able to provide these resources. Add 1 GB RAM for every two nodes (including Control Nodes and Storage Nodes) deployed in your cloud.

See Section 2.2.2.3, “Compute Nodes” for storage requirements.

Usually a single CPU and a minimum of 4 GB RAM are sufficient for the Storage Nodes. Memory requirements increase depending on the total number of nodes in SUSE OpenStack Cloud—the higher the number of nodes, the more RAM you need. A deployment with 50 nodes requires a minimum of 20 GB for each Storage Node. If you use Ceph as storage, the storage nodes should be equipped with an additional 2 GB RAM per OSD (Ceph object storage daemon).

For storage requirements, see Section 2.2.2.4, “Storage Nodes (optional)”.

The monasca Node is a dedicated physical machine that runs the monasca-server role. This node is used for SUSE OpenStack Cloud Crowbar Monitoring. Hardware requirements for the monasca Node are as follows:

The following formula can be used to calculate the required disk space:

200 GB + ["number of nodes" * "retention period" * ("space for log
   data/day" + "space for metrics data/day") ]

The recommended values for the formula are as follows:

The formula is based on the following log data assumptions:

The formula is based on the following metrics data assumptions:

The formula provides only a rough estimation of the required disk space. There are several factors that can affect disk space requirements. This includes the exact combination of services that run on your OpenStack node actual cloud usage pattern, and whether any or all services have debug logging enabled.

SUSE OpenStack Cloud Crowbar can be extended by SUSE Enterprise Storage for setting up a Ceph cluster providing block storage services. To store virtual disks for instances, SUSE OpenStack Cloud uses block storage provided by the cinder module. cinder itself needs a back-end providing storage. In production environments this usually is a network storage solution. cinder can use a variety of network storage back-ends, among them solutions from EMC, Fujitsu, or NetApp. In case your organization does not provide a network storage solution that can be used with SUSE OpenStack Cloud, you can set up a Ceph cluster with SUSE Enterprise Storage. SUSE Enterprise Storage provides a reliable and fast distributed storage architecture using commodity hardware platforms.

You need seven software repositories to deploy SUSE OpenStack Cloud and to keep a running SUSE OpenStack Cloud up-to-date. This includes the static product repositories, which do not change over the product life cycle, and the update repositories, which constantly change. The following repositories are needed:

As explained in Section 2.6, “High Availability”, Control Nodes in SUSE OpenStack Cloud can optionally be made highly available with the SUSE Linux Enterprise High Availability Extension. The following repositories are required to deploy SLES High Availability Extension nodes:

The product repositories for SUSE Linux Enterprise Server 12 SP4 and SUSE OpenStack Cloud Crowbar 9 do not change during the life cycle of a product. Thus, they can be copied to the destination directory from the installation media. However, the pool and update repositories must be kept synchronized with their sources on the SUSE Customer Center. SUSE offers two products that synchronize repositories and make them available within your organization: SUSE Manager (http://www.suse.com/products/suse-manager/, and Subscription Management Tool (which ships with SUSE Linux Enterprise Server 12 SP4).

All repositories must be served via HTTP to be available for SUSE OpenStack Cloud deployment. Repositories that are installed on the Administration Server are made available by the Apache Web server running on the Administration Server. If your organization already uses SUSE Manager or SMT, you can use the repositories provided by these servers.

Making the repositories locally available on the Administration Server has the advantage of a simple network setup within SUSE OpenStack Cloud, and it allows you to seal off the SUSE OpenStack Cloud network from other networks in your organization. Hosting the repositories on a remote server has the advantage of using existing resources and services, and it makes setting up the Administration Server much easier. However, this requires a custom network setup for SUSE OpenStack Cloud, since the Administration Server needs access to the remote server.

The Administration Server provides all services needed to manage and deploy all other nodes in the cloud. If the Administration Server is not available, new cloud nodes cannot be allocated, and you cannot add new roles to cloud nodes.

However, only two services on the Administration Server are single points of failure, without which the cloud cannot continue to run properly: DNS and NTP.

2.6.1.1 Administration Server—Avoiding Points of Failure #

 

To avoid DNS and NTP as potential points of failure, deploy the roles dns-server and ntp-server to multiple nodes.

Note: Access to External Network

If any configured DNS forwarder or NTP external server is not reachable through the admin network from these nodes, allocate an address in the public network for each node that has the dns-server and ntp-server roles:

crowbar network allocate_ip default `hostname -f` public host

Then the nodes can use the public gateway to reach the external servers. The change will only become effective after the next run of chef-client on the affected nodes.

The Control Node(s) usually run a variety of services without which the cloud would not be able to run properly.

2.6.2.1 Control Node(s)—Avoiding Points of Failure #

 

To prevent the cloud from avoidable downtime if one or more Control Nodes fail, you can make the following roles highly available:

• database-server (database barclamp)

• keystone-server (keystone barclamp)

• rabbitmq-server (rabbitmq barclamp)

• swift-proxy (swift barclamp)

• glance-server (glance barclamp)

• cinder-controller (cinder barclamp)

• neutron-server (neutron barclamp)

• neutron-network (neutron barclamp)

• nova-controller (nova barclamp)

• nova_dashboard-server (nova_dashboard barclamp)

• ceilometer-server (ceilometer barclamp)

• ceilometer-control (ceilometer barclamp)

• heat-server (heat barclamp)

• octavia-api (octavia barclamp)

• octavia-backend (octavia barclamp)

Instead of assigning these roles to individual cloud nodes, you can assign them to one or several High Availability clusters. SUSE OpenStack Cloud Crowbar will then use the Pacemaker cluster stack (shipped with the SUSE Linux Enterprise High Availability Extension) to manage the services. If one Control Node fails, the services will fail over to another Control Node. For details on the Pacemaker cluster stack and the SUSE Linux Enterprise High Availability Extension, refer to the High Availability Guide, available at https://documentation.suse.com/sle-ha/15-SP1/. Note that SUSE Linux Enterprise High Availability Extension includes Linux Virtual Server as the load-balancer, and SUSE OpenStack Cloud Crowbar uses HAProxy for this purpose (http://haproxy.1wt.eu/).

Note: Recommended Setup

Though it is possible to use the same cluster for all of the roles above, the recommended setup is to use three clusters and to deploy the roles as follows:

• data cluster: database-server and rabbitmq-server

• network cluster: neutron-network (as the neutron-network role may result in heavy network load and CPU impact)

• services cluster: all other roles listed above (as they are related to API/schedulers)

Important: Cluster Requirements and Recommendations

For setting up the clusters, some special requirements and recommendations apply. For details, refer to Section 2.6.5, “Cluster Requirements and Recommendations”.

If a Compute Node fails, all VMs running on that node will go down. While it cannot protect against failures of individual VMs, a High Availability setup for Compute Nodes helps to minimize VM downtime caused by Compute Node failures. If the nova-compute service or libvirtd fail on a Compute Node, Pacemaker will try to automatically recover them. If recovery fails, or the node itself should become unreachable, the node will be fenced and the VMs will be moved to a different Compute Node.

If you decide to use High Availability for Compute Nodes, your Compute Node will be run as Pacemaker remote nodes. With the pacemaker-remote service, High Availability clusters can be extended to control remote nodes without any impact on scalability, and without having to install the full cluster stack (including corosync) on the remote nodes. Instead, each Compute Node only runs the pacemaker-remote service. The service acts as a proxy, allowing the cluster stack on the “normal” cluster nodes to connect to it and to control services remotely. Thus, the node is effectively integrated into the cluster as a remote node. In this way, the services running on the OpenStack compute nodes can be controlled from the core Pacemaker cluster in a lightweight, scalable fashion.

Find more information about the pacemaker_remote service in Pacemaker Remote—Extending High Availability into Virtual Nodes, available at http://www.clusterlabs.org/doc/.

To configure High Availability for Compute Nodes, you need to adjust the following barclamp proposals:

SUSE OpenStack Cloud Crowbar offers two different types of storage that can be used for the Storage Nodes: object storage (provided by the OpenStack swift component) and block storage (provided by Ceph).

Both already consider High Availability aspects by design, therefore it does not require much effort to make the storage highly available.

When considering setting up one or more High Availability clusters, refer to the chapter System Requirements in the High Availability Guide for SUSE Linux Enterprise High Availability Extension. The guide is available at https://documentation.suse.com/sle-ha/15-SP1/.

The HA requirements for Control Node also apply to SUSE OpenStack Cloud Crowbar. Note that by buying SUSE OpenStack Cloud Crowbar, you automatically get an entitlement for SUSE Linux Enterprise High Availability Extension.

Especially note the following requirements:

For a basic understanding and detailed information on the SUSE Linux Enterprise High Availability Extension (including the Pacemaker cluster stack), read the High Availability Guide. It is available at https://documentation.suse.com/sle-ha/15-SP1/.

In addition to the chapters mentioned in Section 2.6.5, “Cluster Requirements and Recommendations”, the following chapters are especially recommended:

The High Availability Guide also provides comprehensive information about the cluster management tools with which you can view and check the cluster status in SUSE OpenStack Cloud Crowbar. They can also be used to look up details like configuration of cluster resources or global cluster options. Read the following chapters for more information: