J. Enck Research Note 14 May 2003 Commentary Blade and Virtual Servers Can Maximize Efficiencies At first glance, blade servers and virtual servers appear as different as hardware and software, but both technologies have one thing in common. They can be used to maximize server efficiencies. Consolidation takes on many different forms; however, in all cases, one of the major underlying goals is to improve the efficiency of operations. The other major goal, to reduce total cost of ownership, is directly related to efficiency of operations. This efficiency may be measured in square footage, power consumption, heat generation, administration effectiveness, server manageability, server utilization and more. In short, the quest for consolidation is often a goal achieved by meeting efficiency objectives along the way. When it comes to Intel Architecture (IA) servers, enterprises are constantly looking for ways to improve the efficiency of seemingly endless arrays of IA servers. In many cases, enterprises enter into the consolidation fray with the goal of reducing the number of servers they maintain. Although this is a laudable goal, it is extremely difficult given the limited support for mixed workloads and partitioning on IA servers especially those running Microsoft Windows operating systems. Given this barrier, some enterprises turn to other, more productive areas of consolidation such as storage consolidation and others look to increase manageability and administrative effectiveness by introducing third-party software. A third group of enterprises (often enterprises that have already performed storage consolidation and implemented third-party management tools) turns to yet another strategy: increasing the efficiencies of their servers by shaping their servers to match their application workload. Real Hardware vs. Virtual Hardware Two technologies are commonly used to shape servers to application workloads. The first technology is blade servers. The theory is that deploying a series of small servers driven by powerful management software will allow each server to operate at a higher utilization percentage. For example, instead of deploying a Terminal Server solution on a large symmetric multiprocessing (SMP) server, it can be deployed on a blade server "farm" where the processor, memory and input/output (I/O) needs can be more evenly distributed. Gartner 2003 Gartner, Inc. and/or its Affiliates. All Rights Reserved. Reproduction of this publication in any form without prior written permission is forbidden. The information contained herein has been obtained from sources believed to be reliable. Gartner disclaims all warranties as to the accuracy, completeness or adequacy of such information. Gartner shall have no liability for errors, omissions or inadequacies in the information contained herein or for interpretations thereof. The reader assumes sole responsibility for the selection of these materials to achieve its intended results. The opinions expressed herein are subject to change without notice.
The second technology is virtual servers or virtual machines. In this case, a standard IA server is carved up by software into virtual servers, each server running its own copy of the operating system and each operating system hosting its own unique application. Using this technology, multiple applications can indeed be reliably run on one physical server. The caveat is that each application comes with its own operating system and associated overhead. No matter how you look at it, blade servers and virtual servers have the same core mission: run more servers (real or virtual) at higher utilization rates. With this in mind, we have developed eight points of comparison to contrast the similarities and differences between blades and virtual servers. 1. Point of Origin Blade servers are hardware technology. Blade products are offered by all the major server vendors for example, Dell Computer, Hewlett-Packard (HP), IBM and Sun Microsystems as well as small vendors focused on target markets (for example, Egenera and RLX Technologies). Server vendors would like enterprises to believe that blades are just a new form factor that blades are the next evolution of rackoptimized servers; however, we continue to see blades as additive technology and not displacement technology. Simply put, we see more enterprises deploying blades for new applications instead of redeploying existing applications on blades. Virtual-server technology is software-based technology. A controlling program running on a standard IA server creates "virtual" servers, and each virtual server runs its own copy of the operating system. Each operating system running in a virtual server "believes" it is running on a stand-alone server. VMware was the first to market with virtual-server products (ESX Server and GSX Server); however, Microsoft acquired competing technology from Connectix in early 2003. Microsoft will provide its virtual-server technology alongside its operating system. VMware provides its technology in conjunction with hardware partners (such as Dell, HP, IBM and Unisys) as well as some direct sales. 2. Maturity Both of these technologies are young. Blades were invented in 2000 to address the then-growing needs of Internet data centers and have since moved into more "enterprise friendly" solutions. Virtual servers have deep roots they reach to the IBM mainframe Virtual Machine product, to PC emulation products, and finally to virtual-machine implementations for workstations. The manifestation of virtual servers came about in 2001. Blades and virtual servers have some growing up to do. Blades must become friendlier to common storage connections and become more commodity-oriented (in terms of both price and components). Virtual servers must provide better SMP support and gain the support of commercial software application vendors. 3. Manageability Blades and virtual-server technologies introduce new concepts into the world of server management. In terms of blades, the chassis that holds the blades becomes a new point of management in effect, the chassis creates a new entry in the hierarchy of management. Operators and administrators can monitor blades at the chassis level instead of the blade level; if the chassis is fine, then all the blades in it are fine. Blade servers also tend to come with a robust server management stack that allows blades to be provisioned or re-provisioned dynamically. Thus, the applications handled by a set of blades can be changed as the workload needs change. 14 May 2003 2
Virtual servers, on the other hand, present two interesting new concepts. The first concept is introducing a control point for the virtual servers running on the system. Much like the blade server chassis, the control point monitors the individual virtual servers. Furthermore, virtual servers also provide a type of image management the image (operating system and application) of a virtual server can be saved to a file and then used to provision the same virtual-server instance of other instances on other servers. This provisioning capability can drive dynamic resource allocation or it can be used as a software distribution solution. 4. Availability In a virtual-server environment, many virtual servers run in a single physical server. Thus, if the physical server is compromised, some or all of the virtual servers on that server are compromised. For this reason, virtual servers tend to be hosted on servers with redundant hardware features, such as memory mirroring, redundant power supplies, redundant fans and so forth. Also, if a software problem with the virtual-server control software occurs, the entire system, including all virtual guests, may be compromised. For additional protection, failover software (such as Microsoft Cluster Server) can be used to have a second physical server act as a backup. The virtual-server environment allows software problems in an individual guest virtual server to be shielded from the other guests and even makes it possible to set up procedures for fast restart of failed guests. The chassis that holds blades is typically a passive device and, thus, doesn't really pose much of a threat to availability. Therefore, in a blade server environment, the availability granularity is each blade server. Unassigned blade servers can be provisioned quickly to take the place of a failed blade. Also, note that neither virtual servers nor blades protect against operating-system, application or administrative failures, which cause the majority of cases of unplanned server downtime. 5. Scalability Blade servers were originally created to handle stateless workloads that could be scaled horizontally and managed using load-balancing technology. Examples of these workloads include Web serving and terminal server farms. Over time, however, blade servers have grown in SMP capability. In 2003, several vendors will have four-way blades on the market. Thus, the scalability of blades should be roughly comparable to the scalability of rack-optimized servers in the uniprocessor to four-way range. Virtual servers, however, remain somewhat limited in SMP capability. In 2003, the VMware products will only be able to manage up to two processors in a virtual-server image (for example, a four-way machine might run five two-way virtual servers). At present, no SMP road map has been provided for the Microsoft virtual-server technology. With this in mind, virtual-server technology is better-suited for the deployment of dissimilar applications that do not require large SMP hosts. 6. Supportability Blades servers are manufactured and supported by mainstream server vendors, and Microsoft and the major Linux distributors fully support deployment on blades. In terms of local hardware support, each blade is typically handled as a field-replaceable unit manufacturers provide overnight, depot and onsite repair at the blade level. Support for virtual-server technology is more complex. Microsoft does not currently support VMware (either ESX Server or GSX Server); thus, enterprises using VMware products must rely on their server 14 May 2003 3
vendor for support. Microsoft will release its own virtual-server technology by year-end 2003 (0.7 probability), and this technology will be fully supported by Microsoft and the server vendor community. 7. Resource Allocation In a virtual-server environment, the physical resources of the hosting server are shared among all of the virtual-server instances. The way that these resources are shared is configurable virtual servers can be assigned memory limits, they are assigned percentages of CPU capacity, they can have prioritized access to I/O resources, and, in some instances, they can have exclusive access to I/O resources (for example, a dedicated network interface card for a virtual-server instance). Therefore, proper tuning of the configuration is essential to successful deployment. Although some tuning is required, resource sharing is a key capability to drive up utilization and therefore allow consolidated configurations with greater efficiency than the previous unconsolidated servers. In a blade server environment, each blade has its own memory, its own CPU(s) and, typically, its own disks. Network connections can be run to the individual blades, or, more usually, the network connections are routed through the midplane of the chassis and into an internal network switch. External I/O (for example, Fibre Channel connections) is typically handed the same way each blade may have a channel connection, or the connections could be run through the chassis midplane into a Fibre Channel switch. Given that each blade is self-contained, no special tuning is required, but no benefits of resource sharing are available using blade servers. 8. Application Usage Virtual-server technology can host a number of different operating systems. For example, a Windows 2003 host server could run virtual servers running Windows 2003 or other operating systems Windows 2000, Windows NT, Linux, NetWare and so forth. This capability is unique to virtual servers and has some creative uses. For example, use of this technology can enable a Windows NT application running on an old low-performance server to be moved into a new Windows 2000 server without re-qualifying the application under Windows 2000 in this case, the application would continue to run under Windows NT in the virtual-server instance. Of course this is not the only usage model for virtual servers. Virtual servers are often used to host multiple applications (each application in a separate virtual-server instance) to maximize effective utilization of the host server. Here again, blade servers are like rack servers. They run a current operating system (such as Linux, Windows 2000 or Windows 2003) and host applications qualified for that operating system. There are no special hardware or software "hooks" that turn monolithic applications into distributed applications to run on blades. Still to Come As shown in this research, blade servers and virtual servers have a number of similarities and a number of differences. To make matters more interesting, blades will be able to host virtual servers as the SMP ability of blades increases. By year-end 2008, virtual-server technology will be a standard delivery item on blade technology (0.8 probability), allowing enterprises to leverage both technologies to their advantage. Bottom Line: Blade servers and virtual servers can be effective solutions for reducing square footage, improving manageability, increasing administrative effectiveness and sizing servers to meet application 14 May 2003 4
requirements. Enterprises that evaluate one of these technologies should consider the other as well each technology has its own benefits and drawbacks and is better-suited for some workloads than others. However, enterprises should also note that both technologies are still maturing rapidly. Enterprise deployment of blades or virtual servers should be based on tested applications and a two- to three-year return-on-investment period. 14 May 2003 5