Cisco UCS Components
As discussed in previous chapters, data centers are built using different components that represent the physical infrastructure, the layers of operating systems and virtualization, and the different software and applications that provide the needed services, as well as the components used to build and manage the needed level of security and to monitor and respond to issues in the environment at different levels, in addition to the procedures in place to define the smooth operation and disaster recovery. In other words, a data center has multiple building blocks. In the previous chapters, the focus was on the major infrastructure components for building the base of a data center (that is, the network, the storage, and the compute components). Figure 16-1 illustrates the Cisco UCS and HyperFlex, which comprise the Cisco compute component of the data center.

Figure 16-1 Cisco UCS and HyperFlex
Cisco Systems has been in the network and storage business for a very long time, and somewhere around 2008 it entered the server market with the Cisco Unified Computing System (UCS). It was and still remains a revolutionary, next-generation, highly integrated platform for providing the needed computing resources, together with their communication and storage needs, in a flexible way that allows for the optimal utilization of the hardware computing resources through hardware abstraction.
You will investigate how exactly the hardware abstraction happens and what the benefits are in the next chapter, where Cisco UCS management is discussed. However, to be able to understand the Cisco UCS better, first you will learn about the different physical components of the Cisco UCS and how they connect to each other. You will also learn about the whole physical topology of the Cisco UCS as well as the standalone Cisco UCS servers and their benefits and applications.
Based on the Cisco UCS, Cisco Systems has developed Cisco HyperFlex. It is a hyper-converged application-centric solution that integrates not only the hardware and the management of the Cisco UCS but goes further and creates the needed environment for the virtualization supporting containers. It also includes the needed storage and communication.
Before we can dig deeper into the Cisco UCS hardware abstraction or Cisco HyperFlex, you will need to learn about the physical components that make up the Cisco UCS, as illustrated in Figure 16-2.

Figure 16-2 Cisco UCS Components
The Cisco UCS consists of the following components:
- Cisco UCS Manager (UCSM): This application manages and monitors all the components of the Cisco UCS. In the last generation of the Cisco UCS, the UCS-X, the Cisco UCS Manager is replaced by the Cisco Intersight Managed Mode (IMM).
- Cisco UCS Fabric Interconnects (FIs): The communication devices responsible for the physical connectivity of the Cisco UCS to the LAN and SAN infrastructures. An additional important function is that the Cisco UCS Manager runs on them.
- Cisco UCS 5108 Blade Chassis: The chassis that houses the servers with a blade physical form factor and the I/O modules (IOMs). It provides the connectivity, power supply, and cooling for the blade servers. The new chassis (the UCS X9508), the new blade servers, and their management are discussed in a separate topic.
- CISCO I/O Modules (IOMs): These are the connectivity modules, also known as Fabric Extenders (FEX). They are installed in the blade chassis and provide for the physical connectivity of the servers and the chassis up to the Fabric Interconnects.
- Cisco UCS B-series servers: The Cisco B-series servers are servers that have a smaller physical form factor. This is the blade form factor, and that’s why they are usually called blade servers. The Cisco B-series servers are installed in the Cisco UCS blade chassis.
- Cisco UCS C-series servers: The Cisco C-series servers are designed as rack-mount servers. They can be either a component of the Cisco UCS, under the management of the Cisco UCSM, or they can be used in the standalone mode, where they are managed by their own Cisco Integrated Management Controller (CIMC).
- Cisco virtual interface cards (VICs): The Cisco VIC is a purpose-built converged network adapter to be used in the Cisco UCS servers. It allows for Ethernet, FC, or FCoE communication but also supports the Cisco VM-FEX technology, which allows for the network connectivity awareness and management up to the level of the virtual machines.
These are the major components that build a Cisco Unified Computing System. The whole system appears as one component in front of the rest of the data center components.
All Cisco UCS blade servers come with Cisco UCS Manager capability. Cisco UCS with Cisco UCS Manager provides the following:
- Embedded integration of LAN, SAN, and management
- Auto-discovery, with automatic recognition and configuration of blades
- Local (and optionally global) server profiles and templates for policy-driven server provisioning
The Cisco UCS components are connected and managed in a specific way, as the resulting integrated system is known as a Cisco UCS domain (see Figure 16-3). The Cisco UCS domain consists of two Cisco UCS Fabric Interconnects, which fulfill two roles:

Figure 16-3 Cisco UCS Logical View
- Communication: All the servers’ communication to and from the data center goes through them. Upstream, facing the data center LAN and SAN infrastructures, the Cisco UCS Fabric Interconnects provide the connectivity for both the Ethernet data communication and the storage communication. The storage communication can be based on block or file storage protocols; as for the block-based communication, the supported options cover the FC, FCoE, and the iSCSI protocols. The Fabric Interconnects connect upstream to the LAN and SAN infrastructures. And, downstream, the FIs are responsible for the connectivity to the servers. Inside the Cisco UCS, all the communication is based on Ethernet. There is only Ethernet physical connectivity from the Fabric Interconnects down to the servers. That’s why the servers are equipped with converged network adapters, which externally use Ethernet connectivity but also support FCoE communication between them and the Fabric Interconnects.
- Management: The Cisco UCS Manager application, which takes care of the discovery, provisioning, configuration, and management of the Cisco UCS servers, runs on the Fabric Interconnects. Additionally, the Fabric Interconnects provide a separate out-of-band dedicated interface to connect to the management network in your data center.
In a Cisco UCS domain, the two Fabric Interconnects are connected to each other with a special link, called the cluster link. By bringing in the cluster link, you create a cluster; the link is used for configuration synchronization and the exchange of state information between the two Fabric Interconnects. In this way, if one of them becomes unresponsive, the other takes care of the management. However, keep in mind that this applies to the management application (that is, on which of them the management application will run in active mode). When it comes to processing the data communication from the servers, the two Fabric Interconnects are both active. This is another way of adding redundancy and reliability.
South from the Fabric Interconnects are the Cisco UCS servers. There are two types of servers: the B-series and C-series servers (that is, the blade and rack servers). The C-series servers are equipped with their own power supplies, cooling, and communication adapters. Therefore, when the C-series servers are part of a Cisco UCS domain, they are connected directly to the Fabric Interconnects. There are other supported connectivity options between the C-series servers and the Fabric Interconnects that involve the use of special Cisco 2000 Fabric Extenders (FEXs).
The B-series servers are installed in the Cisco UCS 5108 blade chassis. The blade chassis takes care for the redundant power supply, cooling, and connectivity. For the connectivity, the chassis has installed the Cisco UCS 2000/2200/2300/2400 FEXs, or I/O Modules (IOMs). The series represents the generation of the Cisco UCS FEX. For example, the Cisco UCS 2000 FEXs are the first generation and are no longer sold or supported. The Cisco UCS 2200 FEXs are the second generation and are to be used with the second generation of the Fabric Interconnects and such.
The IOMs, or FEXs, are not switches. This means they do not perform L2 lookups as any other Ethernet switch, so there’s no processing of the Ethernet frames. The IOMs are communication devices that physically connect the uplinks between the FI and the blade chassis with the servers installed in the blade chassis. This happens through mapping of the server-facing (internal) ports to the uplink (external) ports of the IOM. Depending on the model of the IOM/FEX, there’s a different number of internal ports. They are hard-pinned to the slots in which the blade servers are installed. This is a huge advantage, as it means the communication of the servers will always have to reach the Fabric Interconnects, and they will be the first networking-capable devices to process the Ethernet frames. In this way, the number of hops in the communication between the servers inside the UCS will be kept to a minimum.
But here’s one very important thing: the Fabric Interconnects can operate in two modes when it comes to Ethernet processing. The default mode is end host mode, which again is passive. In it, the servers’ traffic is pinned (mapped) to an uplink port and sent to an upstream switch, and there the Ethernet frame is processed. This is because in this mode the Fabric Interconnects do not learn MAC addresses from the LAN infrastructure. Therefore, the upstream network switch will have the knowledge of where to switch to the Ethernet frame from the Cisco UCS server. But the Fabric Interconnects in this mode still learn MAC addresses on their server ports. These are the Ethernet ports that connect to the Cisco UCS IOMs in the blade chassis, or to the C-series servers connected directly to them. In this way, if two Cisco UCS servers want to communicate with each other, and communicate through the same FI, this traffic will be switched locally, without exiting the FI, and without the burden of additional network processing hops, as the IOMs are passive devices.
The second mode of operation for the Fabric Interconnects is the switching mode, where, with some minor limitations, they operate as network switches.
Note that previously, the terms Fabric Extender (FEX) and I/O Module (IOM) have been used interchangeably. In general, from Cisco’s perspective, they mean the same thing, but historically they are used for different things. For example, the Cisco Nexus 2000/2200/2300 FEXs are standalone devices, intended to increase the port density when they are connected to upstream Nexus 5500, 5600, 6000, 7000, 7700, or 9000 Series switches. They belong to the LAN component of the data center, and their role is to extend the access layer. The Cisco UCS FEXs/IOMs devices are installed in the blade chassis (they do not have their own power supply or cooling) for the purpose of providing connectivity for the blade servers. As it is true that they operate the same way as the Nexus 2K FEXs, they are different devices with different application.
Now we reach the blade servers. As already mentioned, they are installed in the server slots of the blade chassis, which has a total of eight server slots. The blade servers come in two physical forms: half-width and full-width servers. The half-width blade server occupies one server slot on the chassis. If equipped only with half-width blade servers, the chassis can accommodate up to eight servers. The full-width blade servers occupy two horizontal slots in that chassis, for a maximum of four full-width servers per chassis.
The B-series servers are equipped with a blade connector, which connects the server to the chassis in the slot. The blade connector provides the power to the server, enables the management communication, and also communicates with the server’s converged network adapter. With that said, there are two major types of communication for the blade server:
- The data plane communication: This includes both the LAN and SAN communication. It happens through the converged adapter or in the Cisco UCS, also called a mezzanine adapter. The mezzanine adapter provides the server separate NICs and HBAs, and externally it has only 10Gbps or higher Ethernet interfaces and supports FCoE for the FC communication.
- The management plane communication: Every Cisco server is equipped with separate, autonomous management hardware. It is called the Cisco Integrated Management Controller (CIMC) and it has its own CPU, memory, flash storage, and network interface. When the server is part of a Cisco UCS domain, the CIMC is under the control of the Cisco UCSM, or put a different way, you as an administrator connect to the Cisco UCSM and work from there. The UCSM is responsible for communicating with the CIM and pushing down configurations and commands and retrieving back status and monitoring information. When you work with a C-series server, and it operates in standalone mode, there is no Cisco UCSM. You connect directly to the server’s CIMC in order to manage and configure it.
Both the management and data plane communication on the B-series server go through the blade connecter, using separate paths of communication through the blade chassis midplane, and through the chassis’ IOMs to reach the Fabric Interconnects.
With all of that said, it is very important to understand that Cisco UCS’s physical connectivity creates the underlying communication infrastructure. This infrastructure is available for the communication needs of the servers. How it will be used by each of the servers is a matter of configuration in the Cisco UCSM and relates to the hardware abstraction principle on which the Cisco UCS is based, discussed in the chapter “Describing Cisco UCS Abstraction”.