We define some terms that will be used throughout this document.
The CPU architecture relevant in this guide is characterized by the addressing width of the CPU (32 or 64 bit) and the instruction set (x86, Sparc, etc.). If these two characteristics are identical, the CPU architecture is identical for the scope of this guide.
A fabric is an independent, closed communication network that connnects a number of machines (here: all nodes in your cluster). Thus, with one adapter in each node and all cable connections set up, the cluster is using a single fabric. Adding another adapter to each node and connecting them would make the cluster run with 2 fabrics.
A The cable between two adapters or between an adapter and a switch.
The single computer that is running software that monitors and controls the nodes in the cluster. For increased fault tolerance, the frontend should not be part of the Dolphin Express interconnect it controls, although this is possible. Instead, the frontend should communicate with the nodes out-of-band, which means via Ethernet.
The installation script is typically executed on the frontend, but can also be executed on another machine that is neither a node nor the frontend, but has network (ssh) access to all nodes and the frontend. This machine is the installation machine.
The interconnect drivers are kernel modules and thus need to be built for the exact kernel running on the node (otherwise, the kernel will refuse to load them). To build kernel modules on a machine, the kernel-specific include files and kernel configuration have to be installed - these are not installed by default on most distributions. You will need to have one kernel build machine available which has these files installed (contained in the
kernel-devel RPM that matches the installed kernel version) and that runs the exact same kernel version as the nodes. Typically, the kernel build machine is one of the nodes itself, but you can choose to build the kernel modules on any other machine that fulfills the requirements listed above.
All nodes constitute the cluster.
The node manager is a daemon process that is running on each node and provides remote access to the interconnect driver and other node status to the network manager. It reports status and performs actions like configuring the installed adapter or changing the interconnect routing table if necessary.
The service name of the network manager is dis_nodemgr.
The Windows Installer is an engine for the installation, maintenance, and removal of software on modern Microsoft Windows systems. The installation information, and often the files themselves, are packaged in installation packages, loosely relational databases structured as OLE Structured Storage Files and commonly known as "MSI files", from their default file extension.
Scalable Coherent Interface is one of the interconnect implementations that can be used with Dolphin Express software, like SuperSockets and SISCI. SCI is an IEEE standard; the implementation offered by Dolphin are the D33x and D35x series of adapter cards.
This is the latest interconnect from Dolphin and offers increased performance. It requires a dedicated switch to build clusters larger than 2 nodes.
This is a new PCI Express Gen 2 interconnect from Dolphin that will be introduced on the market beginning of 2011. It use a dedicated switch to build clusters larger than 2 nodes.
SISCI (Software Infrastructure for Shared-Memory Cluster Interconnect) is the user-level API to create applications that make direct use of the low level Dolphin Express interconnect shared memory capabilities.
To run SISCI applications, a service named dis_sisci has to be running; it loads the required kernel module and sets up the SISCI devices.
PCI express slots, and also Dolphin Express, combine multiple lanes (serial high-speed communication channels using few electrical connections) into communication paths with a higher bandwidth. With PCI Express Gen. 1, each lane carries 2.5Gbit/s of traffic and with PCI Express Gen 2, each lane carries 5.0 Gbit/s. Combining 4 lanes into a single communication path is called x4 and thus delivers 10Gbit/s Bandwidth for Gen 1 or 20Bbit/s Bandwidth for Gen 2, while x8 doubles this bandwidth using 8 lanes.