| 1.1 Identifying the Networks | |
| HP Cluster Platform building blocks contain HP ProCurve switches organized into three discrete switch networks: | Back |
| • Administrative Ethernet network | TOC |
| • Console Ethernet network | |
| • Interconnect Ethernet network | |
| These networks are organized in a flat tree topology, which is a two-level hierarchy of root switches with links to branch switches. For more information on these networks, refer to the HP Cluster Platform Overview and the HP Cluster Platform Site Planning Guide. | |
| The network consists of links between the HP ProCurve switches and following port types: | |
| • Nodes | |
| Each node in the cluster provides two or more network interface cards (NICs). Depending on the server’s format, these NICs might be embedded into the motherboard or installed as option cards in a PCI bus slot. One NIC is dedicated to a console or server management function, enabling you (or a system process) to connect to the node remotely and perform basic system operations such as boot and shutdown. This NIC is part of a group of ports identified on the server’s rear bezel by the following labels: | |
| – Management Processor | |
| – Integrated Lights Out | |
| A second NIC is a general-purpose NIC that is used for the cluster administrative network for operations such as job management. | |
| Both 10/100 Base-T and Gigabit Ethernet are supported, depending on NIC availability, and the type of switch installed in the cluster. The actual role and function of these NICs is determined by the operating environment software and you usually access these links only by using the operating environment. | |
| • Interconnect Controller Card | |
| The interconnect’s control card provides a network port that lets you connect to and administer the interconnect by using its integrated control processor and system firmware. A generic term of controller card is used to describe these cards. For example, the Myrinet Interconnect’s monitoring line card and the InfiniBand™ interconnect’s SMB card are examples of controller cards. | |
| • Network Switch Port-to-Port | |
| In addition to the connections between the network switches and the NICs in each node, there are connections between the network switch ports to combine the switches into a flat tree network topology. In some instances, these links have two physical port-to-port links (trunked links) providing one logical connection. Such trunked links increase the available link bandwidth, providing the operating environment (such as HP XC) can utilize this type of connection. | |
| The root switches are configured to optimize system management. Due to this, the ports must be wired per the tables in this document. Deviations will likely result in functional and/or performance issues. The order for wiring is as follows: | |
| 1. Reserved ports are not to be used unless specifically identified in the tables. | |
| 2. Uplinks from branch switches within the CBBs begin with port 1 of the root switch. CBB1 connects to port 1 (and 2 if trunking applicable). | |
| 3. Uplinks from the administrative network branch switches to the administrative root switches are set for trunking (two ports) when a ProCurve 2848 is used within the CBBs; CBBs containing ProCurve 2824 branch switches uplink through a single Ethernet port. | |
| 4. The control node is connected to port 42 (on the ProCurve 2650 and ProCurve 2848) on the administrative and console root Ethernet network switches. In single cabinet high-density solutions, the less dense ProCurve 2824 and ProCurve 2626 switches are used and the control node is connected to port 22. | |
| 5. Uplinks from Scalable Visualization Array (SVA) branch switches (always ProCurve 2824 switches) are treated as CBBs and connected sequentially following the Cluster Platform CBB uplinks. In these tables, SVA branch switches are not specifically identified due to the variability in their port placement. | |
| 6. Utility nodes, or nodes within the UBB and UXR, connect immediately below the control node (utility node 1 on port 41, utility node 2 on port 40, and so on). | |
| 7. In many configurations, there will be a gap between utility nodes and compute node branch switches. | |