How to improve elasticity of virtualized data centers: Network Architecture

As organizations become increasingly dependent on information systems, people are looking for ways to make information systems more flexible. Data Center administrators have made many attempts, this includes high-availability solutions that have been used for many years, but the entire industry has begun to transition to virtual data centers, forcing organizations to rethink their high-availability policies.

During the transition to a virtual data center, many interesting phenomena have emerged. For example, small companies that have never used cluster policies are also forced to deploy some form of high-availability architecture, that is because small companies believe that they can achieve resource integration through virtualization to reduce costs.

Although servers are integrated through virtualization, risks are also integrated. For example, enterprises may think that an infrastructure server is faulty and will not have disastrous consequences, however, the failure of servers hosting three or four virtual infrastructures is another matter.

The problem is that if an organization does not have a high availability policy, even if a server component fails, it may cause serious service interruption and may affect many virtual servers, this is why elasticity is important even for small virtual data centers.

Redundancy vs Separation

There are two main ways to improve the elasticity of the virtual data center. The first is to use redundant hardware, and the second is to achieve elasticity through separation. In reality, the two are usually used in combination.

The redundancy of a virtual data center is similar to the traditional high availability method. For example, an organization may choose to use servers with redundant components, such as redundant power supplies and redundant network adapters, or deploy a cluster. Although there is no error in using redundancy, additional hardware is always costly, Server clusters are also complicated, expensive, and difficult to deploy and maintain.

Separation is to split all components of the virtual data center into slices, and then use management software to allocate these slices. Note that the degree of separation is high or low, some virtual data centers only separate hardware and use physical servers as a resource pool. However, some virtual data centers separate application systems.

High Availability achieved through separation cannot work without redundant hardware. Unlike traditional hardware redundancy, High Availability Based on separation usually uses low-cost commercial hardware, this is just the opposite of traditional cluster solutions.

Network Architecture

When talking about the network architecture, it is well planned to achieve redundancy in the virtual data center. Many virtualization-related websites suggest installing as many NICs on the server as possible, and then for each Virtual Machine (VM) assign a NIC so that you don't have to worry about a virtual opportunity to compete for the server network bandwidth with other virtual machines.

Although this design optimizes network performance, it does not help with high availability. Because our goal is to achieve redundancy, each host server still needs multiple NICs, but these NICs are used differently.

Ideally, each host server should have at least four NICs. You can consider using the Dell PowerEdge R900 server.

Although it is recommended that the server have four network interfaces, the current server generally only has two ethernet ports. Although most servers provide expansion slots to add NICs, however, these expansion slots should be used for other purposes, such as optical fiber cards.

In addition to one Nic for each Virtual Machine (dedicated for dedicated cards), it is better to connect each Nic to an independent switch. Two switches should be connected to the primary network, the other two switches are connected to the backbone network between the host servers. In this case, when a switch, network card, or a line fails, each host server can still connect to the backbone network and the primary network.

If your host server runs VMware, you also need to configure a vSwitch for each server, which is used to transmit data between the physical network and the virtual network. Figure 1 shows how the four NICs are used on the VMware ESX Server. The physical network and the backbone network paths must be redundant.