As Published In September/October 2003

Part 2: Bringing Computing Power to the Grid

By Michael Miley

The Future of Grid Computing

Grid computing is poised to dramatically lower the cost of computing and to extend the availability and utility of computing resources across departments and organizations. Within the enterprise, the various strategies of sharing and pooling resources will be tried, often with the attendant problems that derive from sharing fixed CPU cycles, as well as those that derive from human reluctance to share limited resources. Between enterprises, security issues will remain a big stumbling block for some time, though Oracle believes that the proper use of its own security technology will provide a strong basis for extending internal grids outside the firewall. And finally, if distributed resources are to be made available on demand, with the response time needed for the most demanding applications, network speeds must be more than just adequate for participating partners.

Dedicated server blade clusters now provide a way for businesses to sidestep many of the barriers to building a grid. The new Oracle Database is up to the task of provisioning resources on an industry-standard grid. Plus, all the operational goals for a grid—such as flexibility, portability, high availability, security, and scalability—are met by an Oracle grid, and system managers have a full arsenal of Oracle tools to manage it.

Michael Miley (mmiley@pacbell.net) is a freelance writer living in Sonoma, California.

Glossary Of Grid-Related Technologies Blades. A computing system that includes processors and memory on a single board, but where other resources such as power, cooling, network access, and storage services are shared among a collection of blades. Blades are easily installed and removed and are smaller than rack-optimized servers. Clustering. Connecting two or more computers together in such a way that they behave like a single computer. Clustering is used for parallel processing, load balancing, and fault tolerance. Clustering is a popular strategy for implementing parallel processing, distributed computing, and grid applications because it enables companies to leverage the investment already made in computers. In addition, it's relatively easy to add new CPUs simply by adding a new PC to the network; or, in the case of blade servers, a new blade or a server composed of blades. Fibre Channel. A serial data transfer architecture developed by a consortium of computer and mass storage device manufacturers, now standardized by ANSI. The most prominent Fibre Channel (FC) standard is Fibre Channel Arbitrated Loop (FC-AL). FC-AL was designed for new mass storage devices and other peripheral devices that require very high bandwidth. Using optical fiber to connect devices, FC-AL supports full-duplex data transfer rates of 100MBps. FC-AL is compatible with, and is expected to eventually replace, SCSI for high-performance storage systems. Gigabit and 10-Gigabit Ethernet. Ethernet is the world's most pervasive networking technology. Gigabit Ethernet is 100 times faster than the original Ethernet (10Mbps) and 10 times faster than Fast Ethernet (100Mbps). It offers 1,000Mbps (1Gbps) raw bandwidth and is compatible with existing Ethernets. Gigabit Ethernet is already being deployed in large numbers in both corporate and public data networks. Meanwhile, an even faster 10-Gigabit Ethernet standard is nearing completion. Grid (grid architecture). A virtualization of an enterprise's computational resources, including servers, networks, storage, and information, into one or more large pools of resources. These resources can be dynamically provisioned on demand to various enterprise applications and users, allowing enterprises to dynamically align their IT resources to their business needs. InfiniBand. Both an I/O architecture and a specification for the transmission of data between processors and I/O devices that has been gradually replacing the PCI bus in high-end servers and PCs. Instead of sending data in parallel, InfiniBand sends data in serial and can carry multiple channels of data at the same time in a multiplexing signal. InfiniBand channels are created by attaching host channel adapters (HCAs) and target channel adapters (TCAs) through InfiniBand switches. HCAs are I/O engines located within a server. TCAs enable remote storage and network connectivity into the InfiniBand interconnect infrastructure, called a fabric. InfiniBand architecture is capable of supporting tens of thousands of nodes in a single subnet. LDAP. Short for Lightweight Directory Access Protocol, LDAP is a set of protocols for accessing information directories. LDAP is based on the standards contained within the X.500 directory standard, but it is significantly simpler. And unlike X.500, LDAP supports TCP/IP, which is necessary for any type of internet access. LDAP makes it possible for almost any application running on virtually any computer platform to obtain directory information, such as e-mail addresses and public keys. NAS. A network-attached storage device is a server that is dedicated to nothing more than file sharing. NAS does not provide any of the activities a server in a server-centric system typically provides, such as e-mail, authentication, or file management. NAS allows more hard-disk storage space to be added to a network that already utilizes servers without shutting them down for maintenance and upgrades. Node. In networks, a node is simply a processing location. A node can be a computer, a set of clustered blades, or some other device, such as a printer. Every node has a unique network address, sometimes called a Data Link Control (DLC) address or a Media Access Control (MAC) address. RAID. An acronym first used in a 1988 paper by Berkeley researchers Patterson, Gibson, and Katz to describe a redundant array of independent disks for providing fault tolerance (redundancy) and improved access rates. RAID provides a method of accessing multiple individual disks as if the array were one larger disk, spreading data access out over these multiple disks, thereby reducing the risk of losing all data if one drive fails, and improving access time. SAN. A storage area network is a high-speed subnetwork of shared storage devices. A storage device is a machine that contains nothing but a disk or disks for storing data. A SAN's architecture works in a way that makes all storage devices available to all servers on a LAN or WAN. As more storage devices are added to a SAN, they too will be accessible from any server in the larger network. SMP. Symmetric multiprocessing is a computer architecture that provides fast performance by making multiple CPUs available to complete individual processes simultaneously (multiprocessing). Unlike with asymmetrical processing, any idle processor can be assigned any task, and additional CPUs can be added to improve performance and handle increased loads. VPN. Short for virtual private network, a VPN is a private network that is constructed by using public wires to connect nodes. There are several systems that enable you to create networks using the internet. These systems use encryption and other security mechanisms to ensure that only authorized users can access the network.

The Role of Linux in the Grid Linux is the OS of choice for an industry-standard grid for several reasons. First, it's cheap, with a variety of vendors providing a range of low-cost supporting services. It's also a multiuser, multitasking networked operating system that's extremely stable and can run for years without a crash. It provides very high performance on lower-cost commodity systems; companies have built Linux clusters out of cheap PCs and blades that rival the computational power of multimillion-dollar supercomputers. And finally, Linux developers are fanatics for unwavering compliance with public and industry standards—which is a critical feature. Any business building an internal grid will need to talk to other external grids in the future. Oracle supports all the components and standards for a standard grid architecture—the blades, the interconnects, and the Linux OS (whether Red Hat or United Linux)—in addition to other possible grid components within heterogeneous settings; for example, Sun 64-CPU SMP servers running Solaris. When using Oracle, the platform is not an issue.