Faced with power limitations and limited real estate in the server room, many organizations have no choice but to consolidate physical system resources while their logical resources continue to expand. With server consolidation being such a hot topic, many shops are turning to VARs for help in deciding both what to consolidate, as well as the applications and services that should be used in the consolidated infrastructure.
If navigating to the end of a consolidation project was the same as reaching the destination of a family trip, you would just pull out a map, choose your path and go. Unfortunately, consolidation projects are not that simple. Also, if you are a member of the male species
like me, then you would be compelled by your male DNA to not ask for directions.
Today's server consolidation map contains two primary roads. The first path allows you to consolidate via server virtualization. By taking the second road you consolidate servers via clustering. When evaluating consolidation candidates, the performance and application-specific requirements of the hosts identified as consolidation candidates should factor heavily into your vision for a consolidated network architecture.
For I/O and resource-intensive applications, clustering remains the top server consolidation choice. Server virtualization applications typically induce overall resource latency of 8% to 20%, making the performance sacrifice of consolidation via server virtualization too high a price to pay.
On the flip side, consolidating distributed resources to an active-active or shared data cluster can result in an overall performance improvement. A shared data cluster, for example, can allow you to load balance access to resources as well as offer application failover. While server virtualization applications, such as VMware's Virtual Infrastructure 3, provide failover as well, there will still be the cost of reduced latency for the sake of virtualizing.
Server virtualization, on the other hand, is well suited for your many applications that require a dedicated box and typically use very little system resources. Servers that host network services as well as domain controllers with CPU utilization under 15% are also excellent candidates.
A major benefit of virtualizing proprietary application servers is that, when managed by an application that supports virtual machine failover, applications can remain highly available and tolerant to the loss of a physical host server. For applications that do not support clustering, failure of their host system would result in the application no longer being available. In these instances products that support VM failover such as VMware's Virtual Infrastructure 3 can offer significant benefits in virtualizing.
Table 1 compares the differences between consolidating via clustering or server virtualization.
|Consolidation candidate resource consumption||Low CPU and disk I/O||I/O-intensive applications and services with high CPU overhead|
|Typical server roles||Dedicated application, legacy application and low I/O network applications and services||Database, email, Web and file servers with high overhead and I/O requirements|
|Major benefits||Little to no reliance on specific hardware; VMs can be moved to different hosts as performance or scalability needs dictate and virtual machine failover can allow applications that do not natively support clustering to remain highly available.||Increased application availability and overall performance|
|Primary drawbacks||Virtual hardware abstraction adds 8-15% latency to host I/O.||Most available clustering solutions are designed to support specific enterprise-class applications.|
|Product vendors||VMware, Microsoft, SWsoft, XenSource||PolyServe, Symantec, Microsoft, Novell, Red Hat|
Table 1: Consolidation via clustering vs. server virtualization
For organizations with diverse application requirements, often two roads toward consolidation are needed: clustering enabling consolidation for high performance applications and server virtualization handling the remainder. While server virtualization does induce latency, the fact that virtual system hardware is emulated allows a virtual machine to be relocated between host systems with different hardware configurations.
While most available clustering solutions will allow you to see improved performance and availability, application support is limited. Major database (Oracle, SQL Server) and email (Exchange, Lotus Domino) systems are supported by both OS vendors and some cluster application vendors. OS vendors such as Microsoft, Novell and Red Hat also provide cluster services for file, print, Web and network service applications. While third-party application support may be limited, all of your major network applications and services can be clustered.
Ultimately, your consolidation path should be decided by your customers' specific needs. While some organizations can use server virtualization alone to solve immediate consolidation needs, others have turned to a combination of both server virtualization and clustering. In aligning your consolidation path to the technology that is most suitable for the systems you plan to consolidate, you should be able to confidently drive down the road to consolidation without having to get out a map or having stop and ask for directions.
About the author: Chris Wolf is a Microsoft MVP for Windows Server – File System/Storage, and an MCSE, MCT and CCNA. He's a senior analyst for Burton Group who specializes in the areas of virtualization, high availability, enterprise storage and network infrastructure management. Chris is the author of Virtualization: From the Desktop to the Enterprise(Apress), Troubleshooting Microsoft Technologies (Addison-Wesley), and a contributor to the Windows Server 2003 Deployment Kit (Microsoft Press). Reach him at firstname.lastname@example.org.