Virtual sprawl is a very serious issue. Virtualization is very beneficial in that it has the power to remove physical dependencies from network resources. This can often ease some aspects of network management. However, as more and more virtual resources are added to the network, keeping track of them can become increasingly difficult. Like with other growth problems, you may not be able to stop sprawl, but you can manage and contain it.
VMware and Microsoft, for example, have both stepped up to the plate with centralized management tools. VMware's Virtual Center provides excellent centralized management features for distributed virtual machines and VM hosts. Microsoft's competing product, System Center Virtual Machine Manager, is currently in beta. Naturally, you cannot overlook the IBM or HP tools that I previously mentioned as well. So the problem has been identified, and both the virtualization software and hardware vendors have stepped up to the challenge of easing management as virtualization solutions continue to grow. What about hardware decisions -- should data center managers be considering scale up instead of scale out?
This completely depends on the application. For high performance database service, scaling out and clustering still makes the most sense. For consolidation efforts and virtual machines, than scaling up is a logical choice. What are the risks of virtualization in the data center? What are the rewards?
Probably the greatest risk of virtualization is product maturity. Sure, we've had server virtualization around for decades, going back to the days of IBM mainframes. However the x86 virtualization products are still relatively new. VMware is certainly the most mature of the x86-class products and offers very predictable results in the data center, but most of its competitors are still at a 1.0 product status and don't have the maturity or the track record yet in the data center.
All three of VMware's major competitors, Microsoft Virtual Server, Virtuozzo, and Xen, show plenty of promise, but still have some catching up to do. While all three competitors have produced data center success stories (Virtuozzo is probably the closest to VMware at the moment), they need to secure a growing list of data center clients to provide proof that their products are data center worthy. What is another potential risk?
Another risk is the need for understanding performance latency in the planning and deployment process. Today's virtualization engines will induce some form of I/O latency, which can impact application performance. So over allocating virtual machine resources to a host server is often a big misstep. The initial disk configuration is another important consideration. Serious performance differences exist between VMs mapped directly to physical disks on the host or SAN and VMs that utilize virtual hard disks. This determination should be made early in the planning process.
One of the major rewards of virtualization in the data center is failover support. For some organizations, administrators are faced with managing critical applications that do not support clustering. I have seen this with some of the smaller database applications out there that many sales and service teams have to rely on. Normally in these situations, failover is manual to a staged warm standby server. Server virtualization products such as VMware ESX Server and SWsoft's Virtuozzo support VM failover to another host. So suppose that a non-clusterable application is run inside a virtual machine. Now if the VM's host system fails, the VM can simply relocate to another online host. Now the failover is automatic, as it would be in a typical cluster environment. How will storage virtualization play into data center management?
Many data centers are employing in-band storage virtualization and are also using policy-based storage management as part of their data protection operations. From a day-to-day operations perspective, storage virtualization is great, as it allows systems to see storage resources without necessarily requiring the administrator or operator to be aware of a storage resources physical location. This is especially useful in backup and recovery operations. Recovery is much easier when you don't have to know which tape a backup is on, for example. While to some degree these issues are more significant in planning and initial deployment, keep in mind that as storage resources are added, knowledge of the virtualization layer is significant since it will be involved in mapping storage resources to servers.
With the sheer volume of data in many data centers, storage virtualization is a major reward. Many shops today have found many several terabytes of data to be extremely difficult without virtualization. In short, storage virtualization can save up to dozens of man hours of labor on almost a weekly basis.