Cisco UCS Blade: A New Approach to Building a Data Center
Almost every year in the IT world, a new trend sets in. For example, in 2007 it was data consolidation on storage systems, in 2008-2009 it was a blade system, 2010 was a year of virtualization, and since 2011 the idea of consolidating data transmission networks has been actively developing. UCS includes all of the above features, and also has improved features of fault tolerance and scalability.
First, let's look at the architecture of Cisco blade systems:
We’ll highlight the main components:
1) Fabric Interconnect - Nexus 5000-based switches with built-in UCS Manager functionality that allows you to manage all servers, BIOS settings, component firmware, switches, routing, virtualization and automation. They also allow you to work with Ethernet and SAN networks and fully support the FCoE protocol.
2) UCS 5108 Blade Chassis - the chassis in which blade servers (blade servers) are installed. It is used as a centralized server cooling system and power supply to the servers.
3) Fabric Extender (FEX) - Extends Fabric Interconnect in the blade chassis, laying multiple 10 Gb / s connections between the blade servers and Fabric Interconnect. Not a switch. Managed as an extender to the Fabric Interconnect switch. It controls power supplies and fans. Mounted in the rear of the blade chassis.
Based on the architecture, we can connect up to 40 blade chassis to a single pair of Fabric Interconnect switches, managing all of them from a single point (UCS Manager). We also do not need to purchase separate switches for each blade chassis, which allows us to save money when expanding our fleet of blade servers.
4) Virtual Interface Card (VIC) - specialized network adapters from Cisco. They allow several network links to pass through a single communication channel. In this case, the processing of all network traffic is performed by the adapter itself. A detailed diagram is shown in the figure:
Each adapter, whether it is 10, 20 or 40 Gb / s, can be divided into 128 virtual ones. The technology is based on the IEEE 802.1bn standard. When creating a new network adapter, you can specify it:
- speed, with a move of 1 Mbps.
- QoS priority.
- type: Ethernet or FCoE.
- Failover working mechanism.
- VLAN / VSAN.
What do we have in the end? Easy administration of each adapter, reducing the number of cables for connecting between the server and the switch, and a flexible system for adding new devices to the server.
This system is ideal for virtualization tasks. Indeed, with a large number of network adapters in the server, we do not need to use a virtual hypervisor switch. It is enough to make a direct forward of each of the adapters to virtual machines.
To understand the principles of Cisco UCS fault tolerance, let's analyze the basic structure of server traffic transfer. UCS consists of a blade chassis and Fabric Interconnect (FI). The blade chassis accommodates the blade servers, and FI performs LAN, SAN switching and manages the chassis / servers.
The number 1 indicates the connection between FEX and FI, which can be from one to eight. Each connection has a speed of 10 Gbit / s. Thus, between eight and 160 Gbps from Fabric Interconnect central switches can be connected to eight servers located inside each chassis.
Connections can be configured to operate in EtherChannel mode or a separate interface can be allocated for each server. Depending on the type of task, network adapters with speeds of 20, 40 or 80 Gbit / s can be installed on the server. In any case, if one of the links fails, its traffic will be distributed between neighboring links. The speed of each virtual adapter can be limited by Quality of Service (QoS) and Enhanced Transmission Selection (ETS) mechanisms.
UCS has a unique feature available on some VIC adapters, which allows you to identify a failed network section and perform a “failover” on the adjacent Fabric Interconnect. Some mezzanine cards (VICs) have a built-in mini-switch that allows you to switch gears from path A to path B and vice versa. Schematically, this is shown in the figure:
Hardware UCS failover provides better fault tolerance than traditional NIC-teaming, thanks to the intelligent switching system built into FI. In the picture above, the HW Failover identifies: a failure in the chassis, in the FEX, or on the network adapter interface. In addition to everything, if FI loses an external connection to a LAN network, it will send a message to the VIC adapter and allow switching to the backup path. In the picture, any failure in the places marked with X will switch Ethernet traffic to path B. Hardware UCS failover is used only for Ethernet networks, since SAN networks have their own failover mechanism, thanks to the multipath mechanism.
MUK-Service - all types of IT repair: warranty, non-warranty repair, sale of spare parts, contract service
Scalability
First, let's look at the architecture of Cisco blade systems:
We’ll highlight the main components:
1) Fabric Interconnect - Nexus 5000-based switches with built-in UCS Manager functionality that allows you to manage all servers, BIOS settings, component firmware, switches, routing, virtualization and automation. They also allow you to work with Ethernet and SAN networks and fully support the FCoE protocol.
2) UCS 5108 Blade Chassis - the chassis in which blade servers (blade servers) are installed. It is used as a centralized server cooling system and power supply to the servers.
3) Fabric Extender (FEX) - Extends Fabric Interconnect in the blade chassis, laying multiple 10 Gb / s connections between the blade servers and Fabric Interconnect. Not a switch. Managed as an extender to the Fabric Interconnect switch. It controls power supplies and fans. Mounted in the rear of the blade chassis.
Based on the architecture, we can connect up to 40 blade chassis to a single pair of Fabric Interconnect switches, managing all of them from a single point (UCS Manager). We also do not need to purchase separate switches for each blade chassis, which allows us to save money when expanding our fleet of blade servers.
4) Virtual Interface Card (VIC) - specialized network adapters from Cisco. They allow several network links to pass through a single communication channel. In this case, the processing of all network traffic is performed by the adapter itself. A detailed diagram is shown in the figure:
Each adapter, whether it is 10, 20 or 40 Gb / s, can be divided into 128 virtual ones. The technology is based on the IEEE 802.1bn standard. When creating a new network adapter, you can specify it:
- speed, with a move of 1 Mbps.
- QoS priority.
- type: Ethernet or FCoE.
- Failover working mechanism.
- VLAN / VSAN.
What do we have in the end? Easy administration of each adapter, reducing the number of cables for connecting between the server and the switch, and a flexible system for adding new devices to the server.
This system is ideal for virtualization tasks. Indeed, with a large number of network adapters in the server, we do not need to use a virtual hypervisor switch. It is enough to make a direct forward of each of the adapters to virtual machines.
fault tolerance
To understand the principles of Cisco UCS fault tolerance, let's analyze the basic structure of server traffic transfer. UCS consists of a blade chassis and Fabric Interconnect (FI). The blade chassis accommodates the blade servers, and FI performs LAN, SAN switching and manages the chassis / servers.
The number 1 indicates the connection between FEX and FI, which can be from one to eight. Each connection has a speed of 10 Gbit / s. Thus, between eight and 160 Gbps from Fabric Interconnect central switches can be connected to eight servers located inside each chassis.
Connections can be configured to operate in EtherChannel mode or a separate interface can be allocated for each server. Depending on the type of task, network adapters with speeds of 20, 40 or 80 Gbit / s can be installed on the server. In any case, if one of the links fails, its traffic will be distributed between neighboring links. The speed of each virtual adapter can be limited by Quality of Service (QoS) and Enhanced Transmission Selection (ETS) mechanisms.
UCS has a unique feature available on some VIC adapters, which allows you to identify a failed network section and perform a “failover” on the adjacent Fabric Interconnect. Some mezzanine cards (VICs) have a built-in mini-switch that allows you to switch gears from path A to path B and vice versa. Schematically, this is shown in the figure:
Hardware UCS failover provides better fault tolerance than traditional NIC-teaming, thanks to the intelligent switching system built into FI. In the picture above, the HW Failover identifies: a failure in the chassis, in the FEX, or on the network adapter interface. In addition to everything, if FI loses an external connection to a LAN network, it will send a message to the VIC adapter and allow switching to the backup path. In the picture, any failure in the places marked with X will switch Ethernet traffic to path B. Hardware UCS failover is used only for Ethernet networks, since SAN networks have their own failover mechanism, thanks to the multipath mechanism.
MUK-Service - all types of IT repair: warranty, non-warranty repair, sale of spare parts, contract service