* Resource virtualization is great, but requires different management thinking
Data center managers have already discovered that consolidation and virtualization of data center resources can produce significant ROI. Part of that return is directly attributed to increased utilization of the resources.
If you can use a single physical server to run several operating system images and applications, you can postpone purchases of new servers, sometimes to the tune of millions of dollars.
Wouldn’t it be a waste if the savings were consumed by increased management complexity?
Managing a pool of virtual servers is more complex in some ways than managing dedicated servers. With dedicated servers, you can fine-tune the infrastructure to the requirements of the application. You can allocate enough capacity to deal with the highest level of demand, even if that means that most of the time that capacity is unused. But in a virtualized server environment, you only allocate the resources you need, when you need them. That means that your management and provisioning system needs to respond to fluctuations in demand.
A management system therefore must work in a bi-directional way. First, the management system must translate application requirements into a set of provisioning commands to “orchestrate” the resources needed. Based on the level of user demand, the management system must balance the supply of resources to the demand in real-time. Secondly, the management system must also be able to evaluate the impact to applications of changes in the infrastructure. If a physical server goes down, the management system must determine which business services and applications were affected by the failure in order to take corrective measures. In a dedicated-server environment, a server failure affects one application, whereas in a shared pool of servers such a failure could affect several applications and business processes.
That is the crux of the matter. Where there is a direct correlation between applications and the infrastructure (that is, the infrastructure is fine-tuned for a single application), managing the infrastructure elements has a one-to-one correspondence with managing the application. In a shared environment, however, the relationship between the infrastructure and the application is a complex map of dependencies. Rather than taking a bottom-up approach, in a virtualized environment you have to manage top-down, from the business process down to the elements.
Here are some of the considerations for managing virtualized infrastructures:
* You must manage the business service, not the elements. While there is obviously a correlation between the two, it is not a one-to-one mapping.
* Provisioning and orchestration is not enough – you need to be able to evaluate the business impact of element failures and adjust resource allocations accordingly.
* Provisioning needs to encompass more than just the servers – the system must also provision network elements (routing, server load balancing, virtual LANs, etc.), security elements (firewalls, intrusion prevention, network encryption, authentication) and storage elements (SAN, NAS).
* Every part of the management system must be based on policies rather than fixed rules. The policies encode the application or business service requirements and are used to dynamically generate infrastructure dependencies.
Management is the biggest stumbling block to successful data-center consolidation and virtualization initiatives. While many vendors offer management products, their most important features often only apply to systems by the same vendor. Data center managers should carefully evaluate their management platform to see how it can support their virtualization strategies.
