||Add To My Personal Library
November 20, 2009
Vol.31 Issue 28|
Page(s) 38 in print issue
Skillful Server Selection
Find The Right Server For Your Data Center’s Needs
Wading through today’s dense selection of servers to identify the perfect server for your data center’s needs can be a daunting task. A server might seem like a great fit based on its specifications and benchmarks, but is it truly the right match for your applications? Will it play nicely with the rest of your IT environment? Will it accommodate future changes?
This series of tips can help you refine not only the server selection process but also the task of self-evaluation, which ultimately goes a long way toward finding the right server.
Focus On Size & Availability
The days of one server per application are mostly in the past when it comes to noncritical applications, according to Denny Lane, director of product management at Stratus Technologies (www.stratus.com). But because data centers increasingly push their servers to the highest utilization levels through the use of virtualization and other technologies, greater demand is placed on the server’s processing performance. In turn, this places more importance on proper sizing and server availability.
“Practically every new server today is designed to support virtual machines, but be sure,” Lane says. “Most end users are reticent to trust critical applications to virtualization just yet. The ‘one application, one server’ model still rules because of the control and security it affords. With availability, security, and performance looming as major issues with cloud computing, most companies will opt for putting only routine, noncritical business processes out there. Business-critical apps stay behind the firewall, which elevates the importance of the hardware selected to run these apps.”
Choose Dynamic Servers
According to Clint Eschberger, technical director at Egenera (www.egenera.com), being dynamic is priority No. 1 for today’s data centers. Traditional servers are difficult to change to accommodate future needs because they’re configured for a single purpose. However, servers need to have the ability to change along with business needs, so it’s important to select hardware that supports modern technologies.
“Being efficient is important in the current economy. Being able to reduce the footprint by adding modern technology like N+1 HA [high availability] and DR [disaster recovery] at the physical layer can reduce the TCO but also create an environment that is easier to manage,” Eschberger says.
IT shops traditionally have several processes that define their overall management of high availability and disaster recovery between different systems—particularly systems running on bare metal or on a hypervisor, he adds. However, by using servers that provide universal HA and DR, data centers can reduce complexity by eliminating third-party clustering, load balancing, and silo methodology.
Look In The Mirror
Self-assessment is a critical requirement for data centers that attempt to identify the right server for their needs. Andrew Hillier, co-founder and CTO of CiRBA (www.cirba.com), recommends assessing three key areas: the configurations of servers currently in use, along with the requirements and constraints that govern these configurations; the business, process, security, and other “nontechnical” attributes of the servers and what policies and constraints must be adhered to; and the utilization levels, patterns, and trends of the workloads running in the environment (including power consumption).
“These areas will impact what kind of servers will be optimal for a given environment,” Hillier says. “For example, if an organization is divided into many small departments and there is a policy that departments cannot share infrastructure, then purchasing large servers may not be optimal, as the business constraints will prevent these servers from being fully utilized.”
Be Inquisitive In Evaluation
When seeking the perfect server, Eschberger relays several rules of thumb garnered from his experience in evaluating servers for multiple types of environments. The first is determining which vendor will offer the best support.
“Look beyond the speeds and feeds of the servers,” Eschberger says. “Most enterprise-class vendors are going to provide similar offerings at each level. What happens when there is a problem? If something breaks in the server, how can you find out? External monitoring tools can only tell you so much. Once the server is down, it is up to the embedded server management to tell you what is wrong. Most vendors have embedded diagnostics or out-of-band systems to help with this. Make sure you understand how this works with the server you are evaluating.”
He also recommends determining how the management system will affect the data center’s current infrastructure. For example, multivendor server infrastructures are popular, but it remains important during server evaluation to ensure that additional complexity won’t be added to the management model. Additionally, managers should identify any constraints (such as space, power, and cooling) in their data centers, as these can heavily influence server aspects such as blades vs. standard rackmounts, number of CPUs, number of hard drives, and even the number of systems.
Server evaluators should also keep in mind the operating system that will be used on the servers, because Eschberger notes that it’s easy to overlook whether test servers actually support the intended OS. Finally, he advises that evaluators perform research to discover what industry experts and other IT professionals are saying about the servers being evaluated. And don’t rely on only one or two reviews—consider the consensus.
by Christian Perry
Bonus Tips |
Balance efficiency and risk. Although it’s not easy to tell business groups that their applications don’t quite warrant fancy servers, the alternative is maintaining inefficient IT environments where capacity goes un-used, says Andrew Hillier, co-founder and CTO of CiRBA (www.cirba.com). By striking a balance between server efficiency and risk, you can avoid inefficient environments as well as risks of workload contention and availability caused by cramming more workloads onto fewer boxes.
Break and evaluate. A favorite tactic of Clint Eschberger, technical director at Egenera (www.egenera.com), is breaking the servers he’s evaluating. He’s not referring to physically breaking them, but instead seeing what happens when something unexpected happens during operation. “Pull the plug [or] pull out the memory, hard drive, etc. Don’t be shy to see what the server will tell you when it has issues,” he says.
Most Valuable Tip: Don’t Overvalue Benchmarks |
Although benchmarks are crucial for determining a server's performance, don't take them as gospel, says Denny Lane, director of product management at Stratus Technologies (www.stratus.com), who relates a server's performance to a car's gas mileage—actual results may vary. Instead, ask for the names of customers with similar deployments, because they'll tell it like it is, and focus more on determining whether the server will fit the application. "Is it transaction-driven or compute-intensive? Does it require high I/O bandwidth? Can it scale to accommodate intermittent spikes in processing demands? What are its availability and reliability track records? Does it integrate easily into the existing IT environment, and is it easy to use? Is it rackfuls of stripped-down, throwaway servers you’re looking for, or robust, highly configurable, low-latency servers you need? There is a server for every application, but no one server is good for every application,” Lane says.
Tip: Best Return On Investment: Match The Server To The Workload |
When assessing utilization and performance, a server’s ability to perform well for a certain type of workload is critical, says Andrew Hillier, co-founder and CTO of CiRBA (www.cirba.com). For example, architectures that provide many logical cores and hyper-threading are well-suited for highly concurrent applications with many active threads, whereas architectures with fewer cores and higher per-core performance provide better homes for single-threaded applications.
“When virtualization is added to the mix, then further criteria are added, including virtual symmetric multiprocessing limitations, I/O overhead, etc.,” Hillier says. “All of these can have an impact on selecting the right server, and the right answer will often be different for different areas of IT environments.”