At the heart of each Oracle PCA installation is a pair of management nodes. They are installed in rack units 5 and 6 and form a cluster in active/standby configuration for high availability: both servers are capable of running the same services and have equal access to the system configuration, but one operates as the master while the other is ready to take over the master functions in case a failure occurs. The master management node runs the full set of services required, while the standby management node runs a subset of services until it is promoted to the master role. The master role is determined at boot through OCFS2 Distributed Lock Management on an iSCSI LUN, which both management nodes share on the ZFS storage appliance installed at the bottom of the rack. Because rack units are numbered from the bottom up, and the bottom four are occupied by the ZFS Storage Appliance, the master management node is typically the server in rack unit 5. It is the only server that must be powered on by the administrator in the entire process to bring the appliance online.

For details about how high availability is achieved with Oracle PCA, refer to Section 1.5, “High Availability”.

When you power on the Oracle PCA for the first time, you can change the factory default IP configuration of the management node cluster, so that it can be easily reached from your data center network. The management nodes share a Virtual IP, where the management web interface can be accessed. This virtual IP is assigned to whichever server has the master role at any given time. During system initialization, after the management cluster is set up successfully, the master management node loads a number of Oracle Linux 6 services, in addition to Oracle VM and its associated MySQL database – including network, sshd, ntpd, iscsi initiator, dhcpd – to orchestrate the provisioning of all system components. During provisioning, all networking and storage is configured, and all compute nodes are discovered, installed and added to an Oracle VM server pool. All provisioning configurations are preloaded at the factory and should not be modified by the customer.

For details about the provisioning process, refer to Section 1.4, “Provisioning and Orchestration”.

The compute nodes in the Oracle PCA constitute the virtualization platform. The compute nodes provide the processing power and memory capacity for the virtual servers they host. The entire provisioning process is orchestrated by the management nodes: compute nodes are installed with Oracle VM Server 3.2.11 and additional packages for InfiniBand and Software Defined Networking. When provisioning is complete, the Oracle PCA software expects all compute nodes in the same rack to be part of the same Oracle VM server pool.

For hardware configuration details of the Oracle Server X6-2, Oracle Server X5-2, Sun Server X4-2 and Sun Server X3-2 compute nodes, refer to Server Components in the Oracle Private Cloud Appliance Installation Guide. Different generations of servers may be mixed within the same installation. In such configurations the version of the Oracle PCA controller software must support the most recent server model installed. Compute nodes of different hardware generations operate within the same server pool but belong to different CPU compatibility groups. Since live migration between CPU compatibility groups is not supported, virtual machines have to be cold-migrated between compute nodes of different generations. An exception to this rule can be implemented, but only if the migration occurs from an older model to a newer model. For more information about CPU compatibility groups, please refer to the section “Server Processor Compatibility Groups” in the Oracle VM User's Guide.

The Oracle PCA Dashboard allows the administrator to monitor the health and status of the compute nodes, as well as all other rack components, and perform certain system operations. The virtual infrastructure is configured and managed with Oracle VM Manager.

The Oracle PCA offers modular compute capacity that can be increased according to business needs. The minimum configuration of the base rack contains just two compute nodes, but it can be expanded by one node at a time up to 25 compute nodes. Apart from the hardware installation, adding compute nodes requires no intervention by the administrator. New nodes are discovered, powered on, installed and provisioned automatically by the master management node. The additional compute nodes are integrated into the existing configuration and, as a result, the Oracle VM server pool offers increased capacity for more or larger virtual machines.

Because of the diversity of possible virtualization scenarios it is difficult to quantify the compute capacity as a number of virtual machines. For sizing guidelines, refer to the chapter entitled Configuration Maximums in the Oracle Private Cloud Appliance Release Notes.

The Oracle ZFS Storage Appliance ZS3-ES installed at the bottom of the appliance rack should be considered a 'system disk' for the entire appliance. Its main purpose is to provide storage space for the Oracle PCA software. A portion of the disk space is made available for customer use and is sufficient for an Oracle VM storage repository with a limited number of virtual machines, templates and assemblies.

The hardware configuration of the Oracle ZFS Storage Appliance ZS3-ES is as follows:

The storage appliance is connected to the management subnet ( 192.168.4.0/24 ) and the InfiniBand (IPoIB) storage subnet ( 192.168.40.0/24 ). Both heads form a cluster in active-passive configuration to guarantee continuation of service in the event that one storage head should fail. The storage heads share a single IP in the storage subnet, but both have an individual management IP address for convenient maintenance access. The RAID-Z2 storage pool contains two projects, named OVCA and OVM .

The OVCA project contains all LUNs and file systems used by the Oracle PCA software:

The OVM project contains all LUNs and file systems used by Oracle VM:

In addition to offering storage, the ZFS storage appliance also runs the xinetd and tftpd services. These complement the Oracle Linux services on the master management node in order to orchestrate the provisioning of all Oracle PCA system components.

The Oracle Private Cloud Appliance relies on a combination of Ethernet connectivity and an InfiniBand network fabric. The appliance rack contains redundant network hardware components, which are pre-cabled at the factory to help ensure continuity of service in case a failure should occur.

Ethernet

The Ethernet network relies on two interconnected Oracle Switch ES1-24 switches, to which all other rack components are connected with CAT6 Ethernet cables. This network serves as the appliance management network, in which every component has a predefined IP address in the 192.168.4.0/24 range. In addition, all management and compute nodes have a second IP address in this range, which is used for Oracle Integrated Lights Out Manager (ILOM) connectivity.

While the appliance is initializing, the InfiniBand fabric is not accessible, which means that the management network is the only way to connect to the system. Therefore, the administrator should connect a workstation to the available Ethernet port 19 in one of the Oracle Switch ES1-24 switches, and assign the fixed IP address 192.168.4.254 to the workstation. From this workstation, the administrator opens a browser connection to the web server on the master management node at http://192.168.4.216 , in order to monitor the initialization process and perform the initial configuration steps when the appliance is powered on for the first time.

InfiniBand

The Oracle PCA rack contains two NM2-36P Sun Datacenter InfiniBand Expansion Switches. These redundant switches have redundant cable connections to both InfiniBand ports in each management node, compute node and storage head. Both InfiniBand switches, in turn, have redundant cable connections to both Fabric Interconnects in the rack. All these components combine to form a physical InfiniBand backplane with a 40Gbit (Quad Data Rate) bandwidth.

When the appliance initialization is complete, all necessary Oracle PCA software packages, including host drivers and InfiniBand kernel modules, have been installed and configured on each component. At this point, the system is capable of using software defined networking (SDN) configured on top of the physical InfiniBand fabric. SDN is implemented through the Fabric Interconnects.

Fabric Interconnect

All Oracle PCA network connectivity is managed through the Fabric Interconnects. Data is transferred across the physical InfiniBand fabric, but connectivity is implemented in the form of Software Defined Networks (SDN), which are sometimes referred to as 'clouds'. The physical InfiniBand backplane is capable of hosting thousands of virtual networks. These Private Virtual Interconnects (PVI) dynamically connect virtual machines and bare metal servers to networks, storage and other virtual machines, while maintaining the traffic separation of hard-wired connections and surpassing their performance.

During the initialization process of the Oracle PCA, five essential SDNs are configured: a storage network, an Oracle VM management network, a management Ethernet network, and two VLAN-enabled virtual machine networks.

Finally, the Fabric Interconnects also manage the physical public network connectivity of the Oracle PCA. Two 10GbE ports on each Fabric Interconnect must be connected to redundant next-level data center switches. At the end of the initialization process, the administrator assigns three reserved IP addresses from the data center (public) network range to the management node cluster of the Oracle PCA: one for each management node, and an additional Virtual IP shared by the clustered nodes. From this point forward, the Virtual IP is used to connect to the master management node's web server, which hosts both the Oracle PCA Dashboard and the Oracle VM Manager web interface.