According to a 2005 Gartner Inc. survey, 60% of users surveyed had consolidation projects underway with an additional 28% looking into it. Cost reduction pressure and the need for improved management were cited as the key drivers for unifying servers.
Implementing server virtualization in conjunction with blade systems not only helps in the consolidation project but also lays a strong foundation for next-generation data centers. Many users are implementing virtualization software, such as VMware Inc.'s ESX Server and its related management products, at the same time as they roll out blade servers.
One user, when asked about the biggest benefits of ESX Server, immediately pointed to the IBM BladeCenter. "If possible, deploy VMware on blade servers," he advised. "A lot of people are afraid of blades because there is a slight learning curve. But the configuration becomes so much more manageable and redundant, and that boosts efficiency even more."
And why is that? There are abundant reasons: Blades not only take up half the space of traditional 1U servers, but they also allow for more simplified, centralized management because they share vital resources like switches and interface cards. In addition, blades save space, improve scalability, reduce complexity and offer better manageability.
Blade servers pack more computing power into a smaller physical footprint. Going a step beyond rack servers, blade systems offer the benefits of both space savings and preconfigured modularity. You can fit 60 blades into the same space required for 42 rack-mounted servers (though numbers vary by vendor). Consolidation (particularly with virtualization) reduces the number of physical servers needed, and blades then reduce the footprint per server.
Flexible options: Scale-out and scale-up applications
Although many think of blades as low-end servers (early blades were based on low-speed chips), blades have evolved to include faster processors and dual-core chips (with multi-core soon to come). Those factors make blades even better candidates for consolidation, especially when virtualizing multiple servers onto a blade. Most blades now come with two CPUs on the blade. There are also four-way blades available with dual-core chips, yielding eight cores (requiring blade slots). Blade server use has moved from low-end, scale-out applications only (e.g., basic Web services) to higher-end, scale-out and even scale-up applications including high-performance clustered computing (HPCC) and applications like Oracle RAC.
Blades reduce complexity by decreasing the number of components to be managed inside the data center. Within the blade chassis, blades share common power and cooling components with built-in redundancy. They also share other common modules in the back plane, including switch modules, KVM modules and sophisticated redundant management modules that allow remote management of all the components in the chassis. In addition to simplification, sharing these components brings down the cost per server.
Another area that blades simplify is cabling. The blade chassis' passive mid-plane includes pre-wired networking connections from each blade server to each switch module, as well as to each management module, KVM switch and any virtual media. Switch modules can be Gig Ethernet or 10GigE, and either two or four Gig Fibre Channel or high-speed InfiniBand. This pre-wiring eliminates the cabling nightmares (not to mention cabling costs) of rack servers. It can be particularly helpful in consolidation projects, especially if multiple servers are being virtualized onto blades, since all blades automatically talk to everything.
Management modules provide access to each blade and shared module in the chassis, down to the bare metal. That allows for full software and hardware management of the blade server system components. For example, if the operating system on a particular blade becomes unstable or inaccessible, the management module provides out-of-band management, which would allow reboots or even powering off the blade.
On the storage front, although blades can be configured with disk storage on each blade, more and more users are implementing diskless blades. Diskless blades, also called anonymous blades, connect to a SAN or NAS for their storage. Again, this is particularly useful as part of consolidation and virtualization in which you don't want to tie a blade to particular storage, but you do want any blade to be able to access the necessary storage for the virtual machines and applications that will run on it. This offers many advantages, both in initially provisioning servers and in re-provisioning or physically swapping out blade servers, and you don't have to worry about access to storage and data.
Anonymous blades, shared networked storage and software like VMware's VirtualCenter (which creates a pool of virtualized servers) work together to offer a flexible, modular infrastructure that yields high availability, serviceability and ease of management. VirtualCenter's single console manages the pool of virtual servers, while systems management software such as IBM Director manages the blade server farm, including complete remote management down to the bare metal. Adding VMotion technology to the mix yields enough management flexibility so that virtual machines can be moved from one blade to another. For example, if a problem starts to heat up on a blade, triggering a thermal indicator, an alert would be sent. The administrator could then move the VMs on that blade to another blade, have someone local physically swap the failing blade, and then remotely bring the new blade online and move the VMs back.
Blade server systems offer significant advantages, particularly when used in combination with virtualization products as part of a consolidation effort. Although most people think of blades for space savings, the modularity, pre-wiring, sharing of common modules and powerful remote management offer great benefits in cost savings, simplification and reduced management costs.
The benefits will grow even greater with upcoming enhancements in blade chips: faster processors, multi-core chips, hardware assists for virtualization, power management and cooling functions, among others. If you're consolidating and virtualizing servers and haven't yet implemented blades, it's time to take a serious look.
About the author: Barb Goldworm has spent over twenty-five in the computer industry in various technical, marketing, sales, senior management and industry analyst positions with IBM, Novell, StorageTek, Enterprise Management Associates and multiple successful startups. She is the founder and president of Focus Consulting, a research, analysis and consulting firm that focuses on storage and systems.