SQL Server 2008 : Physical server design - Server consolidation and virtualization

1. Networking components

Of the four main hardware components, network I/O isn't likely to contribute to SQL Server performance bottlenecks. Tuning disk I/O, memory, and CPU resources is far more likely to yield bigger performance increases. That being said, there are a number of important network-related settings that we need to take into account, such as maximizing switch speed, building fault tolerance into network cards, and manually configuring network card settings.

1.1. Gigabit switches

Almost all servers purchased today come preinstalled with one or more gigabit network connections offering roughly 10 to 100 times the bandwidth of previous generations.

The speed of the network card is only as good as the switch port it's connected to. It's common for SQL Servers with gigabit network cards to be connected to 100Mbps switch ports. Gigabit switches should be used where possible, particularly at the server level between database and application servers.

In smaller networks, it's not uncommon for hubs to be used instead of switches. Hubs broadcast traffic to all network nodes. In contrast, switches intelligently route traffic as required, and should be used to reduce overall network traffic.

1.2. NIC teaming

To increase network bandwidth and provide fault tolerance at the network level, a technique known as NIC teaming can be used. NIC teaming involves two or more physical network interface cards (NICs) used as a single logical NIC. Both cards operate at the same time to increase bandwidth, and if one fails, the other continues operating.

Although NIC teaming is a useful technique from both a performance and a fault tolerance perspective, there are some known limitations and restrictions regarding its use in a clustered environment.

1.3. Manually configuring NIC settings

Finally, most network cards offer an autosense mode that permits self-configuration to match its speed with the connected switch port speed. It's not uncommon for autosense to get it wrong, artificially limiting throughput, so NIC settings (speed and duplex) should be manually configured.

2. Server consolidation and virtualization

The arrival of personal computer networks marked a shift away from "big iron" mainframes and dumb terminals to a decentralized processing model comprised of many (relatively cheap) servers and lots of personal computers. In many ways, we've come full circle, with a shift back to a centralized model of SANs and virtualized servers running on fewer, more powerful servers.

In recent years, the shift toward server consolidation and virtualization has gained pace, and in the years ahead, the idea of a dedicated, nonvirtualized server may seem quite odd. From a SQL Server perspective, this has a number of important ramifications. Before we look at the considerations and differences, let's begin by taking a look at the goals of consolidation and virtualization.

2.1. Goals of consolidation and virtualization

The plummeting cost and increased power of today's server components has moved the real cost of modern computing from hardware to people, processes, power, and space. As such, in a never-ending quest to minimize operating costs, businesses have embraced consolidation and virtualization techniques. The major goals are to avoid server sprawl and minimize costs.

Server sprawl

The term server sprawl is used to describe the uncontrolled growth of servers throughout the enterprise, making administration very difficult. SQL Server sprawl is a particularly nasty problem; consider an example of a SQL Server instance running on a PC sitting under someone's desk, installed by a DBA who is no longer with the organization and who never documented its existence. Does this instance need backing up? Does it contain sensitive data? Do you even know that it exists?

Given the ease with which new SQL Server installations can be deployed, and the power of today's commodity PCs, SQL Server sprawl is an all-too-common problem. While there are tools for discovering the presence of SQL Server instances, and new SQL Server features such as policy-based management that make administration much simpler, the sprawl issue remains a significant problem, and one of the prime reasons for consolidation and virtualization projects.

Operating costs

Each physical server consumes space and power, the natural enemies of a data center. Once the data center is full, expanding or building a new one is a costly exercise, and one that makes little sense if the existing servers are running at 50 percent utilization.

As computing power increases and virtualization software improves, the march toward fewer, more powerful servers hosting multiple virtual servers and/or more database instances is an industry trend that shows no sign of slowing down. The real question for a DBA is often how to consolidate/virtualize rather than whether to consolidate/virtualize. In answering that question, let's look at each technique in turn, beginning with consolidation.

2.2. Consolidation

Although virtualization can be considered a form of consolidation, in this section we'll take consolidation to mean installing multiple SQL Server instances on the one server, or moving multiple databases to the one instance. Figure 1 shows an example of consolidating a number of database instances onto a failover cluster.

Support of NUMA hardware, the ability to install multiple instances and cap each one's memory and CPU, and the introduction of Resource Governor in SQL Server 2008 all contribute to the ability to effectively consolidate a large number of databases and/or database instances on the one server.

Just as creating a new virtual server is simple, installing a new SQL Server instance on an existing server is also easy, as is migrating a database from one instance to another. But just because these tasks are simple doesn't mean you should perform them without thought and planning. Let's take a look at a number of important consolidation considerations for SQL Server.

Figure 1. Among other benefits, SQL Server consolidation centralizes administration and combats server sprawl.
 

Baseline analysis

From a consolidation perspective, having this data at hand helps you make sensible decisions on the placement of instances. For example, consolidating a number of CPU-starved servers on a single CPU core box doesn't make any sense. In contrast, consolidating servers that consume very little resources does make sense. Accurate baseline data is a crucial component in making the right choice as to which servers and/or databases should be consolidated.

When examining typical usage as part of a consolidation process, take care to ensure batch processes are considered. For example, two SQL Server instances may coexist on the one server perfectly well until the end of the month, at which point they both run a large end-of-month batch process, potentially causing each process to exceed the required execution window.

Administrative considerations

Consolidation brings with it an even mix of benefits and challenges. We've just covered the importance of consolidating complementary instances from a performance perspective. Equally important is considering a number of other administration aspects:

• Maintenance windows—How long is each server's maintenance window, and will the combination of their maintenance windows work together on a consolidated server?

• Disk growth—Is there enough disk space (and physical disk isolation) to ensure the database growth and backup requirements can be met?

Tempdb

Apart from the ability to affinitize CPUs and cap memory usage, choosing to install multiple instances has one distinct advantage over placing multiple databases in the one instance: each instance has its own tempdb database. Depending on the databases being consolidated, installing multiple instances allows more than one tempdb database to be available, enabling the placement of databases with heavy tempdb requirements in the appropriate instance.

Let's turn our attention now to virtualization, a specialized form of consolidation.

2.3. Virtualization

Unlike the SQL Server instance/database consolidation techniques we've outlined so far, virtualization occurs at a lower level by enabling a single server to be logically carved up into multiple virtual machines (VMs), or guests. Each VM shares the physical server's hardware resources but is otherwise separate with its own operating system and applications.

Virtualization platforms, also known as hypervisors, are classified as either Type 1 or Type 2. Type 1 hypervisors, commonly referred to as native or bare-metal hypervisors, run directly on top of the server's hardware. VMware's ESX Server and Microsoft's Hyper-V (see figure 2) are recent examples of Type 1 hypervisors.

Figure 2. Hypervisors such as Microsoft's Hyper-V are used to virtualize multiple guest servers. Guests share the host's resources while appearing on the network as normal servers.
 

Type 2 hypervisors such as VMWare Workstation and Microsoft's Virtual Server run within an existing operating system. For example, a laptop running Windows Vista could have VMware Workstation installed in order to host one or more guest VMs running various operating systems, such as Windows Server 2008 or Novell SUSE Linux.

Type 1 hypervisors are typically used in server environments where maximum performance of guest virtual machines is the prime concern. In contrast, Type 2 hypervisors are typically used in development and testing situations, enabling a laptop, for example, to host many different guest operating systems for various purposes.

Let's take a look at some of the pros and cons of virtualization as compared with the consolidation techniques covered above.

Advantages of virtualization

Virtualization offers many unique advantages:

• Resource flexibility—Unlike a dedicated physical server, resources (RAM, CPU, and so forth) in a VM can be easily increased or decreased, with spare capacity coming from or returned to the host server. Further, some virtualization solutions enable VMs to dynamically move to another physical server, thus enabling large numbers of virtual machines to be balanced across a pool of physical servers.

• Guest operating systems—A single physical server can host many different operating systems and/or different versions of the one operating system.

• Ability to convert physical to virtual servers—The major virtualization products include tools to create a VM based on an existing physical machine. One of the many advantages of such tools is the ability to preserve the state of an older legacy server that may not be required anymore. If required, the converted virtual server can be powered on, without having to maintain the physical server and the associated power, cooling, and space requirements while it's not being used.

• Portability and disaster recovery—A VM can be easily copied from one server to another, perhaps in a different physical location. Further, various products are available that specialize in the real-time replication of a VM, including its applications and data, from one location to another, thus enabling enhanced disaster-recovery options.

• Snapshot/rollback capability—A powerful aspect of some virtualization platforms is the ability to snapshot a VM for later rollback purposes. At the machine level, this feature can be considered as backup/restore. It enables you to make changes, safe in the knowledge that you can restore the snapshot if necessary. An example is performing an in-place upgrade of SQL Server 2005 to 2008. Should the upgrade fail, the snapshot can be restored, putting the system back to its pre-upgrade state.

Despite these clear advantages, there are a number of issues for consideration before you decide to virtualize SQL Server environments.

Considerations for virtualizing SQL Server

Arguably the single biggest issue for consideration when virtualizing SQL Server is that of support, particularly for mission-critical production systems. It's not uncommon to hear of organizations with virtualized SQL Server environments having difficulty during support incidents due to the presence of a virtualization platform. A common request during such incidents is to reproduce the problem in a nonvirtualized environment. Such a request is usually unrealistic when dealing with a critical 24/7 production SQL Server environment.

The support issue is commonly cited as the prime reason for avoiding virtualization in production environments (and associated volume/load test environments). Those who take this approach often use virtualization in other less critical environments, such as development and testing.

Other considerations for SQL Server virtualization include the following:

• Scalability—The maximum resource limitations per VM (which varies depending on the virtualization platform and version) may present an issue for high-volume applications that require maximum scalability. In such cases, using physical servers, with scalability limited only by the hardware and operating system, may present a more flexible solution.

• Performance overhead—Depending on the hypervisor, a commonly cited figure in terms of the performance overhead of the virtualization platform is approximately 10 percent.

• Baseline analysis—As with the server consolidation techniques we discussed earlier, consideration should be given to the profiles of the individual virtual servers running on the same machine—for example, placing many CPU-intensive VMs together on a single CPU core host machine.

• Licensing—Licensing can be a tricky and complex area, so I'm not going to discuss the pros and cons of licensing virtualization. But I do recommend that before deciding on a server consolidation technique, understand the licensing implications fully.

• Toolset—One of the things that becomes apparent when troubleshooting a performance problem on a VM is that in order to get the full picture of what's happening on the server, you need access to the virtualization toolset in order to determine the impact from other VMs. Depending on the organization, access to such tools may or may not be granted to a DBA.

Virtualization and consolidation techniques are here to stay; it's vitally important that the pros and cons of each technique are well understood.