Figure 4-22 shows a typical large Enterprise Data Center infrastructure design. The design follows the Cisco multilayer infrastructure architecture, including core, aggregation, and access layers.
NOTE In the Enterprise Data Center, the distribution layer is known as the aggregation layer.
Figure 4-22 Sample Data Center Infrastructure
Figure 4-22 Sample Data Center Infrastructure
Layer 2 Clustering Blade Chassis Blade Chassis Mainframe Layer 3
and NIC Teaming with Pass-Through with Integrated with OSA Access
*-Data Center Access-
OSA = Open Systems Adapter
The data center infrastructure must provide port density and Layer 2 and Layer 3 connectivity for servers at the access layer, while supporting security services provided by ACLs, firewalls, and intrusion detection systems (IDS) at the data center aggregation layer. It must support Server Farm services, such as content switching, caching, and Secure Sockets Layer (SSL) offloading while integrating with multitier Server Farms, mainframes, and mainframe services (such as TN3270, load balancing, and SSL offloading). Network devices are often deployed in redundant pairs to avoid a single point of failure.
The following sections describe the three layers of the Enterprise Data Center infrastructure.
The Data Center Access layer provides Layer 2, Layer 3, and mainframe connectivity. The design of the Data Center Access layer varies depending on whether Layer 2 or Layer 3 access switches are used; it is typically built with high-performance, low-latency Layer 2 switches, allowing better sharing of service devices across multiple servers and allowing the use of Layer 2 clustering, which requires the servers to be Layer 2-adjacent. With Layer 2 access switches, the default gateway for the servers can be configured at the access or aggregation layer.
Servers can be single- or dual-attached; with dual-attached NICs in the servers, a VLAN or trunk is required between the two redundant access layer switches to support having a single IP address on the two server links to two separate switches. The default gateway is implemented at the access layer.
A mix of both Layer 2 and Layer 3 access switches using one rack unit (1RU) and modular platforms results in a flexible solution and allows application environments to be optimally positioned.
The Data Center Aggregation (distribution) layer aggregates the uplinks from the access layer to the Data Center Core layer and is the critical point for control and application services.
Security and application service devices (such as load-balancing devices, SSL offloading devices, firewalls, and IDS devices) provide Layer 4 through Layer 7 services and are often deployed as a module in the aggregation layer. This highly flexible design takes advantage of economies of scale by lowering the total cost of ownership (TCO) and reducing complexity by reducing the number of components to configure and manage. Service devices deployed at the aggregation layer are shared among all the servers, whereas service devices deployed at the access layer benefit only the servers that are directly attached to the specific access switch.
Although Layer 2 at the aggregation (distribution) layer is tolerated for legacy designs, new designs should have Layer 2 only at the Data Center Access layer. With Layer 2 at the Data Center Aggregation layer, physical loops in the topology would have to be managed by STP; in this case, as for other designs, RPVST+ is a recommended best practice to ensure a logically loop-free topology over the physical topology.
The Data Center Aggregation layer typically provides Layer 3 connectivity from the data center to the core and maintains the connection and session state for redundancy. Depending on the requirements and the design, the boundary between Layer 2 and Layer 3 at the Data Center Aggregation layer can be in the multilayer switches, the firewalls, or the content-switching devices in the aggregation layer. Depending on the data center applications, the aggregation layer might also need to support a large STP processing load.
Implementing a Data Center Core layer is a best practice for large data centers. The following should be taken into consideration when determining whether a core is appropriate:
■ 10-Gigabit Ethernet density: Without a Data Center Core, will there be enough 10-Gigabit Ethernet ports on the Campus Core switch pair to support both the campus Building Distribution layer and the Data Center Aggregation layer?
■ Administrative domains and policies: Separate campus and data center cores help isolate the campus Building Distribution layers from Data Center Aggregation layers for troubleshooting, maintenance, administration, and implementation of policies (using QoS and ACLs).
■ Anticipation of future development: The impact that could result from implementing a separate Data Center Core layer at a later date might make it worthwhile to install it at the beginning.
The data center typically connects to the Campus Core using Layer 3 links. The data center network addresses are summarized into the Campus Core, and the Campus Core injects a default route into the data center network. Key Data Center Core layer characteristics include the following:
■ A distributed forwarding architecture
■ Low-latency switching
■ 10-Gigabit Ethernet scalability
■ Scalable IP multicast support
Density and Scalability of Servers
Some scaling issues in the data center relate to the physical environment.
The most common access layer in enterprises today is based on the modular chassis Cisco Catalyst 6500 or 4500 Series switches. This topology has also proven to be a very scalable method of building Server Farms that provide high-density, high-speed uplinks and redundant power and processors. Although this approach has been very successful, it results in challenges when used in Enterprise Data Center environments. The typical Enterprise Data Center experiences high growth in the sheer number of servers; at the same time, server density has been improved with 1RU and blade server solutions. Three particular challenges result from this trend:
■ Cable bulk: Typically, three to four interfaces are connected on a server. With a higher density of servers per rack, cable routing and management can become quite difficult.
■ Power: The increased density of components in a rack is driving a need for a larger power feed to the rack. Many data centers do not have the power capacity at the server rows to support this increase.
■ Cooling: The number of cables lying under the raised floor and the cable bulk at the cabinet base entry is blocking the airflow required to cool equipment in the racks. At the same time, the servers in the rack require more cooling volume because of their higher density.
These challenges have forced customers to find alternative solutions by spacing cabinets, modifying cable routes, or other means, including not deploying high-density server solutions. Another way that customers seek to solve some of these problems is by using a rack-based switching solution. Using 1RU top-of-rack switches keeps the server interface cables in the cabinet, reducing the amount of cabling in the floor and thus reducing the cabling and cooling issues. Another option is to place Cisco Catalyst 6500 Series switches like bookends near the ends of the row of racks so that there are fewer switches to manage.
Was this article helpful?