Blade servers may revolutionize the data center, but they bring up some new and old storage challenges.
By Alex Gorbansky
Blade servers provide a modular computing architecture aimed at improving operational efficiencies by reducing overall management and physical data-center costs. The first generation of blades was primarily geared toward edge applications and delivered savings through high density and low power consumption. In 2002, the market was limited almost exclusively to edge applications such as Web serving. More recently, server vendors have released second-generation, multi-processor blades and are taking aim at the data center.
To gain widespread adoption as a platform for middleware and database applications, blades will have to overcome several major challenges. These include the ability to meet the more stringent performance requirements of demanding application workloads, as well as development of comprehensive blade management and provisioning tools. We expect blade hardware and software suppliers to effectively address these issues over the next couple of years.
In addition to their inherent value proposition, several market factors will promote the broad use of blade servers in edge, middleware, and database environments. Initiatives around grid, or utility, computing by the major systems vendors and parallel efforts by software vendors to re-architect their applications for horizontal scaling represent key drivers to widespread blade adoption.
The development of grid-enabling virtualization software is another critical technology driver for blade adoption. This type of breakthrough technology will enable applications—previously capable of scaling only vertically—to scale out across a virtual pool of computing resources and benefit from the economies of scale delivered by blades.
Blades servers have the potential to become ubiquitous in data centers and will have a profound impact on storage architectures. As blades and associated technologies drive the virtualization of server processing resources, server-to-storage I/O will become virtualized as well, blurring traditional boundaries between servers, applications, and storage.
How does a blade server differ from standard 1U or 2U rack servers? Blade servers consist of a chassis that houses a set of modular, hot-swappable, and independent servers or blades. Most chassis come in a 3U, 6U, or 7U form factor.
Each blade is essentially an independent server on a motherboard, equipped with one or more processors, associated memory, networking, I/O connectivity, and possibly disk storage, running its own operating system and application software. To meet strict space requirements, blades usually do not include a CD-ROM, diskette, or USB connections.
Individual blades are plugged into the mid-plane or backplane of the chassis to access shared components such as power supplies, fans, cabling, Ethernet, and in some cases, Fibre Channel switches. The resource sharing of blades provides substantial efficiencies in power consumption and cable management compared to standard rack servers.
The first generation of blade servers (late 2001) was designed primarily to meet the needs of dot.coms, telcos, and ISPs looking for cost- and space-efficient platforms for edge or Tier 1 applications (Web hosting, portal services, calendaring, and other single-threaded applications that can scale effectively across a number of single processor machines). The first blades were ultra-dense and low in power consumption, delivering the following benefits compared to traditional servers:
Density—Optimal rack-space utilization continues to be a major focus for most data-center managers. Blades enable IT organizations to utilize racks more efficiently. While deploying 336 1U servers would require a total of eight 42U racks, the equivalent number of blade servers can fit into one 42U rack.
Power consumption—Ultra-dense blades can dramatically reduce the ongoing electrical costs associated with the operation of server resources. Generally, the power consumed by the CPU of a traditional 1U server ranges from 90 to 100 watts. A blade-server processor, on the other hand, consumes 10 to 15W. Additional power savings are achieved through the elimination of redundant components.
Cable management—A shared chassis design enables blades to greatly simplify the pain associated with cable management in a data center. A rackable 1U server will typically have up to five wires: two for networking, one for power, one serial, and one PCI—15 cables per 3U chassis. In the equivalent 3U space, a blade server, due to its common power and networking components, only requires four cables.
Serviceability—Blade servers are designed for hot-swappability. A failed blade can be easily replaced, and the replacement blade brought online without disrupting the normal operations of the remaining blades in the chassis. This represents a major serviceability improvement over traditional server designs where replacing failed components can be a time-consuming and challenging process.
Linear scalability—Unlike large, symmetric multi-processing servers, which scale processing horsepower vertically (or within the box), blades are designed for horizontal scaling, sometimes referred to as "scaling out." Blades enable processors, I/O, and networking resources to be scaled linearly and in a balanced fashion—ideal for edge applications such as Web serving. In addition, the physical installation process of adding more blades to an existing chassis is much simpler than the installation of standard rack servers.
Beyond the first generation
While end users are beginning to adopt blade servers for edge applications, many vendors hope to extend the reach of blade technology deeper into the data center. To this end, most of the major systems vendors have recently released multi-processor, higher-performance blade servers targeted at Tier 2 (application servers) and Tier 3 (database) applications. But do blades have what it takes to become a suitable platform for these more-demanding workloads?
For blades to gain widespread adoption in middleware and database environments vendors must address the following issues:
Performance—While second-generation, high-end blades promise savings in space, power, and cable management costs like their predecessors, they must also deliver on far more stringent performance levels required for Tier 2 and 3 applications. First-generation blades sacrificed performance in favor of density and power savings, using slower processors. Clearly, these types of tradeoffs are not acceptable for data-center workloads. Users will expect a blade-server configuration to deliver performance similar to an equivalent configuration of rack servers. Blade vendors have begun to address these needs by shipping symmetrical multi-processing (SMP) blades equipped with Intel Xeon processors ranging from 2.2 to 3.0GHz and by embedding high-performance, low-latency I/O subsystems directly on the blade.
Cooling—Increases in blade processor speed in dense SMP configurations create serious challenges for system cooling. Strenuous application workloads across a number of blades may cause hot spots within the chassis. In some cases, vendors have added a layer of sheet metal around each blade to dissipate heat better. In deploying dense SMP blade configurations, users should be wary of thermal overhead and demand guidance from vendors on supported configurations and recommended thresholds.
Application architecture—Many middleware and database applications scale vertically, requiring large amounts of memory and cache coherency. These applications will have to be re-designed to leverage more modular processing units. The trend toward commodity computing resources is particularly attractive to enterprise software vendors, since it may increase IT spending on software versus hardware. As more application vendors embrace this model, wider blade-server adoption will follow. Grid-enabling virtualization software will help reduce the challenges associated with deploying vertically scalable applications in a blade environment.
Economics of server acquisition—While many IT shops acquire edge servers in bulk, these users typically purchase mid-tier or high-end servers in much smaller quantities. Today, when comparing hardware acquisition costs, a small blade configuration will be roughly equivalent to a similar configuration of standard rack servers. Until blade hardware costs decrease, users will derive cost savings primarily through management efficiencies such as rapid server provisioning and the ability to easily deploy and update server software configurations. For this reason, attaining cost efficiencies through blades requires deploying a sizable server environment. So would deploying middleware or database applications on blades yield an attractive ROI?
While these applications may not necessarily require a high node count, users who have already deployed or are considering deploying blades for edge applications can leverage the blade infrastructure across multiple workloads. It is the broad use of blade servers across multiple types of applications that will drive maximum operational efficiencies for users. For this reason, when evaluating a blade-server solution, users should look for products that allow seamless mixing of single-processor and SMP blades. On the blade management side, users must search for solutions that provide capabilities to seamlessly re-allocate blade resources between workloads as workload demands change.
We expect steady growth of blade deployments in front-end applications over the next couple of years. In parallel, deployments for middleware and database applications will begin to appear. As hardware and software vendors address issues related to performance, cooling, and horizontal scaling of applications, we expect blade adoption to spread aggressively beyond the edge to middleware and database environments.
Note: For vendor/product profiles, view the entire report ("Blade Servers: A Revolution in Server Architectures and Its Impact on the Data Center") at www.tanejagroup.com. Next month, see Blade management, the role of blade servers in grid computing, and the impact of blades on storage.
Alex Gorbansky is a senior analyst at The Taneja Group consulting firm (www.taneja group.com) in Hopkinton, MA.