This chapter covers the following topics:
Types of server farms and Data Centers
Data Center topologies
Fully redundant Layer 2 and Layer 3 designs
Fully redundant Layer 2 and Layer 3 designs with services
This chapter focuses on three main properties of Data Center architectures: scalability, flexibility, and high availability. Data Centers are rapidly evolving to accommodate higher expectations for growth, consolidation, and security. Although the traditional Layer 2 and Layer 3 designs have not changed drastically over the last few years, stringent demands for uptime and service availability, coupled with new technology and protocols, make the design efforts more challenging and demanding.
Demands for scalability, flexibility, and high availability can be summarized as follows:
ScalabilityThe Data Center must support fast and seamless growth without major disruptions.
FlexibilityThe Data Center must support new services without a major overhaul of its infrastructure.
High availabilityThe Data Center must have no single point of failure and should offer predictable uptime (related to hard failures).
A hard failure is a failure in which the component must be replaced to return to an operational steady state.
Scalability translates into the capability to sustain rapid growth in performance, the number of devices hosted in the Data Center, and the amount and quality of the services offered. Higher performance implies tolerance to very short-term changes in traffic patterns without packet loss and longer-term plans mapping growth trends to the capacity of the Data Center.
Scalability on the number of hosted devices refers to being capable of seamlessly adding more ports for servers, routers, switches, and any other service devices, such as server load balancers, firewalls, IDSs, and SSL offloaders. Higher density also includes slot density because the number of slots ultimately determines the potential growth of the system.
Flexibility translates into designs that accommodate new service offerings without requiring the complete redesign of the architecture or drastic changes outside the normal periods scheduled for maintenance. The approach to flexibility is a modular design in which the characteristics of the modules are known, and the steps to add more modules are simple.
High availability translates into a fully redundant architecture in which all possible hard failures are predictable and deterministic. This implies that each possible component's failure has a predetermined failover and fallback time, and that the worst-case scenario for a failure condition is still within the acceptable failover limits and is within the requirements as measured from an application availability viewpoint. This means that although the time of failure and recovery of a network component should be predictable and known, the more important time involves the user's perception of the time to recover application service.
After a failure, the recovery time could be measured from the perspective of the Layer 2 environment (the spanning tree) or from a Layer 3 perspective (the routed network), yet the application availability ultimately matters to the user. If the failure is such that the user connection times out, then, regardless of the convergence time, the network convergence does not satisfy the application requirements. In a Data Center design, it is important to measure recovery time from the perspectives of both the network and the application to ensure a predictable network recovery time for the user (application service).
Figure 4-1 presents an overview of the Data Center, which, as a facility, includes a number of the building blocks and components of the larger enterprise network architecture.
This books deals primarily with the engineering of application environments and their integration to the remaining enterprise network. Different types of server farms support the application environments, yet this book focuses on understanding, designing, deploying, and maintaining the server farms supporting intranet application environments. The actual engineering of the different server farm typesInternet, extranet, and intranet server farmsdoes not vary much from type to type; however, their integration with the rest of the architecture is different. The design choices that differ for each type of server farm are the result of their main functional purpose. This leads to a specific location for their placement, security considerations, redundancy, scalability, and performance. In addition to the server farm concepts, a brief discussion on the types of server farms further clarifies these points.
The figures in this chapter contain a wide variety of Cisco icons. Refer to the section, "Icons Used in This Book" (just before the "Introduction") for a list of icons and their descriptions.
Figure 4-1 Overview of Data Center Topology
Types of Server Farms and Data Centers
As depicted in Figure 4-1, three distinct types of server farms exist:
All three types reside in a Data Center and often in the same Data Center facility, which generally is referred to as the corporate Data Center or enterprise Data Center. If the sole purpose of the Data Center is to support Internet-facing applications and server farms, the Data Center is referred to as an Internet Data Center.
Server farms are at the heart of the Data Center. In fact, Data Centers are built to support at least one type of server farm. Although different types of server farms share many architectural requirements, their objectives differ. Thus, the particular set of Data Center requirements depends on which type of server farm must be supported. Each type of server farm has a distinct set of infrastructure, security, and management requirements that must be addressed in the design of the server farm. Although each server farm design and its specific topology might be different, the design guidelines apply equally to them all. The following sections introduce server farms.
Internet Server Farms
As their name indicates, Internet server farms face the Internet. This implies that users accessing the server farms primarily are located somewhere on the Internet and use the Internet to reach the server farm. Internet server farms are then available to the Internet community at large and support business-to-consumer services. Typically, internal users also have access to the Internet server farms. The server farm services and their users rely on the use of web interfaces and web browsers, which makes them pervasive on Internet environments.
Two distinct types of Internet server farms exist. The dedicated Internet server farm, shown in Figure 4-2, is built to support large-scale Internet-facing applications that support the core business function. Typically, the core business function is based on an Internet presence or Internet commerce.
In general, dedicated Internet server farms exist to sustain the enterprise's e-business goals. Architecturally, these server farms follow the Data Center architecture introduced in Chapter 1, "Overview of Data Centers," yet the details of each layer and the necessary layers are determined by the application environment requirements. Security and scalability are a major concern in this type of server farm. On one hand, most users accessing the server farm are located on the Internet, thereby introducing higher security risks; on the other hand, the number of likely users is very high, which could easily cause scalability problems.
The Data Center that supports this type of server farm is often referred to as an Internet Data Center (IDC). IDCs are built both by enterprises to support their own e-business infrastructure and by service providers selling hosting services, thus allowing enterprises to collocate the e-business infrastructure in the provider's network.
The next type of Internet server farm, shown in Figure 4-3, is built to support Internet-based applications in addition to Internet access from the enterprise. This means that the infrastructure supporting the server farms also is used to support Internet access from enterprise users. These server farms typically are located in the demilitarized zone (DMZ) because they are part of the enterprise network yet are accessible from the Internet. These server farms are referred to as DMZ server farms, to differentiate them from the dedicated Internet server farms.
Figure 4-2 Dedicated Internet Server Farms
These server farms support services such as e-commerce and are the access door to portals for more generic applications used by both Internet and intranet users. The scalability considerations depend on how large the expected user base is. Security requirements are also very stringent because the security policies are aimed at protecting the server farms from external users while keeping the enterprise's network safe. Note that, under this model, the enterprise network supports the campus, the private WAN, and the intranet server farm.
Notice that Figure 4-3 depicts a small number of servers located on a segment off the firewalls. Depending on the requirements, the small number of servers could become hundreds or thousands, which would change the topology to include a set of Layer 3 switches and as many Layers 2 switches for server connectivity as needed.
Figure 4-3 DMZ Server Farms
Intranet Server Farms
The evolution of the client/server model and the wide adoption of web-based applications on the Internet was the foundation for building intranets. Intranet server farms resemble the Internet server farms in their ease of access, yet they are available only to the enterprise's internal users. As described earlier in this chapter, intranet server farms include most of the enterprise-critical computing resources that support business processes and internal applications. This list of critical resources includes midrange and mainframe systems that support a wide variety of applications. Figure 4-4 illustrates the intranet server farm.
Notice that the intranet server farm module is connected to the core switches that form a portion of the enterprise backbone and provide connectivity between the private WAN and Internet Edge modules. The users accessing the intranet server farm are located in the campus and private WAN. Internet users typically are not permitted access to the intranet; however, internal users using the Internet as transport have access to the intranet using virtual private network (VPN) technology.
Figure 4-4 Intranet Server Farms
The Internet Edge module supports several functions that include the following:
Securing the enterprise network
Controlling Internet access from the intranet
Controlling access to the Internet server farms
The Data Center provides additional security to further protect the data in the intranet server farm. This is accomplished by applying the security policies to the edge of the Data Center as well as to the applicable application tiers when attempting to harden communication between servers on different tiers. The security design applied to each tier depends on the architecture of the applications and the desired security level.
The access requirements of enterprise users dictate the size and architecture of the server farms. The growing number of users, as well as the higher load imposed by rich applications, increases the demand placed on the server farm. This demand forces scalability to become a critical design criterion, along with high availability, security, and management.
Extranet Server Farms
From a functional perspective, extranet server farms sit between Internet and intranet server farms. Extranet server farms continue the trend of using web-based applications, but, unlike Internet- or intranet-based server farms, they are accessed only by a selected group of users that are neither Internet- nor intranet-based. Extranet server farms are mainly available to business partners that are considered external yet trusted users. The main purpose for extranets is to improve business-to-business communication by allowing faster exchange of information in a user-friendly and secure environment. This reduces time to market and the cost of conducting business. The communication between the enterprise and its business partners, traditionally supported by dedicated links, rapidly is being migrated to a VPN infrastructure because of the ease of the setup, lower costs, and the support for concurrent voice, video, and data traffic over an IP network.
As explained previously, the concept of extranet is analogous to the IDC, in that the server farm is at the edge of the enterprise network. Because the purpose of the extranet is to provide server farm services to trusted external end users, there are special security considerations. These security considerations imply that the business partners have access to a subset of the business applications but are restricted from accessing the rest of the enterprise network. Figure 4-5 shows the extranet server farm. Notice that the extranet server farm is accessible to internal users, yet access from the extranet to the intranet is prevented or highly secured. Typically, access from the extranet to the intranet is restricted through the use of firewalls.
Many factors must be considered in the design of the extranet topology, including scalability, availability, and security. Dedicated firewalls and routers in the extranet are the result of a highly secure and scalable network infrastructure for partner connectivity, yet if there are only a small number of partners to deal with, you can leverage the existing Internet Edge infrastructure. Some partners require direct connectivity or dedicated private links, and others expect secure connections through VPN links. The architecture of the server farm does not change whether you are designing Internet or intranet server farms. The design guidelines apply equally to all types of server farms, yet the specifics of the design are dictated by the application environment requirements.
Figure 4-5 Extranet Server Farms
The following section discusses the types of Data Centers briefly mentioned in this section.
Internet Data Center
Internet Data Centers (IDCs) traditionally are built and operated by service providers, yet enterprises whose business model is based on Internet commerce also build and operate IDCs. The architecture of enterprise IDCs is very similar to that of the service provider IDCs, but the requirements for scalability are typically lower because the user base tends to be smaller and there are fewer services compared with those of SP IDCs hosting multiple customers.
In fact, the architecture of the IDC is the same as that presented in Figure 4-2. An interesting consideration of enterprise IDCs is that if the business model calls for it, the facilities used by the Data Center could be collocated in a service provider Data Center, but it remains under the control of the enterprise. This typically is done to lower the costs associated with building the server farm and reducing a product's time to market by avoiding building a Data Center internally from the ground up.
Corporate Data Center
Corporate or enterprise Data Centers support many different functions that enable various business models based on Internet services, intranet services, or both. As a result, support for Internet, intranet, and extranet server farms is not uncommon. This concept was depicted in Figure 4-1, where the Data Center facility supports every type of server farm and also is connected to the rest of the enterprise networkprivate WAN, campus, Internet Edge, and so on. The support of intranet server farms is still the primary target of corporate Data Centers.
Enterprise Data Centers are evolving, and this evolution is partly a result of new trends in application environments, such as the n-tier, web services, and grid computing, but it results mainly because of the criticality of the data held in Data Centers.
The following section discusses the typical topologies used in the architecture of the Data Center.