As I explained in my column last week, IBM's announcement of BladeCenter is one example of a sea change that is taking place in server design. Across the IT industry, hardware manufacturers are looking for ways to maximize the performance of server components while minimizing their size, power consumption, and cost. Blade servers represent a giant leap forward in this effort, and they epitomize what most servers will look like in several years. Here are the two salient features of this "new look."
- Disaggregation of processor, storage, and network components. In the near future, most new servers won't be preconfigured, standalone units that contain most of the components needed to deliver IT services. Such servers will be replaced by "blades" or "bricks" that contain processing, storage, or network capacity. Companies will create servers from these building blocks by sliding them into chassis in much the same way that we put feature cards into our PCs. Such servers will offer instant, on-the-fly reconfiguration to meet changing needs.
- Clustering and virtualization at the component level. Today, most clustering and virtualization occurs at the level of the standalone server. We cluster multiple servers together for failover purposes or to handle very large processes and databases, and we create virtual servers within partitions of these systems. However, as servers become disaggregated into separate processor, storage, and network components, most clustering and virtualization will occur at the component level. To manage a particular Web application, for instance, a company might cluster five processor blades into a single processor image, connect that image to a storage image made up of four clustered storage blades, then link both images to two clustered network blades. As the application's requirements change, the company will simply reconfigure these virtual images.
While this design revolution will make servers far more flexible, it will also make them much cheaper. Once companies can buy only the processor, storage, or network capacity they need and then manage each of these functions as independent pools, they will dramatically reduce the unused capacity in their IT infrastructures. This reduction, coupled with the sharing of resources such as power supplies and other server components across a larger number of components, will significantly reduce hardware costs.
Do Blade Servers Make Sense for Mid-Market Organizations?
By now, you're probably wondering whether your company's next system should utilize a blade server design such as IBM's BladeCenter. If you're a mid-market company and your primary goal is to spend less money on hardware, your company probably should not take the blade server leap...at least for the next year or two. The reason for this is simple: Unless you're an enterprise that can buy dozens of blades, you probably won't realize adequate savings to justify your investment.
While the power, cooling, and floor space costs for blade servers are lower than those for traditional rack-mounted systems, they frequently cost more on an up-front basis. This may seem odd because in theory, a single blade server should cost less than a handful of traditional rack-mounted servers that have the same number and type of processors. After all, each rack-mounted server has its own chassis, power supply, media bays, and other components, while the blade server shares this hardware across multiple blades.
In practice, however, today's blade servers often cost more than an equivalent number of rack-mounted servers. This is because blade server vendors have made considerable investments in designing their own proprietary chassis, blades, and other components. By contrast, traditional rack-mounted servers use cheaper industry-standard components.
As a case in point, let's tally up the price for a BladeCenter that we can compare to a rack-mounted Intel server with similar attributes. Based on IBM's list prices, a BladeCenter that is fully populated with 14 blades--each with a single Intel Xeon processor running at 2.0 GHz, 512 MB of memory, and a 40 GB IDE disk drive--costs $34,280. That equates to $2,448 per single-processor blade. By comparison, IBM offers its xSeries Model 335, a 1U rack-mounted server that, like BladeCenter, offers 1 or 2 Xeon processors, 512 MB of memory, and a 40 GB IDE disk drive. As a uniprocessor, the x335 carries a list price of $2,219--more than $200 less per CPU than a BladeCenter. The Model 335 also offers several things you can't get on BladeCenter, including PCI slots and expandability to 80 GB IDE drives.
To be fair, BladeCenter can pack twice the processing performance into the same space as the Model 335, and it consumes far less power and cooling capacity as well. As a result, BladeCenter will offer a lower overall cost of ownership for enterprises that support hundreds of Intel servers in a space-constrained environment. However, most mid-market organizations support less than 100 Intel servers and are constrained not so much by floor space or power consumption as they are by up-front costs. That makes BladeCenter and most other blade servers hard to justify for smaller organizations.
While blade servers may not make immediate sense for your organization, there's a good chance that they will do so in the future. Though today's products rely on proprietary designs, movements are underway within the industry to establish standards for blade server chassis, bus architectures, interfaces, and other components. In one to two years, these standards could allow third-party vendors to create a host of cheaper blades that will be interchangeable between the chassis of different vendors. At the same time, blade server management applications will emerge and become integrated with existing systems management suites. These developments will help to create a new generation of industry-standard blade servers that cost far less to own and operate than today's models.
In short, today's blade servers probably won't make financial sense for your company unless you need dozens of them. If you do have such a need, you should realize that in two to three years, the next generation of industry-standard chassis may not support the blades you buy today. As such, you should look for a positive return on your investment within that time frame. If you cannot meet this tough requirement, I would advise you to put your blade server plans on the shelf. However, be ready in a year or two to dust those plans off, because the server market is about to become a very different place.