IT channel takeaway: Your customers may look to virtualization to track servers and address rising energy costs, but you must know where and when virtualization is a good fit so you can ultimately help them avoid virtual sprawl dilemmas.
With Andrew Hillier, co-founder and CTO at CiRBA.
Question: Server sprawl, both physical and virtual, can quickly become an IT department's worst nightmare. However, actually identifying underutilized servers and consolidating them is not as easy as one might think. For those organizations that are considering a consolidation, where do you start and how do you proceed?
Hillier: The best place to start is to take a high-level view of the infrastructure and its contents. Often termed "inventory," this is the determination of what you actually have and, although it sounds basic, it is often quite challenging in many environments. From here, you can start to isolate logical groupings of servers by platform, business function, geography or other classifications. These divisions generally represent the "constraints" on a consolidation initiative and can include technical constraints, such as application and server compatibility, as well as business constraints, such as restrictions on consolidating apps from different locations, owners, availability classes, etc. Once you understand the groupings, you can begin to look at the aggregate utilization of these groupings. This tells you how utilized the environments are, where the greatest impact is to be found and, perhaps most importantly, what the maximum consolidation potential is. Taking this even further, utilization levels are also a form of constraint, and measuring detailed workload patterns and scrutinizing what combinations of workloads will fit on what servers is an important step. And finally, by merging different sets of technical, business and workload constraints, it is possible to identify opportunities in a variety of areas, including database consolidation, app stacking, virtualization and others.
Question: As organizations struggle to keep track of their servers and deal with rising energy costs within the data center, more and more are looking to virtualization as the solution. What are the right things to consider virtualizing and what might not suit that strategy? What are some alternate strategies that maximize consolidation potential?
Hillier: The ability to virtualize is very liberating, but does not suit every data center or every application. Keep in mind that when you virtualize, you place more than one application on the same physical server. This in turn makes all of those applications subject to what happens to that server, and windows designated for change, backup and other activities become interdependent. System availability can also become linked, and combining applications with different service-level targets on the same server may be undesirable. Applications can be constrained by non-technical factors such as compliance or a virtual wall that must be maintained between different business groups, applications or data networks. From a performance perspective, some virtualization technologies are not well suited to hosting applications that rely on I/O performance, such as databases. While they will run on a virtualized server, they may not perform optimally. For this reason, it is not common to see SQL or Oracle databases running in certain virtualized environments. And, lastly, another important thing to consider is that many Unix and Linux applications will happily coexist on a single OS image, sharing system resources without the need for virtualization. The same is not necessarily true in the Windows world, where there is a higher risk of contention between apps and therefore a greater requirement for virtualization. In the Unix and Linux world, many organizations find that OS stacking or the use of containment technologies is often a viable alternative.
Question: As we look ahead to 2007, how serious is the threat of virtual sprawl? And how can a manager avoid it?
Hillier: Virtual sprawl is not a big threat in environments that have only virtualized small sets of systems, but it is in larger production deployments. As soon as there are a larger number of virtual systems being administrated by more than just a single administrator or group, then there is a risk that there will be virtual sprawl. Virtual sprawl is when an organization loses track of what systems have been created (running or otherwise), and happens when there is low accountability, insufficient process control, and the lack of proper documentation of images. This has the effect of mirroring the "open systems sprawl" that has occurred over the last decade, which ironically is one of the main drivers to go to virtualization in the first place. In 2007, the increased deployment of virtual machines will certainly mean that many organizations will start to experience this for the first time, but it is an effect that lags virtualization initiatives by some time.
This 3 Questions originally appeared on IT Business Edge.