Oracle Exalogic Elastic Cloud consists of hardware and software engineered together.
Oracle Exalogic Elastic Cloud is the world‘s first engineered system specifically designed to provide enterprises with a foundation for secure, mission-critical private cloud capable of virtually unlimited scale, unbeatable performance, and previously unimagined management simplicity. Exalogic is the ideal platform for applications of all types, from small-scale departmental applications to the largest and most demanding ERP and mainframe applications. While Exalogic is optimized for enterprise Java, Oracle Fusion Middleware, and Oracle‘s Fusion Applications, it is also an outstanding environment for the thousands of third-party and custom Linux and Solaris applications widely deployed today. Simply put, Exalogic is a giant step forward in realizing Oracle‘s vision for the datacenter of the 21rst century.
Exalogic is an engineered system comprising both hardware and software components, each of which is a strategic technology in Oracle‘s product portfolio.
Oracle Exalogic Elastic Cloud Hardware
Exalogic hardware is pre-assembled and delivered in standard 19‖ 42U rack configurations. Each Exalogic configuration is a unit of elastic cloud capacity balanced for compute-intensive workloads. Each Exalogic configuration contains a number of hot-swappable compute nodes, a clustered, high-performance disk storage subsystem, and a high-bandwidth interconnect fabric comprising the switches needed to connect every individual component within the configuration as well as to externally connect additional Exalogic or Exadata Database Machine racks. In addition, each configuration includes multiple 10 Gigabit Ethernet ports for integration with the datacenter service network and Gigabit Ethernet ports used for integration with the datacenter‘s management network. All Exalogic configurations are fully redundant at every level and are designed with no single point of failure.
Each Exalogic compute node is a fully self-contained unit of compute capacity with either multi-core x86 Xeon or SPARC-T3 processors, redundant power supplies, fast ECC DIMM memory, and redundant InfiniBand Host Channel Adapters. Each compute node also contains two solid-state disks (SSDs), which host the operating system images used to boot the node and act as high-performance local swap space and storage for diagnostic data generated by the system during fault management procedures.
Extreme Scalability with Exalogic
InfiniBand is fundamental to the Exalogic Elastic Cloud system. In addition to providing an extremely fast, high-throughput interconnect between all of the hardware units within a deployment, it also provides extreme scale, application isolation, and elasticity.
There are two traditional approaches to growing a datacenter‘s compute capacity, the most basic approach is vertical scaling. Vertical scaling adds compute capacity to an individual computer, but with major limitations. There are limits to how much a single computer can scale with a balanced configuration. Also, vertical scaling increases the impact of a single system failure. More recently, horizontal scaling has become more common and is accomplished by networking together many individual computers using basic networking technologies like Ethernet. Horizontal scaling increases both compute capacity and the tolerance of individual system failure. However, horizontally expanded systems function as a collection of separate computers that require coordination. This coordination is challenged by throughput, latency and lack of high-end cluster features in basic Ethernet networks.
By contrast, the lossless switched InfiniBand I/O fabric used by the Exalogic system connects all systems together in a way that forms a single large computer.
It is possible to start as small or quarter rack, grow to medium or half a rack or a to a large full rack and, beyond this one can connect as many as eight full racks of Exalogic hardware (or any combination of Exalogic and Exadata configurations) together without the need for any external switches. In cases where more than eight racks of Exalogic or Exadata hardware are required, Oracle offers a choice of several high-capacity datacenter switches which allow the creation of Exalogic clouds comprising hundreds of racks and tens of thousands of processors.
Exalogic systems scale horizontally, meaning that there is no degradation of system performance as the size of the cloud increases. Equally importantly, an Exalogic cloud is a resource pool that can be dynamically sub-divided into secure units of capacity. The underpinnings of this capability are in the design of InfiniBand itself, which was designed for precisely this use. InfiniBand supports partitions, in which communication between end-points on the I/O fabric is strictly controlled within the fabric switches. Individual compute nodes, or even specific I/O devices, may be grouped into logical partitions, within which communication is allowed. Communication between logical partitions, however, can be controlled at the lowest level.
Augmenting the security of InfiniBand partitions is another feature of InfiniBand called virtual lanes. Each I/O end-point communicates over the I/O fabric using one or more of these virtual lanes, each of which is fully independent of the others and may be assigned a priority, thereby ensuring that applications may have guaranteed access to shared resources. In the case, for example, of an extremely mission-critical application, it is possible using these features for an administrator to select a number of compute nodes, assign all of the I/O devices on those nodes to a secure partition, and then assign virtual lanes for exclusive access to shared storage, Exadata Database Machine resources, and external service network ports. More importantly, if capacity requirements change, compute nodes may be added to, or removed from, the desired partition dynamically through simple configuration.