1. THE SHIFT TO SERVER VIRTUALIZATION

Of all the innovations in server technology over the last 10 years, in my view virtualization has had the biggest impact, and made the biggest improvements, to server computing. Although 64-bit architectures, multi-core processors, and solid-state drives have revolutionized their niches of the industry, only virtualization has fundamentally changed the way we can choose to deploy, manage, and protect server workloads.

Today, it’s likely then that the IT environments you use have virtualized servers in them. While a few years ago these servers might have run the smaller workloads such as domain controllers and print servers, today the capability of virtualization technology means you are more likely to also find mission critical servers with high workloads, such as database servers, being virtualized.

Finally, we’ll consider how you can deploy SQL Server 2012 successfully in a virtual environment and monitor it post go-live.

2. AN OVERVIEW OF VIRTUALIZATION

A typical textbook definition of virtualization defines the concept of sharing a single physical resource between multiple isolated processes, by presenting each with their own virtual version of the physical resource. For example, several virtualized instances of Windows can run concurrently on a single physical server, each believing they have exclusive access to the server’s hardware. One of the many benefits of doing this is to increase the physical server’s overall utilization, therefore increasing the value the physical server delivers.

A simple real-world example of deploying virtualization is to have a single physical server hosting four virtual servers.

Let’s assume that the physical server has four CPU cores, 16GB of memory, and the necessary virtualization software to run virtual servers installed on it.

In our example, four virtual servers can then be created by the virtualization software and each configured to have four virtual CPUs and 3GB of memory.

By default, none of the virtual servers are aware of each other, let alone that they are sharing the physical server’s hardware between them — nor would they know in our example that each physical CPU core has potentially been allocated twice (8 physical cores but 16 virtual CPUs allocated).

When the four virtual servers are running concurrently, the virtualization software manages access to the physical server’s resources on an “as and when needed” basis.

In a well-configured environment, we could expect the person who configured the virtual servers to know that no more than two of them would ever need to use all of their CPU resources at any one time. Therefore, the physical host should always be able to satisfy requests by the virtual servers to use all of their allocated CPU resources without having to introduce any significant scheduling overhead.

In a badly configured environment, there might be a need for three virtual servers to use all of their allocated CPU resources at the same time. It’s when this happens that performance could begin to degrade for each of the virtual servers, as the virtualization software has to start scheduling access to the physical server’s resources; a quart has to be made out of a pint pot!

However, as you can probably already see, if the virtual server workloads in this example were correctly sized and their workloads managed, then a significant amount of data center space, power, cooling, server hardware, CPUs, and memory can be saved by deploying one rather than four physical servers.

This “deploy only what you actually need” approach provided by virtualization explains why the technology moved so quickly from being deployed in the development lab to enterprise data centers. In fact, other than smartphone technology, it’s hard to find another technological innovation in recent years that has been adopted so widely and rapidly as virtualization has.

This rapid adoption is highly justifiable; virtualization brought IT departments an efficient data center with levels of flexibility, manageability, and cost reduction that they desperately needed, especially during the server boom of the mid-2000s and then the recession of the late 2000s. Moreover, once virtualization is deployed and the benefits of replacing old servers with fewer new servers are realized, the technology then goes on to deliver more infrastructure functionality — and interestingly, functionality that wasn’t available with traditional physical servers.

Indeed, it’s rare to find a SQL Server environment now which doesn’t use virtualization technologies in some way. In larger environments, companies might only be deploying it on developer workstations or in the pre-production environment; but increasingly I am finding small, mid-size, and even large infrastructures that are hosting their entire production environment in a virtualized manner.

History of Virtualization

The concepts of the virtualization technology that people are deploying today are nothing new, and you can actually trace them back to IBM’s mainframe hardware from the 1960s! At the time, mainframe hardware was very expensive, and customers wanted every piece of hardware they bought to be working at its highest capacity all of the time in order to justify its huge cost. The architecture IBM used partitioned a physical mainframe into several smaller logical mainframes that could each run an application seemingly concurrently. The cost saving came from each logical mainframe only ever needing to use a portion of the mainframe’s total capacity. While hardware costs would not have decreased, utilization did, and therefore value increased, pleasing the finance director.

During the 1980s and 1990s, PC-based systems gained in popularity; and as they were considerably cheaper than mainframes and minicomputers, the use of virtualization disappeared from the technology stack for a while. However, in the late 1990s, VMware, a virtualization software vendor, developed an x86-based virtualization solution that enabled a single PC to run several operating system environments installed on it concurrently. I remember the first time I saw this running and was completely baffled! A backup engineer had a laptop running both Windows and Linux on it; from within Windows you could watch the virtual server boot with its own BIOS and then start up another operating system. At the time, very few people knew much about the Linux operating system, especially me, so the idea of running it on a Windows laptop looked even more surreal!

This example was a typical use of VMware’s original software in the late 1990s and early 2000s, and for a few years, this was how their small but growing customer base used their technology. It was only a few years later that a version of their virtualization software hosted on its own Linux-based operating system was released and data center hosted server-based virtualization solutions began appearing.

Fundamentally, this server-based virtualization software is the basis of the platform virtualization solutions we use today in the biggest and smallest server environments.

The Breadth of Virtualization

When we talk about virtualization today, it is mostly in terms of physical servers, virtual servers, and the virtualization software known as a hypervisor. However, your data center has probably had virtualization in it in some form for a long time, for the reasons we mentioned earlier — to help increase the utilization of expensive and typically underused physical hardware assets.

Today, most Storage Area Network hardware, SANs, use virtualization internally to abstract the storage partitions they present a server with from their physical components, such as the different speed hard drives it might use internally for storing data on.

While a system administrator will see an amount of usable storage on a storage partition the SAN creates for them, the exact configuration of the physical disks that store the data are hidden, or abstracted, from them by a virtualization layer within the SAN.

This can be a benefit for system administrators, allowing them to quickly deploy new storage while the SAN takes care of the underlying technical settings. For example, modern SANs will choose to store the most regularly used data on fast disks and the less frequently used data on slower disks. Yet, the data accessed most frequently might change over time, but by using virtualization, the SAN can re-distribute the data based on historic usage patterns to optimize its performance without the system administrator knowing.

Of course, this may not always be appropriate, a DBA might ask to use storage with consistent performance metrics; but like all virtualization technologies, once the product’s options and limitations are known, an optimized configuration can be used.

Cisco and other network vendors also use virtualization in their network hardware. You may wonder how a collection of network cables and switches could benefit from virtualization, but the concept of virtual LANS (VLANs) enables multiple logical networks to be transmitted over a common set of cables, NICs and switches, removing the potential for duplicated network hardware.

Finally, believe it or not, SQL Server still uses memory virtualization concepts that date back to the Windows 3.1 era! Windows 3.1 introduced the concept of virtual memory and the virtual address spaces, it is still core to the Windows memory management architecture that SQL Server uses today. By presenting each Windows application with its own virtual memory address space, Windows (rather than the application) manages the actual assignment of physical memory to applications. This is still a type of virtualization where multiple isolated processes concurrently access a shared physical resource to increase its overall utilization.

Platform Virtualization

Having looked at the background of virtualization and some of the reasons to use it, this section clarifies what the term platform virtualization means.

Platform virtualization is a type of hardware virtualization whereby a single physical server can concurrently run multiple virtual servers, each with its own independent operating system, application environment and IP address, applications, and so on.

Each virtual server believes and appears to be running on a traditional physical server, with full access to all of the CPU, memory, and storage resources allocated to it by the system administrator. More importantly, in order for virtualization technology to work, the virtual server’s operating system software can use the same hardware registers and calls, and memory address space, which it would use if it were running on a dedicated physical server. This allows software to run on a virtual, rather than physical, server without being recompiled for a different type of hardware architecture.

Cloud Computing

It’s almost impossible to read technology news these days without seeing references to cloud computing, and more commonly private clouds and public clouds. One of the advantages of cloud computing is that new servers can be deployed very quickly, literally in just minutes, and to do this they use platform virtualization.

Private Clouds

In summary, private clouds are usually a large and centrally managed virtualization environment deployed on-premise, typically in your data center. The virtualization management software they use often has management features added that allow end users to provision their own new servers through web portals, and for the dynamic allocation of resources between virtual servers. A key benefit for businesses too is the ability to deploy usage-based charging models that allow individual business departments or users to be charged for their actual usage of a virtual server, as well as allowing more self-service administration of server infrastructures.

Public Clouds

Public clouds, more often referred to as just cloud computing, are very similar to private clouds but are hosted in an Internet connected data center that is owned and managed by a service provider rather than an internal IT department. They allow users from anywhere in the world to deploy servers or services, through non-technical interfaces such as a web portal, with no regard for the underlying physical hardware needed to provide them. Microsoft’s Windows Azure service is an example of a cloud computing service.