It's no secret what's driving the move to virtualization in data centers. The demand for new and expanded software systems is growing, but the geographic and carbon footprint required for scaling underutilized dedicated servers is too costly on many levels.

This issue has led to the maturation of the virtual server, where increased reliability and stability means reduced risk, making virtualization of the data center a viable solution. Virtualization makes it possible to replace physical servers running at 10 percent capacity with fewer, more powerful servers running at 60 percent capacity or more.

Figure 1: A-Network cloud connected to four servers running one process each. B-Network cloud connected to two servers each running many processes.

Legacy virtualization platforms support up to four or eight virtual machine (VM) instances per physical server. New virtualization platforms can support 32 VMs or more, with a virtual switch directing traffic from the network interface to the appropriate VM.

This growth has implications for the performance of the physical and virtual servers and for the network.

When working with a dedicated server, it's a fairly straightforward process to characterize performance and isolate factors that affect it. A test system may be used to emulate realistic users, traffic and network conditions and measure the response times of the application with the test system.

When working with dozens of virtual servers in a single physical server, it's not as straightforward. A connection from the test system to the physical server won't provide the granularity of testing required for meaningful results. The single physical interface handles traffic for many VM instances, making it difficult to isolate and measure the performance of each VM instance or the performance of the virtual switch.

Figure 2: A-Test system connected to four servers running one process each. B-Test system connected to two servers running many processes each, with a virtual switch.

Visibility into the performance of each VM is needed, but how can we get it? By creating the capability to capture results and generate traffic from within the virtual server instance. We need a virtual test system inside the virtual server.

What exactly is a virtual test system?

A virtual server is implemented in a virtual machine, a protected partition in a physical server, but to the user it looks and behaves exactly as if it were a dedicated, physical server. In the same way, a virtual test system is a software-based test system implemented in a virtual machine, but to the network devices under test, it looks and behaves exactly as if it were a hardware test system.

Figure 3: Test box connected to a server running 12 processes. These processes contain virtual test systems.

A virtual test system supports the same capabilities as a physical test system. It generates traffic between any number and any combination of virtual and physical interfaces. It reports results per instance and aggregate results for the system under test. It supports realistic layer-2 through layer-7 traffic on virtual and physical interfaces.

The only way to get the granular test results required to reveal the true performance of a virtual server implementation is through a virtualized test system running in VMs on the server.

Another area of concern in the process of virtualizing the data center is the performance of the virtual switches. Ethernet switches are the work horses of the data center. They're not flashy but they consistently and reliably support line-rate throughput, low latency and low jitter. In fact, they can often be taken for granted. Test engineers don't expect switches to introduce problems during testing.

By contrast, virtual switches haven't yet achieved hardware-level performance and reliability. They use the same CPU resources for switching frames as the applications running on the VMs and therefore under heavy loads may not be as reliable as physical switches.

However, that doesn't mean that virtual switches are not ready for business applications. The key is testing. Knowing in advance the performance limits of the virtual switch implemented on your platform means reducing or eliminating the risk of downtime or performance problems. It's the difference between a confident deployment or the fear of customer service issues.

Beyond performance testing, the virtual switch must support the same protocols and functions as a physical switch. Functional testing is just as important for virtual switches as performance and scalability testing, and should be a part of any data center test methodology.

A network manager can reasonably expect any modern Ethernet switch to support features such as virtual LANs (VLANs), access control lists (ACLs), and Internet group management protocol (IGMP) for forwarding multicast traffic. These protocols, and others, are often included as part of physical switch performance testing and should be included when testing virtual switches.

Finally, there is the impact of virtualization on the network itself. A standard rack holds numerous physical servers and dozens of blade servers. Some data centers house hundreds or thousands of racks. With the number of server instances per rack possibly increasing by an order of magnitude, the increase in network traffic can be significant.

In addition, a VM instance often uses more bandwidth than a physical server, due to the traffic generated by managing a virtual machine from a central location. Also, some platforms move virtual machines between physical servers, enhancing uptime and reliability, but also generating considerable network load.

The implications for the data center infrastructure are clear. Virtualization requires that not only the VM solution be tested, including the virtual switch, but also the transport network and elements of the infrastructure supporting functions such as security, load balancing, and others. Establishing performance limits for the components of the network avoids unpleasant surprises and enables planning for growth as the data center expands.

The bottom line is that virtualization promises many benefits for providers and customers, from better margins to better prices to a greener network. Proper testing verifies that it will deliver on those promises.