RTC Magazine

Blade servers have gained ground over the past few years as an efficient, condensed computing solution in large data center implementations. They are widely used as replacements for pizza box servers and large rackmounts (Figure 1).

The advantages inherent in blade server technology include smaller form-factors, denser computing, expandability, hot-swap capabilities, flexible and fail-safe architecture, reduced downtime, increased redundancy, simplified server management, easier hardware and software integration, and lower heat dissipation and power requirements. These improvements over traditional data center servers provide for a massive increase in deployable resource density and an overall reduction in long-term costs.

Each blade in a chassis is typically a self-contained server. Data center consolidation, advanced communications and remote management of servers required to run 24/7 are just a few of the reasons for large-scale deployments. These deployments are becoming more popular for telecom, telephone and cellular carriers, insurance companies, tax preparers, state and local government agencies and educational institutions, among others.

Although blade servers have a long way to go before they are the standard deployed technology throughout the data center, they are beginning to appear in less traditional implementations such as high-performance computing (HPC).

Newer Blade Technology Enables High-Performance Functionality

Many of the characteristics that enable the use of blades in HPC applications have emerged in recently introduced blade servers. As they become increasingly smaller and more powerful, blades are incorporating technology such as open standard architectures, multicore processors, PCI expansion for multiple I/O functionality, the ability to house multiple OSs, low-power processors and innovative cooling techniques, standard AC electrical connectivity, daisy-chaining and Gigabit Ethernet ports. When these are combined with the growing trend of virtualization, blade functionality increases even more.

The great advantage of an open architecture is that anyone can design add-on products for it. An open architecture also increases the potential for partnerships that enable the integration of design cycle management tools. By integrating the tools engineers already use within the same user interface, open design environments enable greater productivity. Innovative products, open standards and interoperability are fundamental to the mainstream adoption of new services in the HPC world.

The emergence of multicore processors gives blade manufacturers even denser computing options. Chips such as Advanced Micro Devices’ dual-core Opteron processor offer significantly greater computing performance than equivalent single-core devices, yet produce no additional heat and do not require additional power. The consolidated processing strength of multiple cores enables next-generation systems and opens up new possibilities for very high-performance blade servers.

The reduction of bottlenecks among processors and system components in AMD’s Direct Connect architecture, for example, makes possible more efficient use of current system resources as well as tomorrow’s high-end components (Figure 2). In particular, dual-core processing is well suited to large clusters. The processing capability of two cores on one chip makes possible server and workstation consolidation. The efficiency of this technology also opens up a new level of high-performance computing and the flexibility to design more innovative solutions that directly address specific customer needs.

The quest for greater performance in embedded and high-performance applications has pushed systems manufacturers to incorporate PCI slots into blades. PCI expansion slots are being implemented in newer blade servers. Higher-performing PCI-X clock speeds up to 133 MHz pump throughput up into the gigabyte/second range. This speedy bus is ideal for critical cards such as frame grabbers and higher-end control cards. PCI-compatible graphics cards can now be added directly to the server for use in backracking, machine vision and imaging implementations.

PCI-X is viewed by many as a logical way to combine more robust bandwidth and greater cost efficiency in data-intensive environments. Applications range from medical imaging to industrial controls to virtual private networking, as well as communications systems and storage area network products such as clustered servers. A PCI-X solution usually offers 32-bit or 64-bit modes and is backward compatible with existing PCI configurations.

The ability to house multiple OSs, previously a convenience, is quickly becoming a requirement in high-performance applications. Within a single rack the newer blades can each run a different OS, either in single or dual boot mode, or, even more efficiently, two concurrent virtualized OSs. Ideally, a blade would provide interoperability with all major OS offerings, including Windows Server and XP Pro, RedHat, SUSE, Fedora Core Linux, and, in some cases, even Solaris 10. Off-the-shelf software applications could be integrated seamlessly and most custom applications could be installed with little or no reconfiguration.

As blades become smaller and more efficient, low-power processors and innovative cooling techniques are emerging that minimize power and cooling requirements while increasing maximum processor core density. A blade can further reduce wattage with a combination of the proper management of power requirements and the customization of hardware configurations to meet the needs of the application and task. A mix of active and passive cooling techniques allows form-factors to be reduced and blades to be housed in much smaller rooms than required previously.

Single-phase AC power sourcing is a major development in blade server technology. It simplifies electrical efficiencies and reduces data center infrastructure requirements by eliminating the need for hard-wiring and expensive transformer circuitry. Older server technology incorporates specialized three-phase, 208V input connections. The expense of rewiring a server room to accommodate this type of connectivity figures largely into the overall total cost of ownership. Newer blades are often designed to run on a standard 110/220V single-phase wall socket by utilizing advanced load-balancing and integrated hot-swappable power distribution buses that results in built-in redundancy and substantially lower cost.

The ability to daisy-chain several blade servers frees up more space and packs more computing horsepower into the same square footage by stacking several chassis in a rack. Some blade manufacturers are now incorporating built-in keyboard/video/mouse (KVM) switches and built-in inter-chassis daisy-chain ports, which eliminate unnecessary cabling and the need for separate external monitors. This can provide an aggregate interconnect rate of up to 10 Gbytes/s. Connectivity is also becoming more efficient to storage area networks via Fibre Channel and by the use of Gigabit Ethernet for creating virtual LANs.

Virtualization and Blades: The Dense Computing Solution

Virtualization, an old idea in the server community, has recently re-emerged as a solution to the ever growing problem of underutilized physical servers. It partitions a server into several “virtual machines,” each capable of running its own OS and application environment. When blade server technology is combined with virtualization, the result is a massive increase in functionality. For example, VMware’s middleware solution allows a host OS to simultaneously run a guest OS as a virtual layer, with a multitude of OS configuration options. The optimal virtualization implementation is applied to the highest-performing processors. When high-performance capabilities are applied to the smallest available blade servers, virtualization is at its most efficient, enabling space minimization, system efficiency and better application utilization (Figure 3).

HPC Blade Implementations

The decision to employ blade servers for non-traditional applications is based on the same benefits that make them attractive for the data center. Distributed computing, rendering/imaging, number-crunching processes, test and measurement data analysis, content manipulation, server appliances or gateways, and heterogeneous computing using mixed OSs are just a few of the areas expected to utilize blades for HPC.

Computing appliance, or dedicated hardware, applications are demanding a move from general-purpose computing to a model that is both powerful and flexible, which combines the strengths of the multi-purpose model with the appliance concept. Simple, reliable devices are required that facilitate repeated tasks. Used as appliances, newer blades provide both the user interface and the “box.” By utilizing open standard architecture and off-the-shelf OSs developed for appliance applications, blades can be fine-tuned to give the optimal performance of a specific desired service, while non-required services are disabled.

As the technology of instrumentation used for signal capture and analysis becomes more complex, so does the need for computers providing immense processing capabilities and flexibility. High-quality measurement and analysis of incoming data is crucial, and high-resolution displays are required for visually rendering much of that data. Aerial and satellite image analysis, automated mapping, detection of human activity, change detection and perceptual organization are some of the processes that require intensive computational processing.

Although signal analysis software that performs these functions often runs on standard OSs such as Windows XP Pro, these complex calculations require more intensive processing than a standard PC can provide. Blades can process large amounts of data, as well as allow multiple monitor hookups. The same requirements, including imaging/rendering needs, apply to test and measurement data analysis.

The integration blade technology brings is poised to define a new generation of distributed computing. Backracking, or centralization, is increasingly being deployed as the architecture of choice for both small and large desktop applications. Its benefits include the elimination of costs for moves, adds and changes, which increases system uptime; the centralization of support; and built-in security for both data protection and disaster recovery. In a backracked environment, a blade acts as a central point that houses the OS and applications, distributing them as requested. In this configuration, 100 or more computers can be stored in a single, centralized rack. With blades in use as desktop PCs, computing power is ensured while maintenance is consolidated and platforms and systems can be quickly upgraded as new technologies become available.

Industry Adoption

Key vertical industries, including oil and gas, are beginning to utilize blades for these non-traditional applications. The newest blade servers are well suited to seismic data analysis, data manipulation, visual rendering via FireWire interfaces and data storage. Military applications are beginning to utilize blades as well, such as signal detection and analysis, surveillance, data analysis and manipulation, and visual rendering of data. Other industries include test and measurement and life sciences, both of which require intensive, number-crunching processes, as well as analysis, manipulation and rendering of data. A less obvious implementation is in the field of media, where digital video capture, digital content creation, editing and visualization are requiring increasingly greater processing capabilities in a smaller space.

The Emergence of Fully Capable Blade Servers

Many vendors are recognizing the need to address these less common deployments of blade server technologies by adopting some, if not all, of the requirements needed to perform these tasks. For example, NextCom’s NextServer 416 (Figure 4) is a high-availability, extreme performance platform that uses the latest technology to provide a solution for server consolidation, cluster and HPC, along with reliable data storage in a small footprint, 4U reconfigurable server.

The platform’s open standard architecture supports AMD’s dual-core Opteron processor and 64-bit Intel EM64T Xeon clustered or independent blade computing. Its performance and form-factor fit the needs of embedded, distributed computing, high-performance imaging and interconnect applications. The aggregate data rate is up to 10 Gbits/s and the platform supports daisy-chained KVM, remote management and alarming, and blade hot-swappability.

Additional blade options include Fibre Channel, Gigabit Ethernet, hard disk, flash disk, PCI-X I/O expansion and SCSI. External monitors can be hooked up to each blade for remote monitoring and visualization. Multiple networks for different functions and/or redundancy can run simultaneously. On-blade storage and external network-attached storage provide flexibility for partitioning applications and data. Blades can be packaged with 32-bit and 64-bit workstation and enterprise server versions of Windows, Linux and Solaris.

High-performance, open standard computing is becoming more commonplace across an increasing number of technology-rich industries. The recent implementation of this technology in blade servers allows non-traditional applications to maximize efficiency and performance. The use of blade servers for non-traditional, high-performance applications will likely increase, just as blade servers will continue to incorporate even more functionality. Smaller form-factors, earlier adoption of new developments in multicore processor technology, increased flexibility and expandability, and increased power efficiency are likely to emerge as technology and application needs evolve.

NextCom Nashua, NH. (603) 886-3874. [www.nextcomputing.com].