|
 |
|
General Information
|
Add To My Personal Library |
July 30, 2010
• Vol.32 Issue 16
Page(s) 36 in print issue
|
Consolidating Servers
Use Virtualization & Other Measures To Stop Server Sprawl
|
As enterprises have become more dependent on information technology—and specifically the client server model—the proliferation of servers has become a common occurrence. The rub is that these servers, even though they fulfill a variety of business purposes, are often underutilized.
Underutilized assets lead to business inefficiencies such as decreased ROIs, excess energy consumption, and misallocation of physical and personnel resources. And, ultimately, these business inefficiencies express themselves in the financial bottom line. Fortunately, server consolidation is a powerful tool that can help administrators get a handle on server sprawl and deliver solid business savings.
 Perform Ample Upfront Planning & Analysis
Just like every other fairly complex business undertaking, performing upfront planning and analysis is the best way to ensure successful data center consolidation, says Jason Dea, product marketing manager at Novell (www.novell.com). A well-conceived consolidation plan, he adds, can help achieve a good ROI for the project and mitigate any associated risks.
It is important to keep in mind, Dea says, that effective server consolidation via virtualization is an ongoing process of refinement and not a one-time event. The goal, he adds, should be successful technology adoption and sustainability. Also, the criticality of applications hosted on a server should drive the process: for example, administrators can be more aggressive when virtualizing noncritical test or development servers.
Perform Virtualization Analysis
When virtualization is used to deliver server consolidation, administrators should perform a “virtualization assessment” to analyze the environment before any action is taken, says Dave Sobel, CEO at Evolve Technologies (www.evolvetech.com). A virtualization assessment should include an inventory of servers and attached peripherals, analysis of server performance characteristics and hardware needs to support virtualized servers, backup and disaster recovery analysis, and an initial virtual resource assignment, he adds.
This analysis enables administrators to zero in on the ideal candidates for virtualization within their environments. Obvious candidates for virtualization include systems with minimal processor utilization, systems with RAM requirements and drive storage needs, systems with redundant or warm-spare servers, limited-use servers, and systems where many partially trusted people need console access.
By the same token, this analysis should also identify candidates that are not well suited for virtualization. Examples of these include systems with constant high processor and/or RAM usage, systems with peripherals (serial, parallel, USB, external SCSI, license key fobs, scanners, bar code readers), systems with exceptionally high network usage (such as Gigabit networking), and systems with specialized hardware requirements, Sobel says.
Once virtualization candidates are identified, administrators must then proceed to implementation. The first step, say Edward Lewis and Jeff Holland, senior technical architects for Systems Alliance (www.systemsalliance.com), is to decide which virtualization hypervisor is best suited for the specific server workloads and management requirements. Once a software hypervisor is selected, they add, the hypervisor’s hardware compatibility list should be consulted to determine supported hardware servers. At that time, Lewis and Holland say, administrators can decide whether they want to use rack-mounted servers or blades.
Next, they add, administrators should utilize a server virtual machine process to determine which utility or new servers get virtualized. The purpose for this is simple: avoiding virtualization sprawl, which defeats the purpose of doing physical server consolidation in the first place. Once these steps are done, administrators can then move on to the actual task of moving physical servers to virtual machines. Hypervisors have features that help with this task.
Use The Right Tool
Simplistic tools that are popular for identifying consolidation candidates do not perform a very deep analysis, says Scott Adams, director of product management at TeamQuest (www.teamquest.com). These tools, he adds, often add up a few key metrics and simply check them against a threshold. Also, he adds, they do not understand the details of resource contention and assume performance degrades linearly as systems become loaded, or they reserve too much capacity to avoid having to account for nonlinear queuing delays.
These simple tools, Adams says, are great for getting a rough idea of potential consolidation candidates; however, when important applications are undergoing consolidation, it makes sense to perform a more sophisticated analysis that accounts for the nonlinear delays that can occur when the inevitable resource bottlenecks happen.
Remember The Old Stuff
In all the discussion about server consolidation, it is easy to forget about the equipment about to be “retired.” Cliffie McKay, director of operations at DMD Systems Recovery (www.dmdsystems.com), says administrators should keep in mind the equipment they are removing whenever they undertake a server consolidation project. This includes understanding the details behind transferring programs and data, creating enough storage space, depreciating equipment from accounting books, recycling the equipment through a legitimate recycler, and destroying any data stored on server hard drives.
For example, McKay adds, whenever programs and data are transferred from consolidated servers, administrators must ensure users still have access to the programs and data during the process so downtime can be avoided.
In terms of timing, McKay says companies should have a retirement plan in place from the time they begin preparing for a server consolidation project. Because some equipment has costs associated with recycling, uninstallation, and data destruction that can impact a server consolidation budget, administrators should receive “worst case” quotes that can be factored into the budget. 
by Bruce Gain
Best Tip: Pay Attention To Networking
Preferably before launching a server consolidation project, administrators should consider the fact that poor application performance can result whenever applications from local branch office servers are consolidated to a centralized data center, says Frederic Hediard, vice president of product marketing at Streamcore (www.streamcore.com). End users, Hediard says, can suffer from latency and network congestion when applications are accessed across the WAN.
The way to alleviate the problem, he adds, is to implement proper bandwidth management and provide priority for critical applications. Full visibility of all network traffic, says Hediard, can help pinpoint root causes of latency and network congestion. |
Easiest To Implement: Use More Server Memory
Uri Cohen, product manager at GigaSpaces (www.gigaspaces.com), says increasing the use of a server’s memory instead of heavily utilizing the disk is one way to consolidate servers. Taking this step, Cohen says, increases application efficiency and throughput, thus allowing them to run on fewer machines. And, he adds, leveraging server RAM for storage eliminates “trips” to slow storage devices, thus maximizing application processing power. |
Bonus Tips
Train the staff. Dan Shauver, manager of systems engineering at Laurus Technologies (www.laurustech.com), says administrators should ensure staff receives training ahead of a consolidation. Even though data center managers are usually experts, training can still provide tips on the new environment that will help in the overall consolidation effort.
Make sure it can be virtualized. If virtualization is part of your consolidation plans, Shauver says administrators should make sure that an application can be virtualized at all. Even though this is becoming less of an issue, he says, there are still applications that don’t work well in a virtualized platform. |
|
|
