Cost-effective, extremely manageable, high-density rack servers September 2002 Why Blade Servers? by Mark T. Chapman IBM Server Group
Page 2 Introduction to Blade Servers Many organizations have begun consolidating servers into centralized data centers, looking to use physical, application or data consolidation as a means of reducing the challenges and costs associated with administering many small servers scattered across the enterprise. To date, physical consolidation has generally involved replacing bulky tower servers with slender 1U or 2U rack systems. They take less space and put the servers and infrastructure within easy reach of the administrator, rather than spread across a large campus. These servers do enable organizations to reap many benefits of consolidation, yet because each server requires its own infrastructure including cables for power, Ethernet, systems management, power distribution units (PDUs), keyboard/video/mouse (KVM) switches and Fibre Channel switches they offer challenges of their own. A rack of 42 1U servers can have hundreds of cables strung throughout the rack, making it difficult to determine which cables attach where and complicating the addition and removal of servers from the rack. In addition, the PDUs and switches consume valuable rack sidewall space. A blade server is a type of rack-optimized server that eliminates many of these complications, thus providing an effective alternative to 1U and 2U servers. There is a range of blade server designs from ultradense, low-voltage, lesser-performing servers to high-performance, lowerdensity servers to proprietary, customized rack solutions that include some blade features. This paper will describe these designs in more detail, discuss how they compare to other server designs and explain how blade servers can fit into your existing datacenter or IT infrastructure. The term blade server refers to a chassis that can hold a number of hot-swappable devices called blades 1. Blades come in two varieties: server blades and option blades. A server blade is an independent server, containing one or more processors and associated memory, disk storage and network controllers and running its own operating system and applications. Each server blade within a system chassis slides into a blade bay 2 and plugs into a midplane or backplane to share common infrastructure components, such as power supplies, fans, CD-ROM and floppy drives, Ethernet and Fibre Channel switches and system ports. Option blades, which may be shareable by the server blades, provide additional features, such as controllers for external I/O or disk arrays, additional power supplies, etc. (The illustration shows how a blade server might be designed to use modular blades and option modules.) 1 When most people speak of blades they are referring specifically to server blades. To lessen the confusion between the terms blade server, server blade and option blade, this paper will refer to blade server enclosures as chassis, server blades as blades, and option blades as options or specific option modules. The term blade server will be used to mean the entire package of chassis, blades and options. 2 Some blade server designs may refer to blade and option bays as slots.
Page 3 This server architecture is the most efficient solution for adding scale and capacity in a data center. Some blade designs support (in a standard 42U rack) at least twice as many processors as in a rack-optimized 1U design. The advantages of the blade architecture can include floor/rack space savings, reduced power usage and lower heat output, as well as a potentially lower acquisition cost versus individual rack-optimized servers. Blade servers can also help reduce administration costs and improve systems management by consolidating dozens of widely distributed servers into one rack. Most blade server chassis can be added to a standard rack along with traditional rack servers. Typically there is no need for a special rack just for blade server chassis. The purpose of this paper is to introduce you to the blade server concept by describing features and benefits of blade servers in general across the industry. Blade Server Benefits What customers want in a blade server is scalability, versatility, performance, availability, easy serviceability, simple systems management and deployment, as well as cost savings all in one package. The modular blade server architecture offers the flexibility needed to accomplish each of these requirements. The following topics explore the benefits of blade servers in these areas: Modular scalability. Unlike a traditional 8-way or 16-way server, which uses a scale up approach, blade servers are designed to efficiently scale out. Their design incorporates many form factors the most common being 3U, 6U and 7U containing varying numbers of blade bays. Adding a new server generally involves nothing more than sliding a new uni- or multiprocessor blade into an open bay in the chassis, thus eliminating the need to physically install and cable individual servers. A larger chassis that allows for greater expansion through option modules helps to balance performance vs. density to properly leverage the infrastructure for optimum utilization. Without this flexibility, the blade server might have great density with poor performance or great performance with low density and not achieve the desired benefits. Option modules permit the addition inside the chassis of shared infrastructure features that typically are externally attached. In addition, by installing option modules containing Gigabit Ethernet switches, KVM (keyboard/video/mouse) switches and Fibre Channel switches, you can save valuable sidewall space in your racks. Because the blades are powered by the chassis, the usual cascading series of PDUs to power everything is not required. This saves both sidewall space and money. Other types of option modules can contain systems management controllers, power supplies or advanced cooling systems for additional redundancy and efficiency, or controllers for direct-attach SCSI disk expansion units. Modules also may include advanced storage controllers for Fibre Channel, iscsi or InfiniBand. These options (along with Gigabit Ethernet) support Network Attached Storage (NAS) and Storage Area Networks (SANs). NAS and SANs are superior architectures for planning and implementing a scalable storage solution with enhanced manageability, a necessity in today s data center. Versatility. Conventional server designs are limited to only one type of processor per server. Blade architecture imposes no such limitations. Advanced chassis designs with sophisticated cooling and power technologies can support a mix of blades containing different types and speeds of processors today, and hotter, faster processors in the future. This provides investment protection as technology evolves. Each blade is a self-contained server, running its own operating system and software. Therefore, a blade containing one Intel Xeon processor and operating as a file/print server could be nestled among blades containing pairs of Intel Itanium 2 processors doing application serving, database operations and Linux clustering and even blades containing completely different chip architectures. This flexibility allows you to consolidate your workloads in one chassis, without regard to whether an application requires a high-performance 64-bit processor or a 32-bit chip, thus eliminating the need for stand-alone servers to perform each function.
Page 4 Performance. Like 1U rack servers, blade servers use a range of processors from low-cost, low-heat-producing chips at one end of the spectrum, to cutting-edge, high-performance chips, such as the Xeon processor at the opposite end. The low-end processors can save you money up front on the purchase price. Over time, they can save you even more on electric bills, if your workload typically doesn t require a tremendous amount of horsepower for example, edge-of-network (EON) servers for caching, load balancing, firewalls, or print and file serving. For enterprise class workloads, blade designs that use the high-end chips are essential. You can get the same processors found in 1U servers, but with up to twice the rack density and at a potentially lower cost. Blades designed around these processors are appropriate for such applications as collaboration (Lotus Notes, Microsoft Exchange, UNIX /Linux OpenMail), advanced Web serving and Web commerce (IBM WebSphere Application Server, WebSphere Commerce Server, MS Content Manager, MS Internet Information Services, BEA WebLogic), computational nodes (Linux clusters, rendering farms, financial analysis), encryption, and workgroup infrastructure (Citrix MetaFrame terminal serving, Novell or MS file/print serving). High availability and ease of serviceability. Doubling the processor density wouldn t be of much benefit if one failing component could bring down a chassis supporting dozens of processors. Therefore it s critical that high-availability features be included to maximize uptime, simplify failure diagnosis, and reduce servicing time. For this reason, server blade designs include high-availability features similar to those found in conventional rack servers, such as redundant and hot-swap components, and even the hot-swapping of entire blades themselves. Removing a server for maintenance simply involves sliding a blade out of the chassis, much like removing a hot-swap hard disk drive. This makes implementing a policy of hot-spare servers simple and effective. Like other types of servers, blades can be configurable to fail over to one another. Advanced blade servers can be designed to an even higher level of availability so that there is no single point of failure. Memory protection methods more sophisticated than standard ECC, as well as built-in RAID 1 mirroring for operating system failover, can be employed. All critical components can be redundant and/or hot-swappable, including the cooling systems, power supplies, Ethernet controllers and switches, mid- or backplanes, hard disk drives and service processors. If a component failure occurs, advanced diagnostics can lead the servicer directly to the failing part, thus restoring full redundancy sooner. Some components can even alert a systems management processor of impending failure, hours or days before the failure occurs. Systems management and deployment. In some blade server designs, integrated systems management processors can monitor the status of the blades, the chassis and the integrated switches all at once. If any conditions warrant an alert, the processors can signal the systems management software, which in turn can alert the administrator by e-mail or pager at any hour of the day or night. The software may also be able to run system diagnostics and integrate with enterprise-wide systems management software. Advanced management software may even be able to do capacity planning, rack management, software rejuvenation, and other critical functions. The ability to slide server blades in and out of the chassis makes new server deployment more efficient. It is far simpler and faster to insert ten new blades into a server chassis than to mount ten new 1U servers, along with their slides or rails. In addition, inserting a blade in an open bay connects it to all infrastructure components in the chassis. With most blade server designs, there is no need to plug multiple cables into each server as it is installed. For example, instead of one KVM cable, one power cable, one Ethernet cable and one systems management cable per server, you may need only one of each cable per blade server chassis 3. Blade servers excel in the physical deployment area, yet they can be just as effective in supporting the software end of the process: deploying/redeploying and repurposing servers by installing or refreshing operating systems and applications. Advanced management functions 3 Some systems may use more than one cable to support redundant components or paths. This still saves many cables per similar configuration.
Page 5 include concepts such as bay profiling: If you insert a blade into a profiled server blade bay, the system can automatically load the operating system and application image designated for that bay onto the blade and get the server up and running without any human intervention. Likewise, hot spare blades waiting in bays can be repurposed under software control to replace failing blades or to handle peak loads. Cost savings. There are several ways blade servers can save you money versus tower and traditional rack servers: Purchase price The purchase price of individual servers may be lower, due to the reduction in duplicate components needed. You may also be able to save on KVM cabling, because you need only one set of cables per chassis, and you may see a corresponding reduction in the number of KVM switches and power distribution units (PDUs) needed. Likewise, if the chassis includes integrated Ethernet switches many Ethernet cables can be eliminated. In a large data center, blade servers could eliminate the cost of hundreds or even thousands of cables. And because of the increased rack density, you may need fewer racks to house the servers. Power consumption The use of low-power processors in some blades can save money on electricity and cooling costs. However, even blades that don t use lower-power processors can reduce the overall power and cooling requirements of a rack by eliminating many duplicate power-consuming infrastructure components that are included in 1U rack servers. Installation cost With a traditional rack server there is a certain amount of assembly needed to unpack the server, install rails, mount the server into the rack (which may require more than one person due to the size and weight), then connect all the cables and perhaps reconfigure the hierarchy of KVM switches and PDUs needed to tie everything together. With a blade server you simply unpack the blade and slide it into the chassis. If you have a large number of servers to install, this can add up to many hours of time and many resources saved. Reduced floor/rack space Due to the increased density of blade servers you may need only half the expensive floor space that you would for traditional rack-optimized servers. By installing blades containing Gigabit Ethernet switches, KVM switches and Fibre Channel switches, you also can save scarce rack sidewall space. Fewer components to fail With hundreds of servers, each containing multiple fans, power supplies, cables and other components, eventually some will fail. In a sufficiently large data center this may mean one or more failures a day. By reducing the number of components in a rack from hundreds to a handful, the number of points of potential failure are reduced as well. Fewer failures mean less time spent servicing the servers. Right-sized purchases Blade servers eliminate the need to purchase excess processors up front to provide for expansion room. They allow you to buy what you need today, then plug in another blade when your processing needs increase, thus spreading your cost of capital equipment over time. Conclusion Taking into account all of these capabilities, blade servers can be an effective part of your data center. By design, they can offer the benefits of tremendous horizontal scalability in a small space, the versatility to mix and match types of blades within a single chassis, a performance spectrum ranging from low cost to high performance/high availability, with quick and easy serviceability, enhanced manageability and simplified deployment, and significant cost savings both upfront and long-term. Additional Information Go to http://ibm.com/servers/eserver/blades (or call 1-888-SHOPIBM) for information about IBM ^ BladeCenter products and services, including part numbers and prices for servers, racks, storage units and other options.
Page 6 IBM Corporation 2002 IBM Server Group Department U2SA Research Triangle Park NC 27709 Produced in the USA 9-02 All rights reserved IBM, the IBM logo, the e-business logo, BladeCenter and WebSphere are trademarks of IBM Corporation in the United States and/or other countries. InfiniBand is a trademark of InfiniBand Trade Association. Intel and Pentium are registered trademarks and Itanium and Xeon are trademarks of Intel Corporation. Linux is a registered trademark of Linus Torvalds. Lotus and Notes are trademarks or registered trademarks of Lotus Development Corporation and/or IBM Corporation, in the Untied States, other countries or both. Microsoft is a registered trademark of Microsoft Corporation. UNIX is a trademark in the United States and other countries licensed exclusively through The Open Group. Other company, product, and service names may be trademarks or service marks of others. IBM reserves the right to change specifications or other product information without notice. IBM makes no representations or warranties regarding third-party products or services. References in this publication to IBM products or services do not imply that IBM intends to make them available in all countries in which IBM operates. IBM PROVIDES THIS PUBLICATION AS IS WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. Some jurisdictions do not allow disclaimer of express or implied warranties in certain transactions; therefore, this statement may not apply to you. IBM ^ xseries servers are assembled in the U.S., Great Britain, Japan, Australia and Brazil and are composed of U.S. and non-u.s. parts. This publication may contain links to third party sites that are not under the control of or maintained by IBM. Access to any such third party site is at the user's own risk and IBM is not responsible for the accuracy or reliability of any information, data, opinions, advice or statements made on these sites. IBM provides these links merely as a convenience and the inclusion of such links does not imply an endorsement.